JP7638532B2 - Dead chicken detection device and chicken farming system equipped with same - Google Patents

Description

The present invention relates to a device that detects dead chickens by observing inside cages in a poultry farm and a poultry farming system equipped with the device.

In recent years, chicken coops have become larger and feeding and other rearing tasks have become more mechanized in order to improve egg production efficiency, that is, to reduce the labor of producers, to be considerate of the surrounding environment, and to deal with chicken infectious diseases. However, even if the coop is isolated from the outside world to prevent the aforementioned infectious diseases, as the number of chickens raised increases and chickens have a short lifespan, a certain percentage of chickens will die. For example, in a large chicken coop that rears 100,000 chickens, 20 chickens may die a day, and in the summer, the number of dead chickens increases due to heat death caused by high temperatures.

Not only do these dead chickens decay themselves, but in some cases they can block eggs laid by other chickens inside the egg collection belt or at the back of the cage from rolling down the slope of the floor, which can cause delays between eggs being laid and the eggs actually being collected. Also, with automated egg collection belts, if dead chickens get in the way, eggs cannot be collected. For this reason, dead chickens need to be removed as quickly as possible.

Meanwhile, today's chicken coops, especially the larger ones, are automated, and chicken coops (chicken farming systems) have been realized in which feeding, watering, egg collection, and manure disposal (replacement of flooring) are all done without human intervention. Therefore, while there is no longer the opportunity to observe the chickens while feeding and collecting eggs, as was the case with farms in the past, dead chickens still need to be surveyed daily. However, with the aforementioned large-scale construction, one chicken coop may have cages that are, for example, 100m long and 3m high, with 10 rows and two floors, and an inspection takes 1.5 to 2 hours, which is hard work.

In addition, the front of the cage has food trays and egg trays that run horizontally like a gutter, making it difficult to see inside the cage. Nowadays, cages of a certain size often house 10 or more chickens together, and dead chickens are sometimes discovered when they are pushed along the slope of the floor by the other chickens. For this reason, it often takes some time for a dead chicken to be discovered. In particular, the top floor of the cage cannot be seen with the naked eye, so the keeper's intuition and experience are required, and it is easy for something to be overlooked.

In order to solve such problems, the dead chicken detection system of Patent Document 1 has been proposed. Patent Document 1 uses a mobile camera vehicle to obtain visible and infrared images of the same location in a chicken coop, identifies pixels from the visible image data that match a specified registered chicken color (the color of dead chickens), and identifies pixels with low temperatures from the infrared image data. If a similar pixel is found near the matching pixel, i.e., if a mass is present, it is determined that a dead chicken is present.

Patent No. 6678898

In the above-mentioned conventional technology, dead chickens are judged based on two parameters, the chicken's color and body temperature, so dead chickens can be detected relatively accurately even with a single captured image. However, since the judgement is based on the time from death until the chicken changes color and the temperature drops, it takes a considerable amount of time to make the judgment, and it is not possible to find chickens that are already weak and not yet dead. Weak chickens not only do not lay eggs and waste feed, but they also get in the way of egg collection.

Tozai Sangyo Boeki Co., Ltd.'s dead chicken navigation system takes pictures of the inside of the cage early in the morning, and uses AI to determine from the images that no movement is a dead chicken. However, even with this conventional technology, the judgment is made from a single captured image, so it is difficult to make an accurate judgment if the image is taken in the early morning when the chickens are laying eggs, in the evening when they are resting, or when feeding activity is low.

The object of the present invention is to provide a dead chicken detection device that can accurately and quickly detect dead or weakened chickens, and a poultry farming system equipped with the same.

The dead chicken detection device of the present invention is a device for detecting dead chickens in cages that house chickens, and is characterized in that the cages are arranged in multiple rows at least horizontally, and the device includes an imaging device that moves horizontally to capture continuous images of the inside of the cages, a first determination means that analyzes the images captured by the imaging device and provisionally determines whether or not the chicken is a dead chicken, and a second determination means that accumulates the determination results of the first determination means for a predetermined number of times by matching the movement position standard of the imaging device, and makes a final determination that the chicken is actually dead if it is equal to or exceeds a predetermined threshold value.

According to the above configuration, when detecting dead chickens in cages, in a large-scale chicken coop in which cages housing many chickens are lined up horizontally (of course, in most cases they are also lined up vertically in multiple layers), it is not practical to constantly observe each cage. Therefore, an imaging device is provided on a self-propelled cart, or preferably on a feeding device that drops food while moving, to capture images of the inside of the cage as it moves horizontally. Then, when analyzing the captured images to determine whether a chicken is dead, since a determination based on a single image would not be able to capture fine details such as movement, the imaging device captures continuous images, and the first determination means tentatively determines whether the chicken is dead or not by, for example, performing image recognition on a frame-by-frame basis (for example, the belly is visible and the chicken is lying down).

However, with this provisional judgment, there is a possibility that a chicken may be judged as dead even if it is resting after eating or lying on its stomach in preparation for laying eggs. Therefore, a second judgment means is provided, and the provisional judgment results of the first judgment means are accumulated a predetermined number of times (in the chronological order). Since the imaging device is moving, the timing is adjusted so that the movement position reference, for example the wall or the wire mesh part of the cage, is matched, and the provisional judgment results are accumulated. The accumulated value is compared with a predetermined threshold value, and if it is equal to or exceeds the threshold value, it is officially judged as a dead chicken.

Therefore, dead chickens are not judged based on a single image captured by a moving imaging device, but are judged based on repeated judgment results at the same location at different times. This requires at least the threshold number of movements (image capture) before a judgment can be made, which takes some time, but still allows accurate detection of dead and weak chickens relatively quickly. Then, if an operator follows the judgment results and goes to the corresponding cage, he or she will be sure to find dead or weak chickens there, which improves the efficiency of the work of removing them, and there are no misses, earning the operator's trust.

The dead chicken detection device of the present invention is also characterized in that the imaging device is provided on the feeding device or on a cart that follows the feeding device.

In a large-scale chicken coop where cages housing many chickens are arranged in multiple tiers horizontally and vertically, an automatic feeding device is used that moves in front of each cage at a predetermined time along rails laid horizontally and then places feed in the feed trays installed in each cage. Although the number of feeding times and the time periods vary slightly depending on the chicken farmer, feeding is typically carried out five to six times during the day, since the chickens are allowed to rest at night. The imaging device is then installed on the feeding device or a cart that follows it, for example, near the feed openings of each tier if the cages are stacked in multiple tiers, or on each tier on both sides if cages are installed on both sides.

Chickens stand and move actively when eating, so if you observe the chickens at feeding time, especially by looking at the chicken's feet from a low viewpoint near the feed tray, you can effectively determine whether the chicken has died or not, and even its health condition, whether it is weak from summer fatigue or the like. Specifically, you can determine this by looking at the chicken's belly or comb at feeding time, or by the movement of its legs, which would move a little if it was standing.

In this way, by simply adjusting the timing of observing the chickens, dead chickens can be found relatively accurately and easily. Also, by observing at frequent feeding times, dead chickens can be found quickly. Furthermore, when an imaging device is provided in the feeding device, there is no need to provide a dedicated device as in Patent Document 1. On the other hand, if a feeding device has already been installed, a cart can be used to follow the feeding device using the same walkways and rails, and the imaging device can be provided on the cart, allowing a dead chicken detection device that can observe chickens at feeding times to be retrofitted.

Furthermore, in the dead chicken detection device of the present invention, the imaging device is characterized in that it observes the inside of the cage through the gap between the food receptacle (feeding tray) located relatively above and the egg receptacle located below.

According to the above configuration, the food receiver (feeding tray) is formed continuously in the horizontal direction like a gutter, which can be an obstacle when observing the area around the chickens' feet from the low viewpoint. However, it is different from the egg receiver, which is also formed continuously in the horizontal direction like a gutter. By installing an imaging device so that it can look through the gap between the two, it is possible to continue to observe the area around the chickens' feet continuously and without obstruction between each cage, which is preferable.

The dead chicken detection device of the present invention is also characterized in that the imaging device is provided in the feeding device, the feeding device further includes a recording device for recording images captured by the imaging device, and the images captured by the recording device are transferred to the first determination means at the starting point for replenishing the feeding device with food.

According to the above configuration, the feeding device travels back and forth in front of the cage, for example, 5 to 6 times a day, as described above, and it is necessary to replenish the feed at the base point after each travel. Therefore, the images captured by the imaging device are recorded in the recording device, and at the replenishment base point, the captured image data is collected and collected from the recording device to the first determination means using simple short-range wireless communication such as Wi-Fi (registered trademark). In this way, there is no need to run wiring to the feeding device for detecting dead chickens, or to install large-scale wireless communication equipment in a large (long) chicken coop; the feeding device only requires the running of conventional power lines and the necessary control lines.

Furthermore, in the dead chicken detection device of the present invention, the first judgment means performs the provisional judgment by analyzing the continuous images of the imaging device and performing a binary judgment for each frame as to whether or not the chicken is dead, as well as a binary judgment as to whether or not the chicken is passing through the moving position reference, and the second judgment means performs the accumulation for the predetermined number of times by shifting the time axis of the binary judgment result as to whether or not the chicken is dead so that the moving position reference is consistent.

According to the above configuration, the first determination means provisionally determines whether or not the chicken is a dead chicken for each frame from the continuous images in a binary value, and determines whether or not the chicken has passed a moving position reference such as a wall or the wire mesh of a cage in that frame. Then, when accumulating the binary determination results of whether or not the chicken is a dead chicken, the second determination means adjusts the time axis of the binary determination results of whether or not the chicken is a dead chicken so that the moving position reference is consistent.

Therefore, even if the travel speed of the feeding device, cart, etc. varies and the number of image frames between the moving position references (one cage) differs, the time difference can be adjusted and the binary provisional judgment results of whether or not a chicken is dead can be accurately accumulated in chronological order for a specified number of frames. The variation in travel speed is caused by the wind direction and volume of the ventilation and cooling air, wheel slippage, or differences in drive voltage and load (amount of feed). This makes it possible to accurately determine the result of the actual judgment of whether or not a chicken is dead.

The dead chicken detection device of the present invention is also characterized in that the second judgment means integrates the binary judgment result of whether or not the chicken is dead a predetermined number of times, in accordance with a predetermined number of bits.

According to the above configuration, the first determination means does not determine whether a chicken is dead from a single image capture by the moving imaging device, but instead uses this as a provisional determination result, and the second determination means accumulates the provisional determination results to make a real determination. The timing is adjusted so that the movement position reference of the imaging device, for example, the wall or the wire mesh part of the cage, coincides with the provisional determination results. During this accumulation, for example, if the binary determination result of whether or not a chicken is dead by the first determination means is less than a predetermined number of bits (frames) between the movement position references, i.e., for one cage, the second determination means inserts dummy bits (frames), and if it is more, deletes bits (frames). Alternatively, without determining a predetermined number of bits (frames) for one cage, dummy bits (frames) are inserted according to the data with the largest number of bits (frames).

This makes it possible to compensate for variations in the moving speed of the imaging device (number of frames) when adjusting the timing and accumulating the judgment results.

Furthermore, in the dead chicken detection device of the present invention, the second judgment means uses the result of the accumulation as the reference data, creates data shifted one frame forward and backward, and compares the result of adding these three data together with the predetermined threshold value.

With the above configuration, even if dead chickens happen to be missed at the same time and this is reflected in the accumulation result (count value), by adding data with a shift in timing earlier or later and comparing it with the threshold, such missed detection can be compensated for and an accurate judgment can be made.

Furthermore, the poultry farming system of the present invention is characterized by being equipped with the dead chicken detection device described above.

The above configuration makes it possible to realize a poultry farming system that can easily and quickly identify weak chickens from dead chickens.

As described above, the dead chicken detection device of the present invention and the poultry farming system equipped with it capture images of the inside of the cage while moving the imaging device laterally, analyze the captured images, and when determining whether the chicken is dead, the first determination means provisionally determines whether the chicken is dead or not based on the state of the chicken in each successive image, such as whether the belly is visible and the chicken is lying down, and the second determination means further adjusts the timing so that the movement position reference of the successive images, such as the wall or the wire mesh part of the cage, coincides, accumulates the provisional determination results, and compares them with a predetermined threshold value to make an actual determination of whether the chicken is dead.

Therefore, although it takes some time to make a judgment, since it is made based on the results of repeated judgments made at different times in the same place, it is possible to accurately detect everything from dead chickens to weak chickens.

1 is a block diagram of a poultry farming system equipped with a dead chicken detection device according to one embodiment of the present invention. FIG. FIG. 1 is a cross-sectional view of a portion of a chicken coop according to one embodiment of the present invention. FIG. 2 is a diagram for explaining an image of determining whether a dead chicken is present using the dead chicken detection device of FIG. 1, and is a diagram showing frame accumulation. FIG. 4 is a diagram for explaining an image of frame timing adjustment in the frame integration in FIG. 3 . FIG. 2 is a functional block diagram for explaining a determination method using the dead chicken detection device of FIG. 1. FIG. 13 is a diagram showing the provisional determination of dead chickens. FIG. 13 is a diagram showing how the center position of the cage wall is determined. FIG. 8 is a diagram showing how frame timing is adjusted based on the determination results of FIG. 7; FIG. 9 is a diagram for explaining a method for determining the presence or absence of dead chickens by integrating data timed in FIG. 8. FIG. 10 is a diagram for explaining a method for determining whether or not there are actual dead chickens by processing the data in FIG. 9 . FIG. 11 is a diagram showing a data string as an example of the determination result of FIG. 10 . FIG. 13 is a diagram showing the search results for parameters for determining the presence or absence of dead chickens. In FIG. 12, a diagram showing preferred parameters is shown.

Figure 1 is a block diagram of a poultry farming system equipped with a dead chicken detection device according to one embodiment of the present invention. Figure 1(a) is a front view of a chicken coop 1, and Figure 1(b) is a side view. The dead chicken detection device of this embodiment is generally used to monitor the health of a large number of chickens, particularly dead chickens, raised in one or more chicken coops 1 from an office 11, and to share this information via mobile terminals 12 carried by breeders, etc. Note that the chickens raised are not limited to egg-laying females, but may also be meat chickens or chickens in the adult growth stage. In addition, in this specification, explanations of normal breeding other than the detection of dead chickens and chickens in poor health (weakened) are omitted.

Figure 2 is a cross-sectional view of a portion of a chicken coop 1 according to an embodiment of the present invention. In the chicken coop 1, multiple rows of cage chicken coops (only two rows (mountains) are shown in Figure 1 for the sake of simplicity) 2 are formed parallel to each other, for example, over 100 m in the width (depth in Figure 1) direction. In each cage chicken coop 2, multiple (four in Figure 1) tiers of cages 3 are stacked, and each cage 3 is divided into a predetermined length in the width direction by a wall 31 such as a wire mesh. Each cage 3 is divided into two in the depth (left and right in Figure 1) direction, so that each cage chicken coop 2 can be accessed from the front and back. Each cage 3 houses, for example, 10 to several chickens 41 to 43 (hereinafter, when collectively referred to, they will be referred to by the reference number 4) in each of the two tiers. In one chicken coop 1, cage chicken coops 2 like those in Figure 1 may be two stories high.

In front of the cages 3 arranged in the horizontal (width, horizontal) direction, a feed tray (feeding tray) 51 is installed at the top and an egg tray 52 is installed at the bottom, like a gutter. A water supply pipe 53 is installed in the horizontal (width) direction in common at the back (center) of the cages 3 arranged back to back, and a water supply port is provided on the water supply pipe 53 for each cage 3 arranged back to back. The water intake port is, for example, a nipple-type waterer, and when the chicken 4 pushes up the nipple-shaped protrusion 54 with its beak, a valve opens and chicken drinking water flows down. The floor of the cage 3 is made of a coarse mesh plate 55, so that the droppings of the chickens 4 pass through the mesh plate 55 and fall onto the droppings removal belt 56 below, which is sent (circulated) at regular intervals to scrape off the droppings and dry them. The mesh plate 55 slopes downward from the rear (center) side of the back-to-back cage 3, allowing laid eggs to roll and making it easier to push out dead chickens.

In the large-scale cage chicken house 2 configured as described above, when detecting dead chickens in the cage 3 as indicated by reference number 43 and weak chickens as indicated by reference number 42, it is generally possible to detect them accurately and quickly by increasing the frequency of observing the inside of the cage 3. However, in a large-scale cage chicken house 2 in which cages 3 containing a large number of chickens 4 are arranged in multiple stages in the horizontal direction and the vertical direction, it is not realistic to observe each cage 3 all the time. Therefore, it is noteworthy that in the dead chicken monitoring device of this embodiment, an imaging device 7 is provided on the feeding device 6 that drops feed while moving, and the inside of the cage 3 is photographed in succession as it moves horizontally. Then, from the photographed images, the judgment means 111 provided in the office 11 judges the dead chickens 43 and weak chickens 42, as will be described in detail later. The imaging device 7 is not limited to the feeding device 6, but may be provided on a cart that follows it.

The feeding devices 6 are provided in each row (mountain) of the cage chicken coop 2, and are controlled by a control device (not shown) to move on their own at a specified time to feed the feed receptacles (feeding trays) 51 of the cages 3 in each row (tier). For this purpose, rails 61 are laid on the upper front surface of each of the cages 3, which are lined up in large numbers in the horizontal direction, and wheels 62 run on these rails 61 to move the frame 63. The frame 63 is provided with feeding openings 64 corresponding to the feed receptacles (feeding trays) 51 of the cages 3 in each row (tier), and an imaging device 7 is installed near each feeding opening 64.

Specifically, when the imaging device 7 captures successive images while traveling in front of the cage 3, the food receptacle (feeding tray) 51 and the egg receptacle 52 are formed continuously in the lateral (horizontal) direction like a gutter as described above, so in this embodiment, the imaging device 7 observes the inside of the cage 3 through the gap between the food receptacle (feeding tray) 51, which is positioned relatively higher, and the egg receptacle 52, which is positioned relatively lower.

In FIG. 2, rails 61 are laid on the top, and the frame 63 runs suspended from the rails 61, making it easy to clean the floor 10 of the chicken coop 1, but the rails 61 may also be laid on the floor 10. The frames 63 may also be stabilized by being interconnected between the rows (mountains) facing each other, and anti-sway rails and wheels or sliding plates that come into contact with the rails may be provided on the top and bottom of the frame 63.

The imaging device 7 is realized by a digital video camera or the like, and each captured image is recorded in a common recording device 71 mounted on the feeding device 6 together with the camera's identification information and date and time information. Then, while the feeding device 6 is returning to the home position (base point) as shown in FIG. 1(b), the recording device 71 transmits the captured image data together to a router 102 installed near the home position by a short-range simple wireless communication 103 such as Wi-Fi (registered trademark). The captured image data received by the router 102 is further transmitted to a server 112 in the office 11 via a communication line 104 in the poultry farming facility and stored there. The home position is set, for example, at a location where a feed refilling machine 101 equipped with a feed tank and a hopper is installed. Although feeding times and periods vary slightly depending on the poultry farmer, the chickens are allowed to rest at night, so the feeding device 6 leaves this home position and travels back and forth in front of the cages 3 at predetermined times, for example, five or six times during the day, and feed is replenished at the home position on each return (round trip).

The captured images in the server 112 are appropriately read out to the analysis personal computer 111, which is the first judgment means. Then, for each frame, the image is recognized by artificial intelligence (AI) to provisionally judge whether it is a healthy chicken 41 or a dead chicken 43. A dead chicken 43 is judged, for example, if it has fallen with its belly visible or its comb is on the floor. However, in this provisional judgment, it is possible that the chicken that appears to be weak, indicated by reference number 42, is resting after eating or laying eggs, in other words, is healthy. Therefore, as the second judgment means, the personal computer 111 accumulates the provisional judgment results for a predetermined number of frames (in the chronological order), compares them with a predetermined threshold, and makes a final judgment that the chicken 43 is actually dead if it is above the threshold.

It should be noted that since the imaging device 7 is moving, when making the actual judgment, the personal computer 111, which serves as the second judgment means, adjusts the timing so that the wall 31 coincides with the moving position reference in the captured image, and accumulates the provisional judgment results. If the actual judgment results in a dead chicken 43 or a weak chicken 42, this is displayed on the screen of the personal computer 111 together with the cage number, and similar information is also shared with the mobile terminal 12 carried by the zookeeper via a public wireless communication line 113 such as a mobile phone line.

Figures 3 and 4 illustrate the image of determining dead chickens. Figures 3(a) and 3(b) show the result of determining dead chickens by the image recognition of the artificial intelligence (AI) in images taken at different times at the same position by the same imaging device on different dates or different imaging devices. In the example of Figure 3, the number of accumulated frames n is 6 and the determination threshold t is 5 when determining dead chickens. Figures 3(a) and 3(b) show the six frames F1, F2, ..., F6 arranged in two rows and three columns. In Figures 3(a) and 3(b), a feed tray 51 is shown at the top and an egg tray 52 is shown at the bottom, but the angle of view of the imaging device 7 can be adjusted to cut these out. In this embodiment, these parts are excluded in advance from the image recognition of the artificial intelligence (AI), so that the angle of view can be expanded even when a general-purpose imaging device 7 is used.

In Fig. 3(a), a dead chicken 43 is detected five times in frames F1, F2, F3, F4, and F6. Although it was not detected in frame F5, it is possible that the detection was missed, so it is determined that a dead chicken 43 is present because the threshold value t is exceeded. On the other hand, in Fig. 3(b), it is detected sparsely in frames F2 and F4, and it is not detected in the previous and following frames either, so it is determined that the threshold value t is exceeded, so there is a possibility of a misdetection, and it is determined that a dead chicken 43 is not present.

Meanwhile, Figure 4 is a diagram that outlines a method for adjusting the timing for the accumulation. Like Figures 3(a) and (b), Figure 4(a) shows images captured at different times at the same position by the same imaging device 7. Like Figures 3(a) and (b), frames F1, F2, ..., F6 are shown in three columns and two rows. However, while Figure 3 shows an image near the center of the cage 3, Figure 4 shows an image near the wall 31, which serves as the reference for the movement position.

To be precise, Figure 4(a) shows images captured at different times at what appears to be the same location. In the example of Figure 4(a), wall 31 appears somewhere in images (frames) F1 to F6. However, there is variation in the position of wall 31 within images (frames) F1 to F6. This variation is often due to differences in the speed of feeding device 6, since imaging device 7 is mounted on feeding device 6 and moves, and more specifically, due to the direction and volume of airflow for ventilation and cooling, slippage of wheels 62 against rails 61, fluctuations in drive voltage, food loading conditions, etc.

Therefore, if frames F1 to F6 were simply added up in the situation shown in Figure 4(a), the positions of chickens 41 to 43 within frames F1 to F6 would be shifted, making it difficult to find dead chickens 43 and weak chickens 42. Therefore, when making an actual judgment by adding up frames F1 to F6 as shown in Figures 3(a) and (b), personal computer 111, which serves as the second judgment means, adjusts the timing so that the wire mesh part of wall 31, which serves as the movement position reference in the captured image, coincides, and adds up the provisional judgment results.

Figure 4(b) shows the results of adjusting the timing of each image (frame) F1 to F6 in Figure 4(a). In the following explanation, it is assumed that the imaging device 7 is moving from right to left toward the screen. And in Figure 4, the wall 31 is in the center of the image, that is, the image is just between the cages 3. In Figures 4(a) and 4(b), the wall 31 is originally in the center of the image in frames F1 and F2, that is, there is no timing deviation, so no adjustment is made. In frames F3, F4, and F5, in Figure 4(a), the wall 31 is on the left side of the screen, that is, the imaging device 7 is slow and the wall 31 has not reached the center position of the image, and in Figure 4(b), the timing is adjusted to be faster. Conversely, in frame F6, in Figure 4(a), the wall 31 is on the right side of the screen, that is, the imaging device 7 is fast and the wall 31 has passed the center of the image, and in Figure 4(b), the timing is adjusted to be slower.

Figure 5 is a functional block diagram for explaining the above-mentioned processing method in the personal computer 111 as the first and second judgment means. Each imaging device 7 captures moving images P1, P2, ..., Pn (hereinafter, collectively referred to as P) for each round trip of the feeding device 6. n is the cumulative number of frames, and for example, for one day, n = 12 for six feedings. Then, as described above, for each image (frame) F1, F2, ..., Fn (hereinafter collectively referred to as F) of each video image P, the artificial intelligence (AI) performs image recognition and creates a sequence of consecutive binary data H1, H2, ..., Hn (hereinafter collectively referred to as H) that determines whether or not a dead chicken 43 is present in the image (frame) F, and creates a sequence of consecutive binary data W1, W2, ..., Wn (hereinafter collectively referred to as W) that determines whether or not a wall 31 is present.

Figure 6 is a diagram explaining how the binary data H and W are created. As in Figure 3, the moving image P is a series of images of a given range at a given time in front of the cage 3. When this is expanded into a continuous screen for easy understanding, it becomes as shown in Figure 6(a). Each image (frame) F is cut out from the given range, for example, as shown in Figure 6(b). There are two images (frames) cut out in Figure 6(b), and they are referenced Fx and Fy. When a binary judgment is made from these images (frames) Fx and Fy as shown in Figure 6(c), the column of data W of the wall 31 is as shown in Figure 6(d), and the column of data H of the dead chicken 43 is as shown in Figure 6(e).

When accumulating data H of dead chickens 43 shown in FIG. 6(e) n times, it is necessary to identify the exact position of which cage and which position the data is from. Therefore, a position identification process is performed to identify the actual center point of wall 31 from data W, which is the binary judgment result of wall 31 shown in FIG. 6(d), and a string of data indicating the judgment result of the center point is obtained as a moving position reference. FIG. 7 shows a method for determining the center point of wall 31. Here, for example, data H is 25 frames, i.e., 25 bits of data per cage 3, and wall data W is 15 frames, i.e., 15 bits of data, but in FIG. 7, the number of bits is reduced to simplify the drawing. Note that the 25 frames of data H are the number of frames in which the wall 31 is not reflected at all, and the 15 frames of data W are the number of frames in which the wall 31 is reflected at least partially.

Figure 7(a) is a column of data on the wall, similar to that in Figure 6(d). In each image (frame) F, "1" indicates that a wall has been detected, and "0" indicates that a wall has not been detected. Figure 7(b) is the actual center point of the wall 31. Figure 7(c) is a graph showing the determination result of Figure 7(a) as a broken line. In Figure 7(c), when a wall 31 is detected (when the binary data becomes "1"), the counter starts counting, and when the count value cnt, which is the number of consecutive counts of frames in which the wall 31 is detected, reaches or exceeds the threshold value t, this section is determined to be a wall 31.

However, since the wall data W is 15 bits, even if the wall is actually continuing, if even one of the 15 bits is missed in the count (wall recognition), it may not be recognized as a wall even if the remaining data is correct, and two cages may be erroneously determined to be one cage. Therefore, a counter that does not detect the wall 31 is provided, and when the wall 31 is no longer detected (when the binary data switches to "0"), the counter starts counting, and if the count value cnt' is less than a threshold value s, for example 1, the wall 31 is determined to be continuing (the count value cnt continues to be added), and when it is 2 or more, the wall 31 is determined to have ended (the count value cnt stops being added). In FIG. 7(c), the center position of the count value cnt thus obtained is determined to be the center position of the wall in FIG. 7(b), as shown in FIG. 7(d).

However, the determination result of FIG. 7(d) may also vary in the number of images (frames) between the walls 31 if there is variation in the travel speed of the imaging device 7. As described above, this variation in travel speed is caused by the wind direction and volume of the ventilation and cooling air, wheel slippage, or differences in drive voltage and load (amount of food). Therefore, in this embodiment, the number of bits is consistent over the n cumulative times of images (frames) for the determination result of the center position of the wall shown in FIG. 7(d).

Specifically, an example of a data string extracted for the video images P1, P2, and Pn based on the judgment results of FIG. 7(d) is shown in FIG. 8(a). In FIG. 8(a), the width of room (cage) 1 is the same, and the position of the wall between room (cage) 2 is the same between videos P1, P2, and Pn. In contrast, there is a difference in the width of room (cage) 2, and the position of the wall between room (cage) 3 is the earliest in video P1 and the latest in video P2, with video Pn being between them. Then, in this embodiment, as shown in FIG. 8(b), dummy data d is inserted into the data string up to the length of the latest video P2. Specifically, based on video P2, dummy data d is inserted for two frames into video P1 and one frame into video Pn. By doing so, the timing of the appearance of the wall 31 between the rooms (cages) can be matched between videos P1, P2, and Pn. The data string for the center position of the wall after timing adjustment created in this way is indicated in Figure 5 by the reference characters w1, w2, ..., wn (hereinafter, when referring to them collectively, they will be indicated by the reference character w).

In the above example, to match the number of bits between the moving images P1, P2, and Pn, a dummy bit (frame) d is inserted to match the data with the largest number of bits (frames) (P2 in Figure 8(a)) without determining a specific number of bits (frames) for one cage. However, it is also possible to determine a specific number of bits (frames) for one cage (25 + 15 = 40 in the above explanation), and insert the dummy bit (frame) d if it is less than that, and delete the bit (frame) if it is more.

When accumulating images (frames) n times for the row of data H of dead chickens 43 shown in FIG. 6(e), timing adjustment is performed for the row of data w at the center position of this wall 31. That is, dummy data d in data w is similarly inserted into the corresponding data H of dead chickens 43 to adjust the bit length, and the data of wall 31 is also superimposed. In this way, data h1, h2, ..., hn of dead chickens 43 after position alignment (hereinafter, collectively referred to as h) is created.

Then, the data h of the dead chicken 43 after position alignment is accumulated (added in the chronological order) for n frames. An image of this accumulation is shown in Figure 9. Figure 9(a) shows an example of adding four frames with n=4, and to simplify the drawing, only the central position of the 15 bits of data for the wall 31 is shown, with the rest being set to "0" data. Similarly, the data inside the room (cage) is also shown with a reduced amount. Simply accumulating the data h for the four frames in Figure 9(a) results in the added data h0 shown in Figure 9(b). The change in Figure 9(b) is shown in a graph as shown in Figure 9(c).

As is clear from the graph, if the threshold value t for determining that a chicken is dead 43 is 2, then there should be a block, i.e. the period during which the integrated value exceeds 2 should be continuous to a certain extent, but there is a dip below the threshold during the period indicated by reference character B in Figure 9(c). The integrated value in Figure 9(b) is also "0", because at the same time, a non-detected "0" indicated by reference character E1 occurs due to a false detection in the data h for the provisional determination of a dead chicken 43 in each frame in Figure 9(a).

Even if such a false positive occurs, based on FIG. 7, in the counting of the count value cnt, in frames 1, 3, and 4, the count value cnt' is only "1". In other words, the dead chicken 43 is determined from the previous and following count values cnt. In contrast, in frame 2, based on FIG. 7, in the counting of the count value cnt shown by reference symbol E2, the count value cnt' counts four "0"s, and the dead chicken 43 is not detected. The "1" shown by reference symbol E3 is outside the cluster of "1"s, and is simply a false positive.

The false detection indicated by reference symbol E1 in FIG. 9(a) happens to occur at the same timing across the four frames that should be accumulated. However, in the example of FIG. 9, the non-detection "0" only occurs for one frame, so the added data h0 shown in FIG. 9(b) is "0" only for one frame period B, and in accordance with FIG. 7 described above, the count value cnt' is only 1, so no detection error occurs. However, if the non-detection "0" occurs for two frames, and the threshold value for the count value cnt' remains at 1, it will be erroneously determined (overlooked) that there is no dead chicken 43.

As mentioned above, the thresholds for the count values cnt and cnt' may be set appropriately based on the size of the mass of dead chickens 43 relative to the number of images (frames) of the inside of the cage 3, which is 25, and the number of images (frames) of the wall 31, per cage. The image capture device 7 has a lower frame rate than a normal moving image capture device (video camera), with one image (frame) being, for example, 0.1 seconds. However, if the thresholds for the count values cnt and cnt' are set stricter, the detection accuracy of dead chickens 43 and weak chickens 42 will increase, but it will take a long time to reach a judgment, and it will take a long time to judge especially weak chickens 42, although it will be almost certain that the judgment will be made. On the other hand, if the thresholds are relaxed, the judgment will be quicker, but there is a problem that there will be more erroneous judgments.

Therefore, in this embodiment, the above-mentioned added data h0, which is simply added up with the data h obtained by adjusting the reference position of the judgment data H of the dead chicken 43, is processed as follows. Figure 10 is a diagram explaining the data processing method. First, the added data h0 shown in Figure 9(b) above is shown in Figure 10(a) and is used as the reference data. As mentioned above, the reference position (timing) of this reference data is also adjusted based on the center position of the wall, as shown in Figure 8. In this embodiment, data is created by shifting this reference data forward or backward by one image (field). The rear-shifted data is shown in Figure 10(b), and the front-shifted data is shown in Figure 10(c). These three frames of data are then added together. The addition result is shown in Figure 10(d).

When the change in the data in FIG. 10(d) is shown in a graph similar to FIG. 9(c), it becomes FIG. 9(e). In contrast, the added data h0 shown in FIG. 9(b) and FIG. 10(a) is shown in FIG. 10(f). Comparing the graph in FIG. 10(f) with the graph in FIG. 10(e), it can be seen that the dip in the count value due to missed detection is compensated for and becomes equal to or greater than the threshold value t. The median of the count value cnt between the threshold value t and above is determined to be the location of dead chickens 43 and weak chickens 42. The column of data K of the determination result combining the center position of the wall and the location of the dead chickens 43 and weak chickens 42 is as shown in FIG. 10(g).

Then, the screen of the personal computer 111 or mobile terminal 12 displays the cage number in which the dead chicken 43 or weak chicken 42 is located. Therefore, the actual display data K', as shown in Figure 11, is data K in which the room (cage) number is counted each time the wall 31 is passed (determined), and the display data K' indicates in which room (cage) and where the dead chicken 43 or weak chicken 42 is located. In Figure 11, it is near the wall between room (cage) 2 and room (cage) 3.

Note that by adding the shift data shown in Figures 10(b) and (c) to the reference data shown in Figure 10(a), it is conceivable that the accuracy of the determination, such as the resolution, may be reduced. However, for one frame, which in this embodiment is about 0.1 seconds, the impact of such a reduction in accuracy is small, and the effect of compensating for the dip described above is far greater.

Here, the search results for optimal parameters for the count value cnt of the number of consecutive frames and the threshold value t for determining whether or not there are dead chickens 43 or weak chickens 42 are shown in Figure 12. The experiment was conducted on data for five days with four rows (mountains) of cages 2. The number of accumulated frames n was set to 6. That is, the chickens passed in front of the cages 3 12 times during six feedings per day, but only the outward journey was imaged. The correct determination of the presence of dead chickens 43 or weak chickens 42 was made by manually observing the images captured by the imaging device 7 and actually removing them. The count value cnt was changed from 1, i.e., when it exceeded the threshold value t even for a moment, to 16, i.e., when it exceeded the threshold value t in a frame equivalent to the width of the actual wall 31. The maximum value of the threshold value t was set to a maximum of 18, because the accumulated value may be tripled due to n=6 and the shift addition in Figure 10. The score used in Figure 12 is the f1-score, which is the harmonic mean of precision and recall.

From Figure 12, as shown in Figure 13, the count value cnt of the number of consecutive frames is 10, i.e., about 2/3 of the 15 frames in which the wall 31 may be visible, and the threshold value t is 8, i.e., after approximately three judgments, the score is 90 or more. More specifically, the matching rate is 100% or more and the recall rate is 90%. These parameters may be determined according to the actual setup of the chicken coop 1 and the breeding environment, etc.

As described above, in the dead chicken detection device of this embodiment and the poultry farming system using it, to detect dead chickens 43 inside cages 3, multiple cages 3 lined up in a lateral (horizontal) direction are continuously photographed by a moving imaging device 7, and the dead chickens 43 are judged by analysis using AI or the like. The judgment result from one image is treated as a provisional judgment result, and a predetermined number n of times (in the chronological order) are accumulated with the timing adjusted so that the wire mesh parts of the cages match, and if the number is equal to or exceeds the threshold value t, it is judged to be an actual dead chicken.

Therefore, dead chickens 43 are judged from the results of repeated judgments at the same location but at different times, and so at least the number of movements (image capture) equal to the threshold value t is required before a judgment can be made, which takes some time, but still allows for accurate detection of dead chickens 43 through to weak chickens 42 relatively quickly. Then, if an operator follows the judgment results and goes to the corresponding cage, he or she will be sure to find dead chickens 43 and weak chickens 42 there, which improves the efficiency of the work of removing them, and also eliminates false starts, earning the operator's trust.

In addition, in the dead chicken detection device of this embodiment, a large-scale chicken coop 1 is made up of cages 3 housing many chickens 4 lined up in multiple tiers in the horizontal and vertical directions, and a feeding device 6 is provided that automatically feeds the chickens by moving in front of each cage 3 at predetermined times along rails 61 laid across the cages and dropping food into feed receptacles (feeding trays) 51 provided in each cage 3. The imaging device 7 is installed near the feeding opening 64 of each tier of the feeding device 6.

On the other hand, chickens 4 stand and move actively when eating, so if you observe the chickens at this feeding time, especially by looking at the chickens' 4 feet from a low viewpoint near the feed tray 51, you can efficiently determine whether the chickens 43 are dead or not, and even their health condition, whether the chickens 42 are weak from summer fatigue or the like. Specifically, you can determine whether the chickens are dead or not by looking at the abdomen or comb 431 at the feeding time, or by the movement of the legs 411, which should move a little if they are standing. In this way, you can find dead chickens 43 and weak chickens 42 relatively accurately and easily, simply by adjusting the timing of observing the chickens 4. Also, by observing the chickens at frequent feeding times, you can quickly find dead chickens and weak chickens 42.

Moreover, the feed receiver (feeding tray) 51 is formed continuously in the horizontal direction like a gutter, which can be an obstacle when observing the area around the legs 411 of the chickens 4 from the low viewpoint. However, it is different from the egg receiver 52, which is also formed continuously in the horizontal direction like a gutter. Therefore, in the dead chicken detection device of this embodiment, the imaging device 7 is installed so that it can look through the gap between the feed receiver and the egg receiver. This is preferable because it allows continuous, unobstructed observation of the area around the legs 411 of the chickens 4 between each cage 3.

Here, it is not necessary to find dead chickens 43 and weak chickens 42 in real time, and the actual removal work is performed about once a day, so it is sufficient if there are no oversights. In other words, it is desirable to detect dead chickens 43 quickly, but a detection delay of about the above time does not cause a big problem. Therefore, the image captured by the imaging device 7 is recorded in the recording device 71, and when the feeding device 6 returns to the home position and replenishes feed from the replenishment machine 101, the captured image data is collected from the recording device 71 by the wireless communication 103 and sucked up in the determination means 111, and analysis is performed each time, or after a day's worth of data to be accumulated is collected. In this way, there is no need to lay wiring for detecting dead chickens 43 in the feeding device 6, and the feeding device 6 only needs to be laid with conventional power lines and necessary control lines. In addition, the wireless communication 103 can be a simple short-distance wireless communication such as Wi-Fi (registered trademark), and there is no need to install large-scale wireless communication equipment to ensure that radio waves reach every corner of the large (long) chicken coop 1.

In this embodiment, the recording device 71 is not provided for each imaging device 7, and in the example of FIG. 1, it is shared by four devices mounted on the same feeding device 6. In recent years, the image quality of even general-purpose (cheap) imaging devices has improved, and there is no problem with AI judgment even in the case of a composite screen. This ultimately makes it possible to shorten the judgment time per screen and reduce the capacity of the recording device 71.

Furthermore, in the dead chicken detection device of this embodiment, the results of the provisional judgment are accumulated multiple times n times to make the actual judgment, and the provisional judgment involves making a binary judgment for each frame as to whether or not it is a dead chicken 43, as well as a binary judgment as to whether or not it is passing through the wall 31, which serves as the moving position reference, and the accumulation is performed by shifting the time axis of the provisional judgment results, that is, adjusting the timing between frames, so that the moving position references match. Therefore, even if the running speed of the feeding device 6 varies and the number of image frames between the moving position references (one cage) differs, the time difference can be adjusted, and the binary provisional judgment results as to whether or not it is a dead chicken 43 can be accurately accumulated in chronological order for a predetermined number of frames n.

Furthermore, when adjusting the timing and accumulating the provisional judgment results, a dummy bit (frame) d is inserted between the movement position references, i.e., in one cage, in accordance with the data with the largest number of bits (frames), so that the variation in the movement speed (number of frames) of the imaging device 7 can be compensated for.

Furthermore, the result of the accumulation (count value) is used as the reference data, and data shifted one frame forward and backward is created, and these three data are added together and compared with the threshold value t. Therefore, even if a dead chicken happens to be missed at the same timing and this is reflected in the result of the accumulation (count value), the missed detection can be compensated for by the data before and after, allowing for an accurate judgment.

In the above explanation, the cage number is obtained from the counting result of the wall 31, but the cage number may be obtained by reading the numbers or barcode displayed on each cage 3 using the imaging device 7 itself or a separate reader, or may be obtained as travel position data from the control device of the feeding device 6 together with information on the level of the cage 3 that the imaging device 7 corresponds to.

1 Poultry house 10 Floor surface 101 Feed refill machine 102 Router 103 Wireless communication 104 Communication line 11 Office 111 Determination means 112 Server 113 Public wireless communication line 12 Mobile terminal 2 Cage chicken house 3 Cage 4 Chicken 41 Healthy chicken 411 Leg 42 Weak chicken 43 Dead chicken 431 Cockscomb 51 Feed tray (feeding tray)
52 Egg receiver 53 Water supply pipe 55 Net plate 56 Feces removal belt 6 Feeding device 61 Rail 62 Wheels 64 Feeding port 7 Imaging device 71 Recording device

Claims (7)

A device for detecting dead chickens in a cage housing chickens,
A plurality of the cages are arranged at least in the horizontal direction, and an imaging device that captures continuous images of the inside of the cage while moving in the horizontal direction;
A first determination means for analyzing an image captured by the imaging device and provisionally determining whether or not the image is a dead chicken;
and a second determination means for accumulating the determination results of the first determination means for a predetermined number of times by matching a moving position reference of the imaging device, and determining that the chicken is actually dead when the result is equal to or greater than a predetermined threshold value;
The first determination means performs the provisional determination by analyzing continuous images of the imaging device, and performs a binary determination for each frame as to whether or not the chicken is a dead chicken, and also performs a binary determination as to whether or not the chicken is passing through the moving position reference;
The second judgment means is a dead chicken detection device characterized in that it shifts the time axis of the result of the binary judgment of whether or not the chicken is a dead chicken and performs the accumulation for the predetermined number of times so that the moving position standard is consistent. The dead chicken detection device according to claim 1, characterized in that the imaging device is provided on the feeding device or on a cart that follows the feeding device. The dead chicken detection device according to claim 1, characterized in that the imaging device observes the inside of the cage through a gap between a food tray located relatively above and an egg tray located relatively below. 3. The dead chicken detection device according to claim 2, wherein the imaging device is provided in the feeding apparatus, the feeding apparatus further comprises a recording device for recording images captured by the imaging device, and the images captured by the recording device are transferred to the first determination means at a starting point for replenishing feed to the feeding apparatus. The dead chicken detection device according to claim 1, characterized in that the second determination means performs integration of the binary determination result of whether or not the chicken is dead a predetermined number of times in accordance with a predetermined number of bits. The dead chicken detection device according to claim 5, characterized in that the second determination means creates data shifted one frame forward and backward using the result of the accumulation as reference data, and compares the result of adding these three data together with the predetermined threshold value. A poultry farming system comprising the dead chicken detection device according to any one of claims 1 to 3.

Images