JP6606858B2 - Object detection apparatus and object detection method - Google Patents

Description

  The present invention relates to an object detection device and an object detection method.

  Conventionally, a technique called Snakes exists as a technique for estimating the size of an object shown in a captured image captured by an imaging apparatus (see, for example, Non-Patent Document 1). In this method, while changing the size of the object model corresponding to the detection target, the object model is fitted to the silhouette image generated from the captured image, and the size of the object is estimated from the size of the object model that fits the silhouette image. To do. A technique for estimating the size of a vehicle using a vehicle model is also disclosed (for example, see Patent Document 1).

JP2013-148971A

Satoshi Eito, “Overview of Active Contour Model Snakes”, MEDICAL TECHNOLOGY Vol. 12, no. 1. January 1994

  However, an object other than the detection target object may also appear in the captured image. In such a case, the degree of coincidence of the object model with a part of the silhouette of the detection object and a part of the silhouette of the object other than the detection object may increase, and the fitting accuracy of the object model may not be improved. Therefore, it is desired to provide a technique for improving the fitting accuracy of the object model.

In order to solve the above problem, according to one aspect of the present invention, a silhouette image generation unit that generates a silhouette image based on a captured image obtained by capturing a three-dimensional space by an imaging device, and a detection target Based on a plurality of corresponding different predetermined directions and edges extracted from the silhouette image, a direction-specific edge image generation unit that generates a plurality of direction-specific edge images, the plurality of direction-specific edge images, and the detection object And an object detection unit that detects whether or not the detection target exists in the captured image, and the edge image generation unit for each direction uses the contour of the object model. When a direction in which the gradient directions of pixels constituting the edge match or are similar to each other is detected from a predetermined direction corresponding to each of a plurality of line segments to be formed, the pixel is detected. Determining a corresponding pixel in each direction edge image corresponding to the direction, the object detection apparatus is provided.

  The detection target may be a detection target vehicle existing on a road plane reflected in the captured image, and the object detection unit is based on the edge image for each direction and a vehicle model corresponding to the detection target vehicle. It may be detected whether the detection target vehicle exists in the captured image.

  The edge image generation unit by direction is based on a predetermined direction corresponding to each of a plurality of line segments forming an outline of the vehicle model in an orientation according to the road plane, and an edge extracted from the silhouette image. The direction-specific edge image may be generated.

  The object detection unit calculates a degree of coincidence of the vehicle model with respect to an edge of the edge image for each direction for each of the plurality of line segments, and whether or not a total value of the degree of coincidence for the plurality of line segments exceeds a threshold value. You may judge whether.

  The object detection unit may calculate the degree of coincidence of the edge image for each direction with respect to the contour of the vehicle model for each of the plurality of line segments, and search for the vehicle model in which the total value of the degree of coincidence is maximized.

  The object detection unit may search for the vehicle model that maximizes the total value of the degree of coincidence while changing at least one of the position and size of the vehicle model.

  The predetermined direction corresponding to each of the plurality of line segments may be a vertical direction with respect to each of the plurality of line segments.

According to another aspect of the present invention, a silhouette image is generated based on a captured image obtained by imaging a three-dimensional space by an imaging device, and a plurality of different predetermined directions corresponding to a detection target object. based on the edge extracted by the silhouette image includes generating a plurality of directions by the edge image, based on the object model corresponding to the detection object and the plurality of directions by the edge image, the captured image wherein a the detection object is detected whether or not there, only including, from a predetermined direction corresponding to a plurality of line segments that form the outline of the object model, the gradient direction of the pixels constituting the edge If There detecting a direction in which identical or similar, comprising determining the pixel as the corresponding pixels in each direction edge image corresponding to the direction in which the detected object detection method provides It is.

  As described above, according to the present invention, it is possible to improve the fitting accuracy of an object model.

It is a 1st figure for demonstrating the background of embodiment of this invention. It is a 2nd figure for demonstrating the background of embodiment of this invention. It is a figure which shows the structural example of a detection system. It is a figure which shows the function structural example of an object detection apparatus. It is a figure which shows the example of the captured image imaged with the imaging device. It is a figure which shows the example of the silhouette image produced | generated based on a captured image. It is a figure which shows the example of a deformation | transformation of the vehicle model corresponding to a detection object vehicle. It is a figure for demonstrating the example of the predetermined direction corresponding to each of several line segments which form the outline of a vehicle model. It is a figure which shows the example of the edge image produced | generated by extracting an edge from a silhouette image. It is a figure which shows the example of several edge image according to a direction. It is a figure which shows the example which calculates the coincidence degree of the vehicle model with respect to the edge of the edge image according to direction about each of several line segments which form the outline of a vehicle model. It is a figure which shows the example of vehicle model Md3 'from which the total value of a coincidence becomes the maximum. It is a flowchart which shows the example of the flow of operation | movement of the object detection apparatus 10 which concerns on embodiment of this invention.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

  In the present specification and drawings, a plurality of constituent elements having substantially the same functional configuration are distinguished by attaching different numerals to the same reference numerals. Further, similar constituent elements of different embodiments are distinguished by attaching different alphabets after the same reference numerals. However, when there is no need to particularly distinguish each of a plurality of constituent elements having substantially the same functional configuration, only the same reference numerals are given.

(1. Background)
First, the background of the embodiment of the present invention will be described. FIG. 1 is a first diagram for explaining the background of the embodiment of the present invention. Referring to FIG. 1, a captured image Img1 is shown in which a situation in which a vehicle V1 that is a large truck travels in a back lane and a vehicle V2 that is a passenger vehicle travels in a front lane is illustrated. Here, in the captured image Img1, the vehicle V1 and the vehicle V2 are close to each other. Assume that it is detected whether or not there is a detection target vehicle in the lane on the near side in such a situation that the vehicles V1 and V2 that are close to each other are imaged.

  FIG. 2 is a second diagram for explaining the background of the embodiment of the present invention. When detecting whether or not a detection target vehicle exists in the lane on the front side, first, as shown in FIG. 2, a silhouette image Et1 is generated from a captured image Img1 obtained by capturing the vehicle V1 and the vehicle V2. The silhouette image Et1 includes a silhouette Si0 in which the silhouette of the vehicle V1 and the silhouette of the vehicle V2 are continuous. Originally, the degree of coincidence of the vehicle model Md21 with the silhouette of the vehicle V2 traveling on the near lane in the silhouette Si0 should be calculated.

  However, as shown in FIG. 2, in this silhouette Si0, the silhouette of the vehicle V1 traveling in the back lane overlaps the silhouette of the vehicle V2 traveling in the front lane. In such a case, the degree of coincidence of the vehicle model Md22 with a part of the silhouette of the vehicle V1 that travels in the back lane and a part of the silhouette of the vehicle V2 that travels in the front lane increases, and the fitting accuracy is high. May not improve. Therefore, the present specification mainly proposes a technique for improving the fitting accuracy.

  FIG. 2 shows an example in which the silhouette of the vehicle V1 traveling in the back lane overlaps the silhouette of the vehicle V2 traveling in the front lane. In such a case, the technique according to the embodiment of the present invention can be applied effectively, but the silhouette of the vehicle V1 traveling in the back lane may not overlap the silhouette of the vehicle V2 traveling in the front lane. . That is, in the embodiment of the present invention, the degree of approach between the vehicle V1 traveling in the back lane and the vehicle V2 traveling in the front lane is not particularly limited. Further, the vehicle V1 and the vehicle V2 do not need to be separated from each other in the back and the front, and may be separated in any direction.

(2. System configuration)
Then, the structural example of the detection system 1 which concerns on embodiment of this invention is demonstrated. FIG. 3 is a diagram illustrating a configuration example of the detection system 1. As illustrated in FIG. 3, the detection system 1 includes an object detection device 10 and an imaging device 20. As shown in FIG. 3, the object detection device 10 and the imaging device 20 may typically be connected by radio or wire, but may be integrated in the same casing. The imaging device 20 is installed in a three-dimensional space (real space) and can generate a captured image by imaging the three-dimensional space.

  As shown in FIG. 3, a part of the road plane P is within the imaging range of the imaging device 20, and when the vehicle V <b> 3 traveling on the road plane P is within the imaging range, the imaging device 20 It is possible to image. Here, the place where the imaging device 20 is installed is not particularly limited as long as it is a position where the road plane P can be imaged. However, as shown in FIG.

  The object detection device 10 detects whether or not a detection target vehicle exists in the imaging range based on the captured image captured by the imaging device 20. At this time, the object detection apparatus 10 fits a vehicle model corresponding to the detection target vehicle to the silhouette image generated from the captured image. When fitting the vehicle model, parameters calculated by calibration performed when the imaging device 20 is installed (for example, an expression representing the road plane P in the three-dimensional space, the traveling direction of the detection target vehicle in the three-dimensional space, etc.) Can be used.

  Note that a method for obtaining parameters such as an expression representing the road plane P and a traveling direction of the detection target vehicle is not particularly limited, but as an example, the technique disclosed in Japanese Patent Application Laid-Open No. 2013-148971 can be employed. is there. Further, in this specification, an example in which one lane is provided on the road plane P will be described, but a plurality of lanes may be provided on the road plane P. When a plurality of lanes are provided on the road plane P, the traveling direction of each of the plurality of lanes may be obtained as a parameter.

(3. Functional configuration example)
Subsequently, a functional configuration example of the object detection device 10 according to the embodiment of the present invention will be described. FIG. 4 is a diagram illustrating a functional configuration example of the object detection apparatus 10. As shown in FIG. 4, the object detection apparatus 10 includes a control unit 110, an input unit 170, a storage unit 180, and an output unit 190. The control unit 110 has a function of controlling the entire operation of the object detection device 10. The input unit 170 receives a captured image captured by the imaging device 20.

  The storage unit 180 can store a program and data for operating the control unit 110. In addition, the storage unit 180 can temporarily store various data necessary in the course of the operation of the control unit 110. The output unit 190 has a function of performing output in accordance with control by the control unit 110. The type of the output unit 190 is not particularly limited, and may be a display device, an audio output device, or a communication device that transmits information to a server (not shown).

  In the example illustrated in FIG. 4, the input unit 170, the storage unit 180, and the output unit 190 exist inside the object detection apparatus 10, but all or part of the input unit 170, the storage unit 180, and the output unit 190. May be provided outside the object detection device 10. The control unit 110 includes an image acquisition unit 111, a silhouette image generation unit 112, a direction-specific edge image generation unit 113, an object detection unit 114, and an output control unit 115. Details of these functional units included in the control unit 110 will be described later.

(4. Function details)
Subsequently, the function details of the object detection device 10 will be described. FIG. 5 is a diagram illustrating an example of a captured image Img2 captured by the imaging device 20. Referring to FIG. 5, the captured image Img2 shows a vehicle V3 traveling on the road plane P. When the imaging device 20 captures the captured image Img2, the captured image Img2 is output to the object detection device 10. In the object detection device 10, the captured image Img <b> 2 is received by the input unit 170, and the captured image Img <b> 2 received by the input unit 170 is acquired by the image acquisition unit 111.

  Subsequently, the silhouette image generation unit 112 generates a silhouette image based on the captured image Img2. FIG. 6 is a diagram illustrating an example of a silhouette image generated based on the captured image Img2. As illustrated in FIG. 6, the silhouette image generation unit 112 can extract the silhouette Si3 from the captured image Img2. The silhouette Si3 corresponds to the silhouette of the vehicle V3 shown in the captured image Img2.

  The silhouette image generation unit 112 may extract the silhouette Si3 in any way. As an example, the silhouette image generation unit 112 may extract, as the silhouette Si3, a set of pixels whose difference value from the background image is larger than the threshold value in the captured image Img2 based on the background difference method. In addition, although the method by which the silhouette image generation unit 112 generates a silhouette image is not limited, as an example, the silhouette image generation unit 112 binarizes the captured image Img2 depending on whether it is a region of the silhouette Si3, The silhouette image Et2 may be generated.

  Further, in the following description, a case where the silhouette Si3 is extracted from the captured image Img2 by the silhouette image generation unit 112 will be described as an example, but a case where a silhouette is not extracted from the captured image Img2 is also assumed. In such a case, the operation may be shifted to processing for a captured image corresponding to the next frame. Alternatively, when detecting whether or not a detection target vehicle exists in each of a plurality of lanes, the operation may be shifted to processing for the next lane.

  Subsequently, the direction-specific edge image generation unit 113 deforms the vehicle model corresponding to the detection target vehicle according to the road plane P. The detection target vehicle and the vehicle model are associated in advance. FIG. 7 is a diagram illustrating an example of deformation of the vehicle model corresponding to the detection target vehicle. For example, the direction-specific edge image generation unit 113 may generate the vehicle model Md3 by modifying the vehicle model Md3 # according to the equation of the road plane P.

  More specifically, as shown in FIG. 7, the edge image generation unit 113 by direction aligns the direction of the vehicle model Md3 # with the traveling direction R in the three-dimensional space, and the road plane P ( Alternatively, the vehicle model Md3 may be generated by changing the size of the vehicle model Md3 # according to the distance to the lane). For example, the smaller the distance, the larger the detection target vehicle should appear in the captured image. Therefore, the vehicle model Md3 # may be modified such that the smaller the distance, the larger the size of the vehicle model Md3.

  FIG. 7 shows an example in which both the direction and the size of the vehicle model Md3 # are changed. However, both the direction and size of the vehicle model Md3 # need not be changed. For example, at least one of the orientation and size of the vehicle model Md3 # may be changed. Further, when the vehicle model Md3 is prepared in advance, there is no particular process for transforming the vehicle model Md3 # into the vehicle model Md3.

  Further, the distance from the imaging device 20 to the road plane P (or lane) may be set in any way. For example, the distance from the imaging device 20 to the road plane P (or lane) may be set manually in advance. Alternatively, the distance from the imaging device 20 to the road plane P (or lane) may be calculated by automatically analyzing a captured image captured by the imaging device 20 at the time of calibration or the like.

  Moreover, in this specification, since it is assumed that the detection target vehicle is a bus having a simple shape, an example in which the vehicle model Md3 # has a simple shape of a rectangular parallelepiped will be mainly described. However, the shape of the vehicle model Md3 # is not limited to a rectangular parallelepiped. For example, the shape of the vehicle model Md3 # may have a more complicated shape than a rectangular parallelepiped. That is, the shape of the vehicle model Md3 # may be changed as appropriate according to the shape of the detection target vehicle.

  Subsequently, the direction-specific edge image generation unit 113 calculates a predetermined direction corresponding to the detection target vehicle. The predetermined direction corresponding to the detection target vehicle may be one, but may be plural. That is, the direction-specific edge image generation unit 113 may calculate a plurality of different predetermined directions corresponding to the detection target vehicle. At this time, for example, the direction-specific edge image generation unit 113 may calculate a predetermined direction corresponding to each of a plurality of line segments forming the contour of the vehicle model Md3.

  FIG. 8 is a diagram for describing an example of a predetermined direction corresponding to each of a plurality of line segments forming the contour of the vehicle model Md3. Referring to FIG. 8, the outline of the vehicle model Md3 is formed by six line segments. The direction-specific edge image generation unit 113 can calculate the vertical directions D1 to D6 for each of the plurality of line segments as predetermined directions corresponding to the plurality of line segments that form the contour of the vehicle model Md3.

  In the following description, a case where vertical directions D1 to D6 for each of a plurality of line segments forming the contour of the vehicle model Md3 are mainly used as the predetermined direction corresponding to the detection target vehicle will be described. However, a direction different from the vertical directions D1 to D6 may be used. For example, instead of the vertical directions D1 to D6, a predetermined direction corresponding to the detection target vehicle may be used, a plurality of different predetermined directions corresponding to the detection target vehicle may be used, or the vehicle model Md3 A predetermined direction corresponding to each of the plurality of line segments forming the contour may be used.

  In the present specification, a rectangular parallelepiped is used as the shape of the vehicle model Md3. Therefore, the number of line segments forming the contour of the vehicle model Md3 is six. However, as described above, the shape of the vehicle model Md3 is not limited to a rectangular parallelepiped. Therefore, the number of the plurality of line segments forming the contour of the vehicle model Md3 may be changed as appropriate according to the shape of the vehicle model Md3.

  Subsequently, the direction-specific edge image generation unit 113 generates an edge image by extracting an edge from the silhouette image Et2. FIG. 9 is a diagram illustrating an example of the direction-specific edge image Eg2 generated by extracting an edge from the silhouette image Et2. The method for extracting an edge from the silhouette image Et2 is not particularly limited. Subsequently, the direction-specific edge image generation unit 113 calculates the gradient direction of each of some or all of the pixels constituting the edge. FIG. 9 shows gradient directions d1 to d6 in six pixels as examples of gradient directions.

  Subsequently, the direction-specific edge image generation unit 113 generates a plurality of direction-specific edge images based on the vertical direction with respect to each of the plurality of line segments forming the contour of the vehicle model Md3 and the edge extracted from the silhouette image Et2. To do. For example, it is assumed that the direction-specific edge image generation unit 113 detects a direction in which the gradient directions of pixels constituting the edge match or are similar to each other from the vertical direction with respect to each of a plurality of line segments forming the contour of the vehicle model Md3. . A similar range may be determined in advance.

  In such a case, the direction-specific edge image generation unit 113 may determine the pixel as a corresponding pixel in the direction-specific edge image corresponding to the detected direction. FIG. 10 is a diagram illustrating an example of a plurality of direction edge images. For example, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D1 with respect to the line segment forming the contour of the vehicle model Md3 matches or is similar to the gradient direction d1 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg21 corresponding to the vertical direction D1.

  Similarly, for example, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D2 matches or is similar to the gradient direction d2 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg22 corresponding to the vertical direction D2. Further, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D3 matches or is similar to the gradient direction d3 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg23 corresponding to the vertical direction D3.

  Further, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D4 matches or is similar to the gradient direction d4 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg24 corresponding to the vertical direction D4. Further, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D5 matches or is similar to the gradient direction d5 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg25 corresponding to the vertical direction D5.

  Further, it is assumed that the direction-specific edge image generation unit 113 detects that the vertical direction D6 matches or is similar to the gradient direction d6 of the pixels constituting the edge. In that case, the direction-specific edge image generation unit 113 determines the pixel as a corresponding pixel in the direction-specific edge image Eg26 corresponding to the vertical direction D6. In this manner, the direction-specific edge image generation unit 113 classifies some or all of the pixels constituting the edge into one of the vertical directions D1 to D6 depending on the gradient direction.

  Subsequently, the object detection unit 114 detects whether or not the detection target vehicle is present in the captured image Img2 based on the direction-specific edge images Eg21 to Eg26 and the vehicle model Md3 corresponding to the detection target vehicle. As described above, by considering the gradient direction of the edge, among the plurality of pixels constituting the edge, the gradient direction does not match the vertical direction of each of the plurality of line segments forming the contour of the vehicle model Md3 or Pixels that are dissimilar are excluded from the fitting target of the vehicle model Md3. For this reason, it is possible to improve the fitting accuracy of the vehicle model Md3.

  More specifically, the object detection unit 114 calculates the degree of coincidence of the vehicle model Md3 with respect to the edges of the edge images by direction Eg21 to Eg26 for each of a plurality of line segments that form the contour of the vehicle model Md3, and the plurality of lines It is determined whether or not the total value of the degree of coincidence for the minutes exceeds a threshold value.

  At this time, when the total value of the coincidence degree of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 exceeds the threshold, the object detection unit 114 determines that the detection target vehicle corresponding to the vehicle model Md3 is the road plane P ( Alternatively, it may be recognized as existing in the lane). On the other hand, when the total value of the coincidence degree of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 does not exceed the threshold, the object detection unit 114 determines that the detection target vehicle corresponding to the vehicle model Md3 is the road plane P ( Alternatively, it may be recognized that it does not exist in the lane).

  FIG. 11 is a diagram illustrating an example in which the degree of coincidence of the vehicle model Md3 with the edges of the direction-specific edge images Eg21 to Eg26 is calculated for each of a plurality of line segments that form the contour of the vehicle model Md3. FIG. 11 shows edge images Eg21 to Eg26 for each direction. The object detection unit 114 matches the line segment Ln1 that forms the contour of the vehicle model Md3 with how much the edge image Eg21 by direction that is a set of pixels having a gradient direction that matches or is similar to the vertical direction of the line segment Ln1. The degree of coincidence indicating whether to do is calculated.

  Similarly, the object detection unit 114 obtains a directional edge image Eg22 that is a set of pixels having a gradient direction that matches or is similar to the vertical direction of the line segment Ln2 with respect to the line segment Ln2 that forms the contour of the vehicle model Md3. The degree of coincidence indicating how much they match is calculated. The object detection unit 114 also determines which edge image Eg23 by direction is a set of pixels having a gradient direction that matches or is similar to the vertical direction of the line segment Ln3 with respect to the line segment Ln3 forming the contour of the vehicle model Md3. The degree of coincidence indicating whether or not the degree matches is calculated.

  In addition, the object detection unit 114 determines which edge image Eg24 by direction, which is a set of pixels having a gradient direction that matches or is similar to the vertical direction of the line segment Ln4, with respect to the line segment Ln4 that forms the contour of the vehicle model Md3. The degree of coincidence indicating whether or not the degree matches is calculated. In addition, the object detection unit 114 determines which edge image Eg25 by direction is a set of pixels having gradient directions that match or are similar to the vertical direction of the line segment Ln5 with respect to the line segment Ln5 forming the contour of the vehicle model Md3. The degree of coincidence indicating whether or not the degree matches is calculated.

  In addition, the object detection unit 114 determines which direction edge image Eg26 is a set of pixels having a gradient direction that matches or is similar to the vertical direction of the line segment Ln6 with respect to the line segment Ln6 that forms the contour of the vehicle model Md3. The degree of coincidence indicating whether or not the degree matches is calculated. Then, the object detection unit 114 calculates the total value of the matching degrees by summing up the matching degrees. Subsequently, the object detection unit 114 searches for a vehicle model that maximizes the total value of the matching degree while changing at least one of the position and the size of the vehicle model Md3.

  FIG. 12 is a diagram illustrating an example of the vehicle model Md3 ′ having the maximum coincidence value. As shown in FIG. 12, the captured image Img2 ′ shows a vehicle model Md3 ′ having the maximum matching value. Here, the object detection unit 114 may change the position of the vehicle model Md3 in the traveling direction R when changing the position of the vehicle model Md3. In order to change the size of the vehicle model Md3, the vertical, horizontal, and height of the vehicle model Md3 may be freely changed independently, and any one of the vertical, horizontal, and height relationships is , You may have some constraint condition.

  In addition, the object detection unit 114 can determine various sizes from the vehicle model Md3 ′ having the maximum matching value. For example, the object detection unit 114 may determine the vertical direction of the vehicle model Md3 'as the vehicle width of the vehicle V3 and the horizontal direction of the vehicle model Md3' as the vehicle length of the vehicle V3. Alternatively, for example, the object detection unit 114 may determine the height of the vehicle model Md3 'as the height excluding the minimum ground height of the vehicle V3. Furthermore, the object detection unit 114 can also determine the vehicle type of the vehicle V3 based on these sizes.

  Note that a method for calculating the degree of coincidence of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 is not particularly limited. As an example, published literature (Shunta Saito et al., “Model-Based Human Body Trunk 3D Shape Estimation”, Information Processing Society of Japan Research Report, Vol. 2011-CVIM-178 No. 16, September 5, 2011) The degree of coincidence of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 can be calculated by using the evaluation function in the fitting.

  Further, the method for calculating the vehicle model Md3 'that maximizes the total value of the coincidence of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 is not particularly limited. As an example, a vehicle model Md3 in which the total value of the degree of coincidence of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26 is maximized using optimization in the fitting of “model-based human body trunk 3D shape estimation”. 'Can be calculated. However, other methods (for example, downhill simplex method) may be used as the optimization algorithm.

  Subsequently, the output control unit 115 may cause the output unit 190 to output the detection result obtained by the object detection unit 114. For example, the detection result may be information indicating whether or not a detection target vehicle exists in the captured image Im2. Alternatively, the detection result may be the degree of coincidence of the vehicle model Md3 with respect to the edges of the direction-specific edge images Eg21 to Eg26, or the shape of the vehicle model having the maximum degree of coincidence with the edges of the direction-specific edge images Eg21 to Eg26. , Size, vehicle type, etc.

(5. Example of operation)
Subsequently, an operation example of the object detection apparatus 10 will be described. FIG. 13 is a flowchart illustrating an example of an operation flow of the object detection apparatus 10 according to the embodiment of the present invention. Note that the flowchart shown in FIG. 13 merely shows an example of the operation flow of the object detection apparatus 10 according to the embodiment of the present invention. Therefore, the flow of the operation of the object detection apparatus 10 according to the embodiment of the present invention is not limited to the flowchart shown in FIG.

  First, as illustrated in FIG. 13, the image acquisition unit 111 acquires a captured image captured by the imaging device 20 via the input unit 170 (step S <b> 11). Subsequently, the silhouette image generation unit 112 generates a silhouette image based on the captured image (step S12). Subsequently, the direction-specific edge image generation unit 113 extracts an edge from the silhouette image (step S13). Subsequently, the direction-specific edge image generation unit 113 generates a direction-specific edge image based on the predetermined direction and the edge with respect to the detection target vehicle (step S15).

  Subsequently, the object detection unit 114 detects whether or not a detection target vehicle exists in the captured image based on the edge image classified by direction and the vehicle model (step 15). At this time, the output control unit 115 may cause the output unit 190 to output the detection result obtained by the object detection unit 114. If the object detection unit 114 continues the operation (“Yes” in step S16), the object detection unit 114 shifts the operation to step S11. If the operation is not continued (“No” in step S16), the object detection unit 114 ends the operation. Let

  The operation example of the object detection apparatus 10 according to the embodiment of the present invention has been described.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

  For example, in the present specification, the case where the detection target vehicle is a bus will be described as an example in consideration of the simplicity of the vehicle model, but the detection target vehicle may be a vehicle other than the bus. For example, the detection target vehicle may be an automobile or a truck. Further, instead of the detection target vehicle, a detection target other than the detection target vehicle may be used. The detection target may be a train or a ship. At this time, an object model may be used instead of the vehicle model. Further, the object may be moving or may be stationary.

  In the above description, the example in which the silhouette Si3 is unconditionally generated when the captured image Img2 is obtained from the imaging device 20 has been described. However, the silhouette Si3 may be generated when a predetermined condition is satisfied. For example, when a vehicle is detected from the captured image Img2, a silhouette Si3 is generated, and when no vehicle is detected from the captured image Img2, a silhouette may not be generated. At this time, the vehicle detection method is not particularly limited.

  In addition, for example, a control unit (not shown) for controlling the entire operation of the object detection device 10 may exist inside the object detection device 10. The control unit (not shown) may be configured by dedicated hardware, or a program stored in a ROM (Read Only Memory) by a CPU (Central Processing Unit) built in the object detection apparatus 10 is stored in a RAM (Random Access). It may be realized by expanding to Memory and executing. In addition to providing such a program, a storage medium storing such a program may also be provided.

DESCRIPTION OF SYMBOLS 1 Detection system 10 Object detection apparatus 20 Imaging device 110 Control part 111 Image acquisition part 112 Silhouette image generation part 113 Edge image generation part by direction 114 Object detection part 115 Output control part 170 Input part 180 Storage part 190 Output part D1-D6 Vertical Direction d1 to d6 Gradient direction Eg2 Edge image Eg21 to Eg26 Edge image by direction Et1, Et2 Silhouette image Img1, Img2, Img2 'Captured image Ln1 to Ln6 Line segment Md21, Md22, Md3, Md3' Vehicle model P Road plane R Running direction Si0, Si3 silhouette V1-V3 vehicle

Claims (8)

A silhouette image generation unit that generates a silhouette image based on a captured image obtained by imaging a three-dimensional space by an imaging device;
A direction-specific edge image generation unit that generates a plurality of direction-specific edge images based on a plurality of different predetermined directions corresponding to a detection target and edges extracted from the silhouette image ;
An object detection unit configured to detect whether or not the detection target exists in the captured image based on the plurality of edge images by direction and an object model corresponding to the detection target;
Equipped with a,
When the direction-specific edge image generation unit detects a direction in which gradient directions of pixels constituting the edge match or are similar to each other from predetermined directions corresponding to a plurality of line segments forming the contour of the object model, Determine a pixel as a corresponding pixel in a direction-specific edge image corresponding to the detected direction;
Object detection device. The detection target is a detection target vehicle existing on a road plane reflected in the captured image,
The object detection unit detects whether or not the detection target vehicle exists in the captured image based on the edge image classified by direction and a vehicle model corresponding to the detection target vehicle.
The object detection apparatus according to claim 1 . The edge image generation unit by direction is based on a predetermined direction corresponding to each of a plurality of line segments forming an outline of the vehicle model in an orientation corresponding to the road plane, and an edge extracted from the silhouette image. Generating the direction-specific edge image;
The object detection apparatus according to claim 2 . The object detection unit calculates a degree of coincidence of the vehicle model with respect to an edge of the edge image for each direction for each of the plurality of line segments, and whether or not a total value of the degree of coincidence for the plurality of line segments exceeds a threshold value. To determine,
The object detection apparatus according to claim 2 . The object detection unit calculates a degree of coincidence of the edge image for each direction with respect to an outline of the vehicle model for each of the plurality of line segments, and searches for the vehicle model in which the total value of the degree of coincidence is maximized;
The object detection apparatus according to claim 2 . The object detection unit searches for the vehicle model in which the total value of the degree of coincidence is maximum while changing at least one of the position and size of the vehicle model.
The object detection apparatus according to claim 5 . The predetermined direction corresponding to each of the plurality of line segments is a vertical direction with respect to each of the plurality of line segments.
The object detection apparatus according to claim 1 . Generating a silhouette image based on a captured image obtained by imaging a three-dimensional space by an imaging device;
Generating a plurality of direction-specific edge images based on a plurality of different predetermined directions corresponding to the detection object and edges extracted from the silhouette image ;
Detecting whether or not the detection target exists in the captured image based on the plurality of edge images according to directions and an object model corresponding to the detection target;
Only including,
When a direction in which gradient directions of pixels constituting the edge match or are similar to each other is detected from a predetermined direction corresponding to each of a plurality of line segments forming the contour of the object model, the pixel corresponds to the detected direction. Including determining as a corresponding pixel in the edge image according to direction,
Object detection method.

Images