JP4842301B2 - 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 for detecting a pedestrian, for example, a method for detecting a pedestrian by matching a captured image with a pattern image indicating the pedestrian is known. In this method, the end (a) and the support (b) of the guardrail as shown in FIG. 12 may be erroneously detected as a pedestrian.

  Therefore, as an apparatus for detecting a guardrail, Patent Document 1 forms a binarized image including an edge pattern by binarizing density differential values of all the pixels when imaging a vehicle rear and recognizing a moving object. An object detection device for detecting a guardrail by determining whether or not it is a guardrail from an appearance cycle of this edge row, using an edge row aligned with a radial line from an enlarged focus corresponding to one point in the imaging direction as a guardrail candidate Has been proposed.

Further, in Patent Document 2, when recognizing a guard rail or the like as an object having a portion standing on a straight line from the road surface on which the vehicle travels, when images are taken in different time series using two cameras, An area including a linear element extending in a direction different from the temporal change direction caused by the running of the vehicle on the image plane captured by one camera is set as the first selection area, and the first selection area is the other camera. Proposed an object detection device that uses the area estimated as the movement destination at the imaging timing as the second selection area, and determines the object (such as a guardrail) from the presence or absence of correlation between the first selection area and the second selection area Has been.
JP 2003-162710 A JP 2004-013400 A

  However, the object detection devices of Patent Document 1 and Patent Document 2 are mainly intended to prevent erroneous detection of the guardrail as a vehicle traveling in an oncoming lane or an overtaking lane, and in time series processing. Since a plurality of input images are required, there is a problem that the processing becomes complicated.

  The present invention has been proposed to solve the above-described problem, and a pedestrian detection apparatus and program capable of improving the accuracy of detecting a pedestrian by preventing erroneous detection through simple and high-speed processing. The purpose is to provide.

  In order to achieve the above object, a pedestrian detection apparatus according to the present invention includes an extraction unit that extracts an image of a rectangular area from an input image, and collates the extracted image extracted by the extraction unit with a pedestrian pattern image. Candidate image determination means for determining whether or not the extracted image is a candidate image for detecting a pedestrian, vanishing point detection means for detecting a vanishing point of the input image, and candidate image determination means An edge determination unit that determines whether or not an edge exists in an end region close to the vanishing point detected by the vanishing point detection unit of the extracted image determined to be an image, and the edge determination unit Determining means for determining an extracted image determined to have no edge in the partial area as an image indicating a pedestrian.

  According to the pedestrian detection device of the present invention, the extraction unit extracts an image of a rectangular 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 the end portion of the guardrail or the column portion may be determined as a candidate image.

  Therefore, the vanishing point detecting unit detects the vanishing point of the input image, and the edge determining unit is close to the vanishing point detected by the vanishing point detecting unit of the extracted image determined to be a candidate image by the candidate image determining unit. It is determined whether or not an edge exists in the end region. Since the bar portion of the guardrail appears as a straight line toward the vanishing point in the input image, if an edge exists in the end region near the vanishing point of the extracted image, the object indicated by the extracted image is determined to be a guardrail. can do. Therefore, the determining unit determines the extracted image determined by the edge determining unit as having no edge in the end region as an image indicating a pedestrian.

  As described above, a candidate image not showing the characteristics of the guardrail is determined as an image showing the pedestrian from the candidate images of the pedestrian only by determining whether or not an edge exists in the end region of the extracted image. Therefore, erroneous detection can be prevented by simple and high-speed processing, and the accuracy of detecting a pedestrian can be improved.

  The edge determination means of the pedestrian detection device of the present invention may be configured such that when a pair of parallel edges with a predetermined interval extending in the direction of the vanishing point is extracted in the end region, the end region has an edge. It can be determined that it exists. Thus, when a pair of parallel edges with a predetermined interval extending in the direction of the vanishing point considered to represent the bar portion of the guard rail is extracted, it is determined that the edge exists in the end region, thereby Since the feature can be captured more accurately, it is possible to improve the accuracy of detecting a pedestrian by preventing erroneous detection.

  Further, the edge determination means of the pedestrian detection device of the present invention determines that an edge exists in the end region when the edge is extracted at two or more sites in the end region. be able to. Since the guard rail is often provided with two or more bar portions, when the edge determination means extracts edges at two or more parts in the end region, there is an edge in the end region. judge. In this way, since the characteristics of the guardrail can be grasped more accurately, the accuracy of detecting a pedestrian can be improved by preventing erroneous detection.

  In addition, the edge determination unit of the pedestrian detection device of the present invention further determines whether or not an edge exists in a region outside the end region of the extracted image, and the end region and the end portion When an edge is extracted in a region outside the region, it can be determined that an edge exists in the end region.

  The edge determination means determines whether or not there is an edge in the region outside the end region of the extracted image. Since the bar portion of the guardrail appears as a straight line toward the vanishing point, if there is an edge outside the end region of the extracted image, the edge existing in the end region is continuous in the direction of the vanishing point. Judgment can be made. In this way, since the characteristics of the guardrail can be grasped more accurately, the accuracy of detecting a pedestrian can be improved by preventing erroneous detection.

  The pedestrian detection apparatus of the present invention further includes a high brightness area determination unit that determines whether or not a high brightness area exists in the extracted image. The extracted image that has been determined not to be determined and determined to have no high brightness region by the high brightness region determination means can be determined as an image showing a pedestrian.

  In many cases, the guardrail is provided with a reflector, and the reflector is represented as an area occupied by a pixel having a high luminance value in the input image. Therefore, the high brightness area determination means determines whether or not there is a high brightness area in which a pixel having a brightness value greater than or equal to a predetermined value occupies a size greater than or equal to a predetermined range in the extracted image. By determining the extracted image determined by the means that the edge does not exist and the high-luminance area determining means determined that the high-luminance area does not exist as an image indicating a pedestrian, the erroneous detection is further prevented. The accuracy of detecting pedestrians can be improved.

  Moreover, the pedestrian detection program of this invention is for making a computer function as each means which comprises the said pedestrian detection apparatus.

  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 first embodiment includes an imaging device 12 that captures a range including a region to be identified, and a pedestrian based on a captured image output from the imaging device 12. The computer 16 which performs the pedestrian detection process routine which detects this, 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.

  If the computer 16 is described with functional blocks divided for each function realizing means determined based on hardware and software, as shown in FIG. 1, the disappearance for detecting the vanishing point of the captured image captured by the imaging device 12 is detected. The point detection unit 22, the window image extraction unit 24 that extracts a predetermined area from the captured image captured by the imaging device 12, and the image extracted by the window image extraction unit 24 are candidate images for detecting a pedestrian. A candidate image determination unit 26 that determines whether or not there is an edge determination unit 28 that determines whether or not an edge exists in the edge region of the window image that is determined to be a candidate image by the candidate image determination unit 26; A reflector determination unit 30 that determines whether or not an image showing a reflector exists in the window image from the luminance value of the pixel, an edge determination unit 28, and a reflector Based on the determination result of the Kuta determination unit 30, a pedestrian image determination unit 32 that determines a window image indicating a pedestrian from candidate images, and a determination by the pedestrian image determination unit 32 on a captured image captured by the imaging device 12 It can be expressed by a configuration including a display control unit 34 that controls to superimpose the results and display them on the display device 18.

  The vanishing point detection unit 22 detects a vanishing point that is a convergence point (intersection point) of parallel lines in the three-dimensional space from the input image. The vanishing point can be detected as an intersection obtained by estimating the lane and the road surface area from the input image based on the color and position and extending the straight line of the lane and road surface area. Note that the vanishing point may be detected by performing processing such as binarization of the input image and detecting linear components other than the vertical direction and the horizontal direction without detecting the lane or the road surface area.

  When the window image extraction unit 24 detects a pedestrian from the 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. In addition, the window image extraction unit 24 converts the extracted window image into an image having a preset number of pixels (for example, an image of 16 horizontal x 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 24, 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 a guardrail image as shown in FIG.

  The edge determination unit 28 uses the Haar feature to determine whether or not there are a plurality of edges in the end region near the vanishing point of the window image determined as the candidate image of the pedestrian by the candidate image determination unit 26. The determination is made based on the calculated luminance information. Since the image showing the guardrail has a feature that there are a plurality of edges in the end region near the vanishing point and the edges are continuous to the outside of the window image, the edge determination unit 28 has a plurality of edges. Whether or not there is an edge in the region outside the end region where the edge exists is determined by the same method.

  The reflector determination unit 30 determines whether or not there is a high luminance region in which pixels having a luminance value equal to or greater than a predetermined value in the window image occupy a size equal to or greater than a predetermined range. The guard rail is often provided with a reflector, and the reflector shows a high luminance value in the input image. Therefore, if a high luminance area exists in the window image, the high luminance area is a feature of the guard rail. It is determined that it is a reflector.

  The pedestrian image determination unit 32 determines that the window image determined by the edge determination unit 28 as having no edge and determined by the reflector determination unit 30 as having no reflector is an image indicating a pedestrian.

  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, the lane and the road surface area are estimated from the input image, and the straight lines of the estimated lane and road surface area are extended and extended. The intersection of the straight lines is detected as the vanishing point of the input image. For example, in the case of an input image as shown in FIG. 3, the intersection P of the extension line of the center line 36 and the white line 38 on the shoulder side is detected as a vanishing point.

  Next, in step 104, a search window of, for example, 16 × 32 pixels is set as a predetermined area (for example, a left corner area) of the input image with respect to the input image, and from the input image using the set search window 40. For example, a window image 44 of 16 × 32 pixels is extracted.

  Next, in step 106, pattern matching between the extracted window image 44 and the pedestrian pattern image is performed to calculate a matching degree, and in the next step 108, the window image 44 indicates a pedestrian that is a detection target. It is determined whether or not the image is a candidate. If the degree of matching calculated in step 106 is greater than or equal to a predetermined value, it is determined that the image is a pedestrian candidate image, and the process proceeds to step 110. If it is determined that the matching degree is less than the predetermined value, the process proceeds to step 112.

  In step 110, a pedestrian image determination process, which will be described later, is executed. Next, in step 112, it is determined whether or not the search has been completed by scanning the search window 40 for the entire input image. If not completed, the process proceeds to step 114, the position of the search window 40 is moved by a predetermined search pitch, the process returns to step 104, and the processes from step 104 to step 112 are repeated. When the search for the entire image is completed, the process proceeds to step 116.

  The search window 40 is used as a frame for extracting a window image for detecting a pedestrian. However, if the search window 40 has a different size, people of various sizes can be detected. Therefore, search windows 40 of various sizes may be prepared in advance, and the entire image may be searched for in each search window 40.

  In step 116, the information stored in the list in the pedestrian image determination process described later is supplied to the display control unit 34, and the display control unit 34 detects the pedestrian detected for the input image based on the information in the list. Is controlled so as to be displayed surrounded by a window.

  Next, a processing routine for determining a pedestrian image will be described with reference to FIG.

  In step 200, an end region 46 of the window image 44 to be determined whether or not there is an edge is determined based on the position of the vanishing point of the input image. When the window image 44 is extracted in the region on the right side from the vanishing point, the left end portion of the window image 44, and when the window image 44 is extracted in the region on the left side from the vanishing point, the right end of the window image 44 is extracted. Is determined as an end region 46 for determining the presence or absence of an edge. For example, in the case of the window image 44 extracted by the search window 40 shown in FIG. 3, since it is extracted in the region on the left side from the vanishing point, the right end portion of the window image is used as the end region 46 for determining the presence or absence of an edge. It is determined.

  Next, in step 202, luminance information is calculated from the end region 46 of the determined window image 44. Here, the Haar feature is used as the luminance information. The Haar feature is a feature amount using a luminance value of a pixel included in a simple rectangular pattern region, and high-speed processing is possible. Specifically, a rectangular area 42 composed of a white area (A area) and a black area (B area) as shown in FIG. 5 characterized by a difference in luminance value between the bar portion of the guardrail and its background is shown in FIG. As shown in the figure, it is set parallel to the edge area 46 on the top of the edge area 46 of the window image 44, and the sum of the luminance values of each pixel included in the A area and the luminance value of each pixel included in the B area The difference from the sum is calculated as luminance information Diff. Similarly, the luminance information Diff is calculated sequentially for each movement pitch to the bottom of the end region 46 while shifting the rectangular region 42 downward by a predetermined range (for example, one pixel line).

  Next, at step 204, it is determined whether or not a plurality of edges are detected in the end region 46 of the window image 44. With reference to the luminance information Diff calculated in step 202, if there is a portion where the luminance information Diff has changed significantly, it is determined that there is an edge.

  For example, the luminance information Diff (n) for each movement pitch is Diff (1) = 66, Diff (2) = 65, Diff (3) = 67,..., Diff (6) = 65, Diff (7) = 117, Diff (8) = 115, Diff (9) = 64,..., The difference between Diff (1) and Diff (2) is 1, Diff (2) and Diff (3) And the difference between Diff (7) and Diff (8) is 2, whereas the difference between Diff (6) and Diff (7) is 52, and Diff (8) and Diff (8) Since the difference from 9) is as large as 51, it is determined that an edge exists in this window image.

  If there are no more than two locations where the luminance information Diff has changed significantly, the process proceeds to step 218. If two or more edges are detected, the process proceeds to step 206.

  In addition, as shown in FIG. 7, the rectangular region 42 can capture the characteristics of the bar portion of the guardrail more by setting the rectangular region 42 at a predetermined angle. For the predetermined angle, an appropriate value may be determined in advance from the relationship between the average height of the guardrail and the installation position (line-of-sight height) of the imaging device 12.

  In step 206, the luminance information Diff is calculated for the region 48 outside the end region 46 of the window image 44 as in the process of step 202. Next, at step 208, it is determined with reference to the luminance information Diff calculated at step 206 whether an edge exists in the region 48 outside the end region 46. When there is a portion where the luminance information Diff is greatly changed in the outer region 48 corresponding to the portion where the luminance information Diff calculated in the end region 46 is greatly changed, the edge of the window image 44 is also displayed. This represents that the edge of the partial region 46 is continuous toward the outside where the vanishing point exists, and can be understood as a feature of the bar portion of the guardrail. If there is no edge in the outer region 48, the process proceeds to step 218. If there is an edge, go to step 210.

  In step 210, a small window 50 as shown in FIG. 8 is set as a predetermined area (for example, a left corner area) of the window image 44 with respect to the window image 44, and the brightness of each pixel in the set small window 50 is set. Extract the value. The size of the small window 50 is set to a size corresponding to the guardrail reflector based on the size of the window image 44, the extraction position of the window image in the input image, and the like.

  Next, in step 212, it is determined whether or not the small window 50 is a high luminance area. When pixels of a predetermined ratio or more (for example, 80% or more) with respect to the total number of pixels in the small window 50 are pixels indicating a luminance value of a predetermined value or more, it is determined that the pixel is a high luminance region. If it is not the high luminance region, the process proceeds to step 214, and it is determined whether or not the extraction of the luminance value is completed by scanning the small window 50 for the entire window image 44. If not completed, the process proceeds to step 216, the position of the small window 50 is moved by a movement pitch determined in advance by a predetermined number of pixels, and the process returns to step 210.

  If the result in step 204, step 208, and step 214 is affirmative, that is, for the window image that did not show the characteristics of the guardrail, in step 218, information such as the position and size of the search window is saved in the RAM as a list. To return.

  As described above, in the pedestrian detection apparatus according to the first embodiment, the determination of the presence / absence of an edge between the end region of the window image and the region outside the end region, and the determination of the presence / absence of a high-luminance region. An image showing a pedestrian with a window image that does not show the characteristics of the guardrail among the candidate images of the pedestrian by determining the presence or absence of a bar portion or a reflector portion toward the vanishing point, which is a feature of the guardrail, with simple processing Therefore, it is possible to improve the accuracy of detecting pedestrians by preventing false detection with simple and high-speed processing.

  In the first embodiment, the example of determining the presence / absence of an edge using the Haar feature has been described. However, the presence / absence of an edge is determined using edge extraction by differentiation or edge extraction by template matching. Also good.

  Next, a pedestrian detection apparatus according to the second embodiment will be described. In the first embodiment, it is determined whether or not an edge is present using the luminance information Diff of the end region 46 of the window image 44, whereas in the second embodiment, the bar portion of the guardrail is determined. Whether or not there is an edge is different depending on whether or not a straight edge indicating the above is extracted. Since the configuration is the same as that of the first embodiment, description thereof is omitted.

  With reference to FIG. 9, the pedestrian image determination process routine in 2nd Embodiment is demonstrated. In addition, this process is a process performed in step 110 of a pedestrian detection process routine (FIG. 2) similarly to 1st Embodiment. Moreover, about the process same as the pedestrian image determination process routine (FIG. 4) of 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  In step 200, the edge region 46 of the window image 44 is determined. Next, in step 300, the edge value 46 is calculated by calculating the luminance value difference of each pixel in the edge region 46 of the window image 44 determined in step 200. To extract.

  Next, in step 302, it is determined whether or not there are two or more pairs of substantially parallel straight edges with a predetermined interval extending in the direction of the vanishing point on the extracted edges. Whether it extends in the direction of the vanishing point can be determined by the slope of the straight line. The predetermined interval is set to a size that takes into account the thickness of the bar portion of the guardrail based on the size of the window image 44, the extraction position of the window image 44 in the input image, and the like. If two or more sets of straight edges are not extracted, the process proceeds to step 218. If extracted, the process proceeds to step 304.

  In step 304, edges are extracted from the region 48 outside the end region 46 of the window image 44 in the same manner as in step 300. Next, in step 306, it is determined whether or not a straight edge extending in the direction of the vanishing point exists in the edge extracted from the outer region 48 continuously to the straight edge extracted in the end region 46. If not, the process proceeds to step 218. If present, the process proceeds to step 210, and the presence or absence of the reflector is determined in steps 210 to 216.

  As described above, in the pedestrian detection device according to the second embodiment, the presence / absence of an edge in the end region of the window image is determined at a predetermined interval extending in the direction of the vanishing point indicating the feature of the bar portion of the guardrail. Since it is performed based on whether or not a pair of substantially parallel straight edges has been extracted, the characteristics of the guardrail can be captured more accurately, and erroneous determination of the guardrail as a pedestrian can be prevented more accurately.

  In the second embodiment, the edge is extracted based on the difference in luminance value of each pixel in the end region. However, the edge is extracted by the Haar feature used in the first embodiment. May be. In order to extract an edge using the Haar feature, for example, the rectangular area is a rectangular pattern composed of a white area (C area) and a black area (D area) as shown in FIG. The size is smaller than the rectangular pattern (FIG. 5) used in the embodiment. Then, as in the first embodiment, the luminance information Diff is calculated while shifting the rectangular area 42 downward by a predetermined range (for example, one pixel line), and the value of the luminance information Diff increases. The boundary between the C region and the D region can be extracted as an edge.

  In the second embodiment, a case has been described in which it is determined whether or not there is a pair of substantially parallel straight edges with a predetermined interval extending in the direction of the vanishing point. However, a pair of substantially parallel edges with a predetermined interval has been described. As long as it is not a straight line, it may be curved. Further, the term “substantially parallel” includes a case where the predetermined interval is gradually narrowed toward the end portion close to the vanishing point of the window image, that is, corresponding to the depth of the image.

  Further, in the first and second embodiments, the case where the window image is extracted while shifting the search window with respect to the entire area of the input image has been described. For example, as a configuration as shown in FIG. An area such as a road surface may be estimated using vanishing points detected at the exit, and a window image may be extracted based on the estimated area. Thereby, for example, since there are no pedestrians in an empty area or the like, it is possible to speed up the pedestrian detection process by extracting window images along a road surface area where pedestrians may exist. it can.

  In the first and second embodiments, the case where an image that does not show the characteristics of the guardrail is determined from the candidate images as an image that indicates a pedestrian, but the determined image is subjected to more detailed pattern matching or the like. Thus, more accurate pedestrian detection may be performed.

It is a block diagram which shows the functional structure of the pedestrian detection apparatus which concerns on 1st Embodiment. It is a flowchart which shows the processing routine of the pedestrian detection which concerns on 1st Embodiment. It is a figure for demonstrating vanishing point detection. It is a flowchart which shows the processing routine of the pedestrian image determination which concerns on 1st Embodiment. It is a figure which shows an example of the rectangular area of a Haar feature. It is a figure for demonstrating the presence or absence determination of an edge. It is another figure for demonstrating the presence or absence of an edge. It is a figure for demonstrating the presence or absence determination of a reflector. It is a flowchart which shows the processing routine of the pedestrian image determination which concerns on 2nd Embodiment. It is a figure which shows another example of the rectangular area | region of a Haar feature. It is a block diagram which shows the other example of a functional structure of the pedestrian detection apparatus which concerns on 1st and 2nd embodiment. It is an example of the window image of a guardrail.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Pedestrian detection apparatus 22 Vanishing point detection part 24 Window image extraction part 26 Candidate image determination part 28 Edge determination part 30 Reflector determination part 32 Pedestrian image determination part 34 Display control part

Claims (7)

Extracting means for extracting an image of a rectangular 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;
Vanishing point detecting means for detecting the vanishing point of the input image;
Edge determination means for determining whether or not an edge exists in an end region close to the vanishing point detected by the vanishing point detection means of the extracted image determined to be a candidate image by the candidate image determination means;
Determining means for determining, as an image showing a pedestrian, an extracted image determined by the edge determining means to have no edge in the end region;
A pedestrian detection device.   The edge determination means determines that an edge exists in the end region when a pair of substantially parallel edges with a predetermined interval extending in the direction of the vanishing point are extracted in the end region. Pedestrian detection device.   3. The pedestrian detection device according to claim 1, wherein the edge determination unit determines that an edge is present in the end region when the edge is extracted at two or more sites in the end region. .   The edge determination means further determines whether or not an edge exists in a region outside the end region of the extracted image, and an edge is extracted in the end region and a region outside the end region. The pedestrian detection device according to any one of claims 1 to 3, wherein it is determined that an edge exists in the end region. A high-luminance area determination means for determining whether or not a high-luminance area exists in the extracted image;
The determining means determines the extracted image that is determined by the edge determining means that no edge is present and that has been determined by the high brightness area determining means that the high brightness area is not present as an image indicating a pedestrian. The pedestrian detection device according to any one of claims 1 to 4. Computer
Extracting means for extracting an image of a rectangular 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;
Vanishing point detecting means for detecting the vanishing point of the input image;
Edge determination means for determining whether an edge exists in an end region close to the vanishing point detected by the vanishing point detection means of the extracted image determined to be a candidate image by the candidate image determination means;
Determining means for determining, as an image showing a pedestrian, an extracted image determined by the edge determining means to have no edge in the end region;
Pedestrian detection program to make it function.   The pedestrian detection program for functioning a computer as each means which comprises the pedestrian detection apparatus of any one of Claims 1-5.