JP4879229B2 - Pedestrian detection device and program - Google Patents

Description

  The present invention relates to a pedestrian detection apparatus and program, and more particularly to a pedestrian detection apparatus and program for detecting a pedestrian from a captured image.

  2. Description of the Related Art In recent years, an increasing number of vehicles are equipped with an object detection device that performs image processing on an image around a vehicle imaged by an in-vehicle camera, detects an object such as a pedestrian, and presents a detection result to a driver.

As a method of detecting a pedestrian, for example, there is a method of detecting a pedestrian by matching a captured image with a pattern image indicating a pedestrian. In this case, a signboard having a shape similar to the pattern image indicating a pedestrian And signs may be erroneously detected as pedestrians. Therefore, it is determined whether the object in front of the vehicle is a stationary object or a non-stationary object using a radar, and does not exist on the course of the own vehicle based on an image obtained by imaging the front of the vehicle among objects determined to be a stationary object. Proposed an obstacle detection method that prevents stationary objects such as roadside signs and guardrails that do not exist on the vehicle's path from being detected as obstacles (For example, Patent Document 1).
JP 2004-171295 A

  However, the obstacle detection method disclosed in Patent Document 1 has a problem that a pedestrian that is stationary on the roadside is excluded as being out of the detection target, so that the pedestrian cannot be detected appropriately.

  The present invention has been proposed to solve the above-described problems, and an object of the present invention is to provide a pedestrian detection device and a program that can appropriately detect a pedestrian while preventing erroneous detection. .

  In order to achieve the above object, a pedestrian detection device according to the present invention includes an extraction unit that extracts an image of a predetermined area from an input image, and collates the extracted image extracted by the extraction unit with a pedestrian pattern image. , Candidate image determining means for determining whether or not the extracted image is a candidate image for detecting a pedestrian, and at least one direction of the extracted image determined to be a candidate image by the candidate image determining means Detecting means for performing frequency analysis of the distribution of the sum of pixel values for each pixel line and detecting an extracted image having a power spectrum density of a predetermined frequency equal to or higher than a predetermined value as an image showing a pedestrian.

  According to the pedestrian detection device of the present invention, the extraction unit extracts an image of a predetermined area from the input image, and the candidate image determination unit collates the extracted image extracted by the extraction unit with the pedestrian pattern image. Then, it is determined whether or not the extracted image is a candidate image for detecting a pedestrian. Here, an image that clearly does not match the pattern image of the pedestrian is not determined as a candidate image, but an image of an object such as a sign or a sign that resembles a pedestrian is determined as the candidate image. In some cases.

  Therefore, the detection means performs frequency analysis of the distribution of the sum of pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determination means, and the power spectrum density of a predetermined frequency is An extracted image having a predetermined value or more is detected as an image showing a pedestrian.

  Further, the pedestrian detection device of the present invention includes an extraction unit that extracts an image of a predetermined area from an input image, and the extracted image extracted by the extraction unit and a pedestrian pattern image are collated, Candidate image determination means for determining whether or not the image is a candidate image for detecting a pedestrian, and pixels for each pixel line in at least one direction of the extracted image determined as the candidate image by the candidate image determination means Detecting means for detecting an image indicating a pedestrian by performing frequency analysis of a distribution of sum of values and deleting an extracted image having a power spectral density of a predetermined frequency less than a predetermined value from the candidate image. ing.

  According to the pedestrian detection device of the present invention, the detection means performs frequency analysis of the distribution of the sum of the pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determination means. And detecting an image indicating a pedestrian by deleting an extracted image having a power spectral density of a predetermined frequency less than a predetermined value from the candidate image.

  As described above, when a pedestrian and a sign or signboard that is an artifact are similar in shape, if a pixel value such as a density value (or luminance value) is analyzed in detail, a frequency analysis result in the case of a pedestrian By using the fact that the power spectrum density of the predetermined frequency is higher than the predetermined value, it is possible to appropriately extract images indicating pedestrians from candidate images or to detect errors such as signs and signs. A pedestrian can be detected appropriately by deleting an image showing an object to be detected from the candidate image.

  In the pedestrian detection program of the present invention, the computer compares the extraction means for extracting an image of a predetermined region from the input image, the extracted image extracted by the extraction means and the pedestrian pattern image, and Candidate image determining means for determining whether or not the extracted image is a candidate image for detecting a pedestrian, and a pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determining means The frequency analysis of the distribution of the sum of pixel values for each pixel is performed to function as detection means for detecting an extracted image having a power spectrum density of a predetermined frequency equal to or higher than a predetermined value as an image showing a pedestrian.

  In the pedestrian detection program of the present invention, the computer compares the extraction means for extracting an image of a predetermined region from the input image, the extracted image extracted by the extraction means and the pedestrian pattern image, and Candidate image determining means for determining whether or not the extracted image is a candidate image for detecting a pedestrian, and a pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determining means In order to perform a frequency analysis of the distribution of the sum of pixel values for each pixel and delete an extracted image having a power spectrum density of a predetermined frequency less than a predetermined value from the candidate image, thereby functioning as a detection unit that detects an image indicating a pedestrian belongs to.

  As described above, according to the present invention, it is possible to prevent erroneous detection and to appropriately detect a pedestrian.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  As illustrated in FIG. 1, the pedestrian detection device 10 according to the present embodiment detects a pedestrian based on an imaging device 12 that captures a range including an identification target region and a captured image output from the imaging device 12. The computer 16 which performs the pedestrian detection process routine to perform, and the display apparatus 18 for displaying the process result in the computer 16 are provided.

  The imaging device 12 captures an area including an identification target region and generates an image signal (not shown), and A / D conversion that converts an analog image signal generated by the imaging unit into a digital signal And an image memory (not shown) for temporarily storing the A / D converted image signal.

  The computer 16 includes a CPU that controls the pedestrian detection apparatus 10 as a whole, a ROM as a storage medium that stores a program for a pedestrian detection processing routine, which will be described later, a RAM that temporarily stores data as a work area, and a bus that connects them. It is comprised including.

  When the computer 16 is described with function blocks divided for each function realizing means determined based on hardware and software, as shown in FIG. 1, a window for extracting a predetermined area from a captured image captured by the imaging device 12 A candidate image determination unit 26 that determines whether the image extracted by the image extraction unit 22, the image extracted by the window image extraction unit 22 is a candidate image for detecting a pedestrian, or a candidate image by the candidate image determination unit 26 Based on the window image that has passed through the specific object removal unit 28 based on the captured image captured by the imaging device 12 and the specific object removal unit 28 that removes the image indicating the specific object from among the window images determined to be present. It can be expressed by a configuration including a display control unit 30 that controls to display the detection result superimposed on the display device 18.

  When the window image extraction unit 22 detects a pedestrian from a captured image, a window having a predetermined size (referred to as a search window) from the input image is determined for each step, and a predetermined movement amount (referred to as a search pitch). ), And the pedestrian is detected from the cut image (referred to as a window image). The amount of movement is determined by a predetermined number of pixels. Further, the window image extraction unit 22 converts the extracted window image into an image having a preset number of pixels (for example, an image of 16 horizontal pixels × 32 vertical pixels).

  The candidate image determination unit 26 performs pattern matching image processing on the window image extracted by the window image extraction unit 22, and is a candidate image of a pedestrian whose window image is a detection target by matching with the pattern image. It is determined whether or not there is. The pattern image is stored in a built-in or external storage medium such as a hard disk drive (HDD) or a CD-ROM as storage means. The window image determined as the candidate image of the pedestrian by the candidate image determination unit 26 may include an image such as a sign or a signboard having a shape similar to that of the pedestrian.

  The specific object removing unit 28 calculates the sum of the pixel values for each horizontal line and the sum of the pixel values for each vertical line of the window image determined as the candidate image of the pedestrian by the candidate image determination unit 26 as a projection value, The obtained projection values are combined, frequency analysis is performed to calculate the power spectral density (or power spectrum), and if the power spectral density at a predetermined frequency is less than the predetermined value, the window is determined to be a specific object The specific object is removed from the candidate image by deleting the image.

  Next, a processing routine for detecting pedestrians according to the present embodiment will be described with reference to FIG.

  In step 100, an image picked up by the image pickup device 12 is input. Next, in step 102, a search window of 16 × 32 pixels, for example, is input to the input image in a predetermined area (for example, a left corner area). For example, a 16 × 32 pixel window image is extracted from the input image using the set search window.

  Next, in step 104, pattern matching between the extracted window image and the pattern image of the pedestrian is performed to calculate a matching degree, and in the next step 106, an image of the image indicating the pedestrian whose detection target is the window image. Determine whether or not to be a candidate. If the degree of matching calculated in step 104 is equal to or greater than a predetermined value, it is determined that the image is a pedestrian candidate image, and the process proceeds to step 108. If the matching degree is determined to be less than the predetermined value, the process proceeds to step 110.

  In step 108, a process for removing a specific object, which will be described later, is executed. Next, in step 110, it is determined whether the search has been completed by scanning the search window for the entire input image. If not completed, the process proceeds to step 112, the position of the search window is moved by a predetermined search pitch, the process returns to step 102, and the processes from step 102 to step 110 are repeated. When the search for the entire image is completed, the process proceeds to step 114.

  Although the search window is used as a frame for extracting a window image for detecting a pedestrian, people of various sizes can be detected if the sizes of the search windows are different. Therefore, search windows of various sizes may be prepared in advance, and the entire image may be searched in each search window.

  Next, in step 114, the display device 18 is set so that the detected pedestrian is displayed surrounded by a window with respect to the input image based on the information stored in the list in the specific object removal process described later. Control.

  Next, the principle of the specific object removal process will be described. Here, a case where the specific object to be removed is a label will be described as an example. Further, a density value represented by a numerical value of 0 (black) to 255 (white) is used as the pixel value of each pixel.

  As shown in FIG. 3, the sum of the density values of the pixels included in the first pixel line arranged in the horizontal direction (the direction of the arrow A) of the window image 32 is calculated as the first row projection value. Next, the projection value of the second row is calculated in the same manner, and the projection value of each row is calculated in the same manner up to the last row. From the calculated projection value, it is possible to obtain a graph 34 showing the distribution of density values with respect to a change in the vertical position of the window image.

  Similarly, the sum of the density values of the pixels included in the first pixel line arranged in the vertical direction (in the direction of arrow B) of the window image 32 is calculated as the first column projection value, and each column is obtained up to the last column. The projection value is calculated in the same manner. A graph 36 showing the distribution of density values with respect to changes in the position of the window image in the horizontal direction can be obtained from the calculated projection value, and a graph 34 showing the distribution of density values with respect to changes in the position of the window image in the vertical direction. A graph 38 showing the distribution of density values with respect to changes in the position of the window image is created. In the combination, a graph 36 indicating the distribution of density values with respect to a change in position in the horizontal direction may be combined with a graph 36 showing the distribution of density values with respect to a change in the position in the horizontal direction.

  As shown in FIG. 4, the graph 38 showing the distribution of the density value with respect to the change in the position of the window image created in this way is 5 to 5 in comparison with the graph in the case of a pedestrian. Characteristic data is an arc shape around 10 rows and an arc shape around 12-20 rows. In addition to signs, artificial objects such as billboards have similar characteristics. This feature is emphasized, for example, by performing a frequency analysis as shown in FIG. Since the graph 38 showing the distribution of the density value with respect to the change of the position of the window image, which is the original data for frequency analysis, is non-continuous data of the projection value for each position (row number and column number) of the window image. Here, frequency analysis is performed by discrete Fourier transform.

  For a predetermined number of pedestrian and sign sample images, power spectral density is calculated by discrete Fourier transform, and a histogram of power spectral density is obtained for each frequency. A histogram at a certain frequency is shown in FIG. The frequency of the power spectral density is distributed in the range of 0.6 to 0.9 around the power spectral density of 0.7 to 0.8 in the case of signs, but in the case of pedestrians, the frequency of the power spectral density It is distributed in the range of 0.8 to 0.9. Thus, since there is a difference in the distribution of power spectral density between the pedestrian and the sign, the pedestrian and the sign can be identified with high accuracy by setting an appropriate threshold value.

  The predetermined frequency for discriminating between the pedestrian and the sign is, for example, a frequency as shown in FIG. 6 for each frequency, and the frequency at which the discrimination error rate between the pedestrian and the sign is minimized, that is, the pedestrian's The frequency that maximizes the difference between the distribution of the power spectral density and the distribution of the power spectral density of the sign can be determined as the predetermined frequency.

  Further, the threshold value for distinguishing between a pedestrian and a sign is, for example, a power spectrum density (for example, FIG. 6) that is an intersection of the distribution of the power spectrum density of the pedestrian and the power spectrum density of the sign at the predetermined frequency. In this case, the power spectral density 0.83) can be determined as the threshold value. In addition, in order to reduce undetected pedestrians caused by accidentally deleting the pedestrian image that should be detected, a value moved in the left skirt direction of the pedestrian power spectrum density distribution (above) It may be a value smaller than the intersection point). Similarly, in order to reduce erroneous detection that the sign image to be originally deleted is a pedestrian image, the value moved in the right foot direction of the distribution of the power spectrum density of the sign (from the above intersection) It may be a large value.

  Next, the processing routine for removing the specific object will be described with reference to FIG.

  A graph showing a change in density value with respect to a change in position in a window image by obtaining a sum of pixel values for each pixel line in a horizontal direction and a vertical direction of a window image that is a candidate image of a pedestrian in step 200 as a projection value. create.

  Next, in step 202, the power spectrum density is calculated by performing discrete Fourier transform on the graph showing the change in the density value, and in the next step 204, it is determined whether or not the power spectrum density at the predetermined frequency is less than the predetermined value. . If it is less than the predetermined value, the window image indicates an object to be removed that is not a pedestrian, so the process proceeds to step 206 and the window image is deleted from the pedestrian candidate image. If the window image is not less than the predetermined value, the window image is an image showing a pedestrian, so the process proceeds to step 208, and information such as the position and size of the search window is stored in the RAM as a list.

  As described above, in the pedestrian detection apparatus 10 according to the present embodiment, the features that appear in the projection values of the pixel density values of the pedestrian image and the sign image are emphasized by frequency analysis. Since the specific object to be removed from the pedestrian can be appropriately identified, it is possible to prevent the pedestrian from being erroneously detected as a pedestrian and to detect the pedestrian appropriately. .

  In the present embodiment, the case where the specific object to be removed is a sign has been described. However, the distribution of the power spectrum density obtained by frequency analysis of the projection value of the pixel value is different from that of a pedestrian. By indicating the above, the pedestrian detection device of the present embodiment can be applied to an object that can be identified as a pedestrian.

  In the present embodiment, the case where the sum of the pixel values for each vertical line and the sum of the pixel values for each horizontal line are combined as the projection value of the pixel value has been described. However, at least one of the extracted images has been described. It is sufficient to use the sum of pixel values for each pixel line in the direction, use the sum of pixel values only in the vertical direction, use the sum of pixel values only in the horizontal direction, or use the sum of pixel values in any direction. Can do.

  In this embodiment, discrete Fourier transform is used for frequency analysis, but discrete cosine transform, discrete wavelet transform, or the like may be used.

  Further, in the present embodiment, the case where the window image is removed from the candidate image when the power spectral density is less than the predetermined value has been described. However, when the power spectral density is equal to or higher than the predetermined value, the image is extracted as an image indicating a pedestrian You may decide to do.

It is a block diagram which shows the structure of the pedestrian detection apparatus which concerns on this Embodiment. It is a flowchart which shows the processing routine of the pedestrian detection which concerns on this Embodiment. It is a figure for demonstrating calculation of a projection value. It is a figure which shows an example of the projection value of a pedestrian and a sign. It is a figure which shows an example of a frequency analysis. It is a histogram of the power spectral density in a predetermined frequency. It is a flowchart which shows the process routine of the specific target object removal which concerns on this Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Pedestrian detection apparatus 22 Window image extraction part 26 Candidate image determination part 28 Specific target object removal part 30 Display control part

Claims (4)

Extracting means for extracting an image of a predetermined area from the input image;
A candidate image determination unit that compares the extracted image extracted by the extraction unit with a pedestrian pattern image and determines whether the extracted image is a candidate image for detecting a pedestrian;
Perform frequency analysis of the distribution of the sum of pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determination means, and extract a power spectrum density at a predetermined frequency that is greater than or equal to a predetermined value Detecting means for detecting an image as an image showing a pedestrian;
A pedestrian detection device. Extracting means for extracting an image of a predetermined area from the input image;
A candidate image determination unit that compares the extracted image extracted by the extraction unit with a pedestrian pattern image and determines whether the extracted image is a candidate image for detecting a pedestrian;
Perform frequency analysis of the sum distribution of pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determining means, and extract a power spectrum density of a predetermined frequency less than a predetermined value Detecting means for detecting an image showing a pedestrian by deleting the image from the candidate image;
A pedestrian detection device. Computer
Extracting means for extracting an image of a predetermined area from the input image;
A candidate image determination unit that compares the extracted image extracted by the extraction unit with a pedestrian pattern image and determines whether the extracted image is a candidate image for detecting a pedestrian;
Perform frequency analysis of the distribution of the sum of pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determination means, and extract a power spectrum density at a predetermined frequency that is greater than or equal to a predetermined value Detecting means for detecting an image as an image showing a pedestrian;
Pedestrian detection program to make it function. Computer
Extracting means for extracting an image of a predetermined area from the input image;
A candidate image determination unit that compares the extracted image extracted by the extraction unit with a pedestrian pattern image and determines whether the extracted image is a candidate image for detecting a pedestrian;
Perform frequency analysis of the sum distribution of pixel values for each pixel line in at least one direction of the extracted image determined to be a candidate image by the candidate image determining means, and extract a power spectrum density of a predetermined frequency less than a predetermined value Detecting means for detecting an image showing a pedestrian by deleting the image from the candidate image;
Pedestrian detection program to make it function.