JP2010160770A - Pedestrian detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a pedestrian detection device that suppresses the occurrence of erroneous detection for detecting a telegraph pole with a reflecting plate as a pedestrian. <P>SOLUTION: An image processing part 51 determines whether a percentage of a specific direction among edge directions included in a pedestrian candidate area constituting a second pedestrian candidate area extraction image exceeds a predetermined percentage, and when it is determined that the predetermined percentage is exceeded, excludes the pedestrian candidate area from the second pedestrian candidate area extraction image. In addition, the image processing part 51 determines whether labeling density of the pedestrian candidate area constituting the second pedestrian candidate area extraction image is equal to or more than predetermined density, and when it is determined that the predetermined density is exceeded, excludes the pedestrian candidate area from the second pedestrian candidate area extraction image. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a pedestrian detection device that detects a pedestrian existing around a vehicle.

  Conventionally, for example, a pedestrian detection device described in Patent Document 1 is known. In the technique described in this document, reflected light from an object with respect to near-infrared light irradiated to the periphery (front) from a projector is captured by an infrared camera, the infrared image is image-processed by a periphery monitoring control unit, and a threshold value is set. Extract the high brightness area that exceeds. The extracted high luminance area whose aspect ratio is within a predetermined range is extracted as a pedestrian candidate area, and the luminance average value and luminance variance value of the image of the pedestrian candidate area are calculated, and the structure is based thereon The candidate pedestrian area determined as an object is excluded, and the rest is set as a pedestrian area.

JP 2005-159392 A

  By the way, in the above prior art, it is assumed that a pedestrian candidate area having a high luminance average value and a low luminance variance value is not a pedestrian candidate area in which a pedestrian is imaged, but this assumption may not be satisfied. is there.

  Specifically, a power pole having a reflector (electric pole sign guard plate) is installed on the road, and the reflector of the utility pole is often subjected to a surface treatment that easily reflects infrared rays. For this reason, in the infrared image captured by the infrared camera, the average luminance value of the pedestrian candidate area where the utility pole including the reflector is captured is high. Further, the reflector of the utility pole is often provided with a striped pattern. Especially when the distance from the utility pole is short, the infrared camera can capture even the striped pattern. Since it is possible to capture a striped pattern, in the infrared image captured by the infrared camera, the luminance dispersion value of the pedestrian candidate region where the utility pole including the reflector is captured is high. Therefore, it becomes impossible to exclude a pedestrian candidate area in which a utility pole including a reflector is imaged. If it cannot be excluded, it may happen that a utility pole with a reflector is erroneously detected as a pedestrian.

  On the other hand, for example, when the pedestrian is dressed to strongly reflect infrared rays, or when the distance between the pedestrian and the infrared camera is short, the average luminance value of the pedestrian candidate area where the pedestrian is imaged is high. In addition, the luminance dispersion value may be low. In such a case, the pedestrian candidate area where the pedestrian is imaged is excluded. If excluded, a pedestrian may not be detected.

  This invention is made | formed in view of the said situation, The 1st objective provides the pedestrian detection apparatus which can aim at generation | occurrence | production suppression of the false detection which detects the utility pole provided with a reflecting plate as a pedestrian. There is to do.

  This invention is made | formed in view of the said situation, The 2nd objective is to provide the pedestrian detection apparatus which can aim at the undetected generation | occurrence | production suppression which does not detect a pedestrian.

  In order to achieve the first object, according to the first aspect of the present invention, an image obtained by extracting a high-brightness region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle. An image obtained by extracting a pedestrian candidate region from the high-intensity region extraction image based on an aspect ratio of a circumscribed rectangle circumscribing the high-intensity region, and high-intensity region extraction image generation means for generating a certain high-intensity region extraction image Based on the pedestrian candidate area extraction image generation means for generating a certain pedestrian candidate area extraction image and the edge direction included in the pedestrian candidate area, the pedestrian candidate area is determined to be a pedestrian candidate area imaged. A pedestrian candidate area is excluded from the pedestrian candidate area extraction image, and a pedestrian candidate area in which the utility pole reflector is imaged by the utility pole reflector exclusion means is excluded. The on the basis of the pedestrian candidate region extraction image after, characterized in that it comprises a pedestrian detection means for detecting a pedestrian existing around the vehicle.

  As described above, the reflector of the utility pole is often provided with a striped pattern, and this striped pattern can be detected as an edge in an image captured at a short distance. is there. Therefore, based on the edge direction included in the pedestrian candidate area, it is possible to determine whether the pedestrian candidate area is a pedestrian candidate area in which the reflector of the utility pole is imaged. In the above configuration as the pedestrian detection device, it is determined whether or not the pedestrian candidate area is a pedestrian candidate area obtained by imaging the utility pole reflector based on the edge direction included in the pedestrian candidate area. If it is determined that the candidate area is a pedestrian candidate area in which the utility pole reflector is imaged, the pedestrian candidate area is excluded from the pedestrian candidate area extraction image. And the pedestrian who exists in the periphery of a vehicle is detected based on the pedestrian candidate area | region extraction image after the pedestrian candidate area | region where the utility pole reflector was imaged was excluded. In this way, since a pedestrian existing around the vehicle is detected based on the pedestrian candidate area extraction image after the pedestrian candidate area in which the utility pole reflector is imaged is excluded, the electric pole provided with the reflector is walked. Occurrence of erroneous detection detected as a person can be achieved.

  In order to achieve the first object, in the invention described in claim 2, a high-luminance region having a luminance equal to or higher than a predetermined luminance is extracted from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle. A high-brightness region extraction image generation unit that generates a high-brightness region extraction image, and a pedestrian candidate region is extracted from the high-brightness region extraction image based on an aspect ratio of a circumscribed rectangle circumscribing the high-brightness region A pedestrian candidate area extraction image generating means for generating a pedestrian candidate area extraction image that is an image that has been processed, and a labeling density that is a ratio of the area of the pedestrian candidate area to the area of a circumscribed rectangle of the pedestrian candidate area A pedestrian candidate area determined to be a pedestrian candidate area in which the utility pole reflector is imaged, and a utility pole reflector exclusion means for excluding the pedestrian candidate area extraction image; Pedestrian detection means for detecting pedestrians present in the vicinity of the vehicle based on the pedestrian candidate area extraction image after the pedestrian candidate area in which the utility pole reflector is imaged by the board exclusion means is excluded. It is characterized by providing.

  In general, since the circumscribed rectangle area of the pedestrian candidate area is the same as or larger than the area of the pedestrian candidate area, the labeling density that is the ratio of the area of the pedestrian candidate area to the circumscribed rectangle area of the pedestrian candidate area is “ 1 "or less. However, when the pedestrian is a pedestrian candidate area where the pedestrian is imaged, the pedestrian has a lot of unevenness, and thus the labeling density is a value smaller than “1”. On the other hand, in the case of a pedestrian candidate area in which the reflector of the utility pole is imaged, the reflector of the utility pole has almost no unevenness, and thus its labeling density is a value close to “1”. Since the labeling density of the pedestrian candidate area has such a tendency, it is possible to determine whether or not the pedestrian candidate area is an imaged image of the reflector of the utility pole based on the labeling density of the pedestrian candidate area. is there. In the above configuration as the pedestrian detection device, based on the labeling density of such a pedestrian candidate area, it is determined whether or not it is a pedestrian candidate area in which the utility pole reflector is imaged, and the walking in which the utility pole reflector is imaged. If it is determined that the area is a pedestrian candidate area, the pedestrian candidate area is excluded from the pedestrian candidate area extracted image. And the pedestrian who exists in the periphery of a vehicle is detected based on the pedestrian candidate area | region extraction image after the pedestrian candidate area | region where the utility pole reflector was imaged was excluded. Thus, since a pedestrian existing in the vicinity of the vehicle is detected based on the pedestrian candidate area extraction image after the pedestrian candidate area in which the utility pole reflector is imaged is excluded, the utility pole provided with the reflector It is possible to suppress the occurrence of false detection of detecting a pedestrian as a pedestrian.

  In order to achieve the first object, in the invention according to claim 3, a high-luminance region having a luminance equal to or higher than a predetermined luminance is extracted from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle. A high-brightness region extraction image generation unit that generates a high-brightness region extraction image, and a pedestrian candidate region is extracted from the high-brightness region extraction image based on an aspect ratio of a circumscribed rectangle circumscribing the high-brightness region A pedestrian candidate area extraction image generating means for generating a pedestrian candidate area extraction image that is a pedestrian candidate area, and a pedestrian candidate area in which a utility pole reflector is imaged based on an edge direction included in the pedestrian candidate area The utility pole reflector exclusion means for excluding the determined pedestrian candidate area from the pedestrian candidate area extraction image, and the area of the pedestrian candidate area relative to the circumscribed rectangle area of the pedestrian candidate area Based on the labeling density that is a ratio, a pedestrian candidate area that is determined to be a pedestrian candidate area in which a utility pole reflector has been imaged, is excluded from the pedestrian candidate area extraction image; Pedestrian detection means for detecting pedestrians present in the vicinity of the vehicle based on the pedestrian candidate area extraction image after the pedestrian candidate area from which the utility pole reflector is imaged is excluded by the utility pole reflector exclusion means. It is characterized by providing.

  According to such a configuration as the pedestrian detection device, the configuration according to claim 1 and the configuration according to claim 2 are used in combination, so that a utility pole having a reflector is detected as a pedestrian. It becomes possible to further suppress the occurrence of erroneous detection.

  By the way, various methods can be considered as a determination method for determining whether or not the reflector of the utility pole is an imaged pedestrian candidate area based on the edge direction included in the pedestrian candidate area.

  In the configuration according to claim 1 or 3, the utility pole reflector exclusion means determines whether or not the ratio of the most frequent edge direction among the edge directions included in the pedestrian candidate area is equal to or greater than a predetermined ratio. And based on determining that it is more than a predetermined ratio, it is good also as determining that the pedestrian candidate area | region is a pedestrian candidate area | region which imaged the utility pole reflector. Alternatively, in the configuration according to claim 1 or 3, as in the invention according to claim 4, the utility pole reflector excluding means is arranged in a specific edge direction among edge directions included in the pedestrian candidate region. As to determine whether or not the ratio is equal to or greater than a predetermined ratio, and based on determining that the ratio is equal to or greater than the predetermined ratio, it is determined that the pedestrian candidate area is a pedestrian candidate area obtained by imaging the utility pole reflector. Also good. In such a configuration, as in the invention described in claim 5, it is desirable that the specific edge direction is a direction along a stripe formed on the utility pole reflector.

  Various methods are also conceivable as a determination method for determining whether or not the reflector of the utility pole is a pedestrian candidate region imaged based on the labeling density of the pedestrian candidate region.

  In the configuration according to claim 2 or 3, the utility pole reflector exclusion means determines whether or not the labeling density is "1", and based on determining that it is "1", It is good also as determining with a pedestrian candidate area | region being a pedestrian candidate area | region which imaged the utility pole reflector. Alternatively, in the pedestrian detection device according to claim 2 or 3, as in the invention according to claim 6, the utility pole reflector exclusion means determines whether the labeling density is equal to or higher than a predetermined density. And based on determining that it is more than a predetermined density, it is good also as determining that the pedestrian candidate area | region is a pedestrian candidate area | region which imaged the utility pole reflector. In such a configuration, it is desirable that the predetermined density is set to be higher than the labeling density of a pedestrian located in the vicinity of the vehicle. In any method, based on the labeling density of the pedestrian candidate area, it can be determined whether or not the pedestrian candidate area is a pedestrian candidate area obtained by imaging the reflector of the utility pole.

  Moreover, in the structure in any one of the said Claims 1-7, like the invention of Claim 8, the brightness | luminance average value in the inscribed rectangle inscribed in the outer shape of the said pedestrian candidate area | region, and this outer shape And a pedestrian candidate area determined to be a pedestrian candidate area obtained by imaging a road sign installed on a road based on a luminance dispersion value in an inscribed rectangle inscribed in the outer shape. It is desirable to further include road sign excluding means for excluding from the region extraction image.

  In such a configuration as a pedestrian detection device, not only the luminance average value in the inscribed rectangle inscribed in the outer shape of the pedestrian candidate area and the luminance dispersion value in the inscribed rectangle inscribed in the outer shape, but also walking In order to determine whether or not it is a pedestrian candidate area obtained by imaging a road sign installed on a road based on the outer shape of the pedestrian candidate area, a pedestrian candidate obtained by imaging a pedestrian from a pedestrian candidate area extracted image Excluding a region is less likely. As a result, the occurrence of undetected pedestrians can be reduced.

  The configuration described in claim 8 does not need to be used in combination with the configuration described in any of claims 1-7. That is, in order to achieve the second object, as in the invention according to claim 9, a high-luminance region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle. Based on the aspect ratio of the circumscribed rectangle circumscribing the high-luminance region, a pedestrian candidate region is extracted from the high-luminance region-extracted image based on the high-luminance region-extracted image generating means that generates the high-luminance region extracted image that is the extracted image Pedestrian candidate area extraction image generating means for generating a pedestrian candidate area extraction image that is an extracted image, an outer shape of the pedestrian candidate area, an average luminance value in an inscribed rectangle inscribed in the outer shape, and The pedestrian candidate area extracted image is a pedestrian candidate area that is determined to be a pedestrian candidate area obtained by imaging a road sign installed on a road based on a luminance dispersion value in an inscribed rectangle inscribed in the outer shape. Excluding road signs excluding means from on the basis of the pedestrian candidate region extraction image, may be provided with a pedestrian detection means for detecting a pedestrian existing around the vehicle.

It is a block diagram which shows an example of the whole structure about the vehicle periphery monitoring system containing one Embodiment of the pedestrian detection apparatus which concerns on this invention. It is a flowchart which shows the process sequence about the pedestrian detection process performed by the pedestrian detection apparatus. (A) is a figure which shows an example of the infrared image imaged with an infrared camera. (B) is a figure which shows an example of the binary image from which the edge contained in the infrared image shown to Fig.3 (a) was extracted. (C) is a figure which shows that the figure similar to a road sign is contained in the binary image shown in FIG.3 (b). (D) is a figure which shows the inscribed rectangle inscribed in the figure similar to the road sign contained in the infrared image shown to Fig.3 (a). It is a flowchart which shows the process sequence about the road sign exclusion process among pedestrian detection processes. (A) is a figure which shows the infrared image of the utility pole located in a short distance from an infrared camera. (B) is a figure which shows an example of the binary image from which the edge contained in the infrared image shown to Fig.5 (a) was extracted. It is a figure which shows the relationship between the edge direction contained in the binary image shown in FIG.5 (b), and its frequency. (A) is a figure which shows the infrared image of the utility pole located in a long distance from an infrared camera. (B) is a figure which shows the pedestrian candidate area | region and its circumscribed rectangle which are contained in the infrared image shown to Fig.7 (a). It is a flowchart which shows the process sequence about the utility pole reflector exclusion process in a pedestrian detection process.

  Hereinafter, an embodiment of a pedestrian detection device according to the present invention will be described with reference to FIGS. FIG. 1 is a block diagram illustrating an example of the overall configuration of the vehicle periphery monitoring system 1 including the pedestrian detection device 30. First, the configuration and function of the vehicle periphery monitoring system 1 including the pedestrian detection device 30 will be described with reference to FIG.

  As shown in FIG. 1, the vehicle periphery monitoring system 1 includes an infrared camera 10, a projector 20, and a pedestrian detection device 30.

  Of these, the infrared camera 10 is installed at the center of the front of the vehicle in the vehicle width direction so that its optical axis faces the front of the vehicle (not shown), and is connected to the pedestrian detection device 30 (specifically, the image processing unit 51). Has been. The infrared camera 10 captures and captures infrared light emitted from an object located in front of the vehicle or reflected from the object, and the captured infrared image is detected by the pedestrian detection device 30 (specifically, the image processing unit 51). Output to. The infrared camera 10 corresponds to the in-vehicle image pickup device described in the claims.

  One projector 20 is provided at each of a left end portion and a right end portion of a bumper (not shown) disposed in the front portion of the vehicle, and is connected to a pedestrian detection device 30 (specifically, a CPU 41) via a switch relay 21. ing. For convenience of illustration, only one projector 20 is shown in FIG. When the switch relay 21 is turned on by the pedestrian detection device 30, the projector 20 is also turned on, and the front of the vehicle is irradiated with near infrared rays.

  Incidentally, the higher the near-infrared reflectance of the object located in front of the vehicle, the more near-infrared rays irradiated from the projector 20 are reflected. Therefore, in the infrared image imaged with the infrared camera 10, the brightness | luminance of such a target object becomes high. Conversely, the lower the reflectance of the near-infrared ray of the object located in front of the vehicle, the less the near-infrared ray irradiated from the projector 20 is reflected. Therefore, in the infrared image imaged with the infrared camera 10, the brightness | luminance of such a target object becomes low.

  The pedestrian detection device 30 includes a CPU 41, a ROM 42, and a RAM 43. Of these, the ROM 42 stores in a non-volatile manner a program for executing various processes related to the pedestrian detection process described later and data such as threshold values required for executing the various processes, and the RAM 43 stores the calculation results. Etc. are temporarily stored. The CPU 41 is connected to the ROM 42 and the RAM 43, and executes various processes using the RAM 43 in accordance with programs stored in the ROM 42.

  Further, the pedestrian detection device 30 includes an image processing unit 51, and the image processing unit 51 includes an A / D conversion circuit, an image processing processor, a VRAM 52, a D / A conversion circuit, and a display screen ( FIG. 1 shows only the VRAM 52). The A / D conversion circuit captures the analog signal output from the infrared camera 10, converts the captured analog signal into a digital signal, and outputs the digital signal to the image processor. The image processor captures a digital signal (digital image data) output from the A / D conversion circuit, temporarily stores it in the VRAM 52, performs image processing on the stored digital image data, and then performs D / A conversion. Output to the circuit. The D / A conversion circuit captures the digital signal output from the image processor, converts the captured digital signal into an analog signal, and outputs the analog signal to the display screen.

  Incidentally, the digital image data is configured by integrating luminance data of each pixel of the infrared camera 10. Here, the luminance data of each pixel represents the luminance of the pixel by, for example, “8 bits (256 gradations)”, and the luminance data when the luminance is the highest (bright) is “255”. The luminance data of the pixel when the luminance is the lowest (dark) is set to “0”.

  The operation of the vehicle periphery monitoring system 1 configured as described above will be described with reference to FIGS. FIG. 2 is a flowchart showing a processing procedure of pedestrian detection processing executed by the pedestrian detection device 30 (specifically, the CPU 41 and the image processing unit 51).

  As shown in FIG. 2, when the pedestrian detection process S1 is started, the pedestrian detection apparatus 30 first captures an infrared image from the infrared camera 10 as a process of step S10. Specifically, the pedestrian detection device 30 turns on the switch relay 21 and turns on the projector 20. When turned on, the projector 20 irradiates the front of the vehicle with near infrared rays. The pedestrian detection device 30 images the front of the vehicle with the infrared camera 10 while the front of the vehicle is irradiated with near infrared rays by the projector 20. And the pedestrian detection apparatus 30 takes in the infrared image which imaged the front of the vehicle from the infrared camera 10. FIG.

  If the infrared image which imaged the front of a vehicle is taken in, the pedestrian detection apparatus 30 will extract the high-intensity area | region more than predetermined brightness from this infrared image as a process of subsequent step S20. Specifically, the image processing unit 51 extracts pixels whose luminance does not reach a predetermined luminance among pixels constituting the infrared image captured from the infrared camera 10, that is, pixels whose luminance data does not reach “150”, for example. The luminance data is replaced with “0”. On the other hand, the image processing unit 51 extracts a pixel whose luminance is equal to or higher than a predetermined luminance among pixels constituting the infrared image, that is, a pixel whose luminance data is “150” or higher, and holds the luminance data. In this way, the image processing unit 51 converts the infrared image into a high-brightness region extracted image that is an image obtained by extracting only a high-brightness region that is a region having a luminance of a predetermined luminance or higher from the infrared image.

  In the present embodiment, “150” is adopted as the luminance data corresponding to the luminance capable of extracting an object having a certain degree of reflection characteristics from the characteristics of the infrared image captured at night, but is not limited to this value. Moreover, the process of step S20 is equivalent to the high-intensity area | region extraction image generation means as described in a claim.

  When the infrared image is converted into the high-brightness region extraction image, the image processing unit 51 divides a high-brightness region of a predetermined luminance or higher included in the high-brightness region extraction image for each target as the processing of the subsequent step S30. A labeling process, which is a process of attaching a label for the purpose, is executed.

  When the labeling process is completed, the image processing unit 51 executes a pedestrian candidate area extraction process as the process of subsequent step S40. Specifically, the image processing unit 51 first sets a circumscribed rectangle that circumscribes a high luminance area having a predetermined luminance or higher included in the high luminance area extraction image. When the circumscribed rectangle is set, the image processing unit 51 next calculates the aspect ratio of the set circumscribed rectangle and determines whether or not the aspect ratio falls within a predetermined range. In the present embodiment, “vertical length of circumscribed rectangle: horizontal length of circumscribed rectangle” = “4: 1” to “4: 3” is adopted as the predetermined range of the aspect ratio. Then, the image processing unit 51 replaces the luminance data of the pixels located in the high luminance area determined to have an aspect ratio that does not fall within the predetermined range with “0”, while the aspect ratio falls within the predetermined range. The luminance data of the pixel located in the determined high luminance area is held as it is. In this way, the image processing unit 51 extracts only the high luminance area from which the aspect ratio of the circumscribed rectangle falls within the predetermined range from the high luminance area extracted image as the pedestrian candidate area. It converts into the 1st pedestrian candidate area extraction image which is an image. In the present embodiment, “vertical length of circumscribed rectangle: horizontal length of circumscribed rectangle” = “4: 1” to “4: 3” is adopted as the predetermined range. Not limited to. The first pedestrian candidate area extracted image corresponds to the pedestrian candidate area extracted image in the claims, and step S40 corresponds to the pedestrian candidate area extracted image generating means described in the claims.

  When such a pedestrian candidate area extraction process is executed, the image processing unit 51 may have captured a road sign from a pedestrian candidate area included in the first pedestrian candidate area extraction image as the subsequent step S50. A road sign exclusion process for excluding a pedestrian candidate area having a high height is executed.

  Hereinafter, the outline of the road sign exclusion process will be described with reference to FIGS. FIG. 3A is a part of an infrared image captured by the infrared camera 10 and is captured by overlapping two circular road signs. Since the road sign is often subjected to a surface treatment that easily reflects infrared rays, as shown in FIG. 3A, an infrared image obtained by imaging the road sign with the infrared camera 10 The average luminance value of the pedestrian candidate area where the sign is imaged is high, and the luminance dispersion value is low. That is, the pedestrian candidate area where the road sign is imaged becomes bright and uniform. Therefore, in the prior art, it has been assumed that a pedestrian candidate area with a high luminance average value and a low luminance dispersion value is not a pedestrian candidate area in which a pedestrian is imaged.

  However, as already described in the column of the problem, the above assumption may not be satisfied. For example, when the pedestrian is dressed to strongly reflect infrared rays, or when the distance between the pedestrian and the infrared camera 10 is a short distance, the average luminance value of the pedestrian candidate area in which the pedestrian is imaged is high, In addition, the luminance dispersion value may be low. In such a case, the pedestrian candidate area in which the pedestrian is imaged may be excluded from the pedestrian candidate area included in the first pedestrian candidate area extraction image.

  Therefore, in the present embodiment, the image processing unit 51 determines whether or not the luminance average value is high and the luminance variance value is low with respect to the pedestrian candidate region included in the first pedestrian candidate region extraction image. Before judging, it was decided whether or not a figure similar to a road sign was detected in the outer shape of the pedestrian candidate area. When a figure similar to a road sign is detected in the outer shape of the pedestrian candidate area, the pedestrian candidate area is highly likely to be a pedestrian candidate area obtained by imaging the road sign. For this reason, it is determined whether or not the luminance average value is high and the luminance dispersion value is low for such a highly likely pedestrian candidate region.

  Specifically, the image processing unit 51 performs the processes of steps S10 to S40 on the infrared image shown in FIG. 3A, and executes the process of step S51 described later. Then, the image processing unit 51 acquires a binary image from which edges included in the infrared image are extracted, as shown in FIG.

  The image processing unit 51 performs a determination process in step S52 described later on the binary image. That is, the image processing unit 51 determines whether or not a graphic similar to a road sign can be detected in the binary image. In the binary image example shown in FIG. 3B, a circle C that is a figure similar to a road sign can be detected. Therefore, the image processing unit 51 can determine that there is a high possibility that the candidate pedestrian area to be processed is a candidate pedestrian area obtained by imaging a road sign.

  Normally, in the binary image from which the edge is extracted, the luminance data of the pixel located in the portion corresponding to the edge is “255 (bright)”, and the luminance data of the pixel located in the portion not corresponding to the edge Is “0 (dark)”, but in FIGS. 3B and 3C, it is shown inverted for convenience of illustration.

  If it is determined that there is a high possibility that the candidate pedestrian area to be processed is a pedestrian candidate area obtained by imaging a road sign, the image processing unit 51 performs steps S53 to S55 described later with respect to FIG. Execute the process. That is, as shown in FIG. 3D, the image processing unit 51 sets an inscribed rectangle R inscribed in the circle C, and based on the luminance average value and the luminance dispersion value in the inscribed rectangle R, It is determined whether the pedestrian candidate area to be processed is a pedestrian candidate area obtained by imaging a road sign. In the image example shown in FIG. 3D, since the average luminance value in the inscribed rectangle is high and the luminance variance value is low, the determination target pedestrian candidate region is a pedestrian candidate region obtained by imaging a road sign. It will be judged that there exists and will be excluded from the pedestrian candidate area | region contained in the said 1st pedestrian candidate area | region extraction image.

  FIG. 4 shows a detailed processing procedure of the road sign exclusion process. With reference to this FIG. 4, the process procedure of a road sign exclusion process is demonstrated concretely.

  The first pedestrian candidate area extraction image includes a plurality of pedestrian candidate areas. When the image processing unit 51 proceeds to the road sign exclusion process S50, first, as a process of step S51, an edge (outer shape) included in the pedestrian candidate area labeled with the smallest number among the plurality of pedestrian candidate areas. ) Is extracted (see FIG. 3B).

  When the edge (outer shape) included in this pedestrian candidate area is extracted, the image processing unit 51 determines whether or not a figure similar to a road sign can be detected from this pedestrian candidate area as the determination process in step S52. (See FIG. 3C). Specifically, the outer shape of the road sign installed on the road is limited to several types of outer shapes (figures) such as a circular shape, a triangular shape, and a rectangular shape. The image processing unit 51 stores in advance a standard pattern (template) of a figure limited to several types in an appropriate storage holding unit (not shown). For example, the above-described walking is performed using known template matching, Hough transform, or the like. The stored standard pattern is searched from the candidate area. Specifically, the image processing unit 51 determines whether or not a portion that shows a similarity to the standard pattern and that has a so-called reliability equal to or higher than a predetermined reliability Thc is detected in the pedestrian candidate area. The predetermined reliability Thc is set to a value higher than the reliability between the standard pattern and the outer shape of the pedestrian.

  If a figure similar to the standard pattern cannot be detected from the pedestrian candidate area by this search (“No” in the determination process in step S52), the pedestrian in which the pedestrian candidate area has captured a road sign. This means that the possibility of being a candidate area is low. Then, the image processing unit 51 proceeds to a determination process in step S56 described later.

  On the other hand, when the figure similar to the standard pattern can be detected from the pedestrian candidate area by the search (“Yes” in the determination process in step S52), the pedestrian candidate area walks the road sign. This means that there is a high possibility of being a candidate area. Therefore, the image processing unit 51 sets an inscribed rectangle R that is inscribed in a graphic similar to the standard pattern, and in the subsequent step S53, an average value of luminance data of pixels located in the inscribed rectangle R is used. A certain luminance average value A and a luminance variance value V, which is a variance value of luminance data of pixels located in the inscribed rectangle R, are calculated. Then, the image processing unit 51 determines whether or not “the luminance average value A is higher than the predetermined threshold value Tha and the luminance variance value V is lower than the predetermined threshold value Thv” as the determination processing in step S54.

  Here, when it is determined that “the luminance average value A is higher than the predetermined threshold value Tha and the luminance variance value V is lower than the predetermined threshold value Thv” (“Yes” in the determination processing in step S54), the inscribed rectangle It means that the inside is bright and uniform. Therefore, the image processing unit 51 determines that there is a high possibility that this pedestrian candidate area is a pedestrian candidate area obtained by imaging a road sign, and in the subsequent step S55, this pedestrian candidate area is used as the first pedestrian candidate area. Then, the process proceeds to the determination process in step S56.

  Further, in the determination process of the previous step S54, when it is not determined that “the luminance average value A is higher than the predetermined threshold value Tha and the luminance dispersion value V is lower than the predetermined threshold value Thv” (in the determination process of step S54, “ No ")," the inside of the inscribed rectangle is dark "or" the brightness of the inscribed area is not uniform ". Therefore, the image processing unit 51 determines that this pedestrian candidate area is unlikely to be a pedestrian candidate area obtained by imaging a road sign, and excludes this pedestrian candidate area from the first pedestrian candidate area extracted image. Instead, the process proceeds to the subsequent determination process in step S56.

  When it is determined “No” in the determination process of step S52, when it is determined “No” in the determination process of step S54, or when the process of step S55 is executed, the image processing unit 51 continues. As the determination processing in step S56, it is determined whether or not the processing in steps S51 to S55 has been executed for all pedestrian candidate regions included in the first pedestrian candidate region extraction image.

  When it is determined that the processes in steps S51 to S55 are not executed for all the pedestrian candidate areas included in the first pedestrian candidate area extraction image (“No” in the determination process in step S56), image processing is performed. The unit 51 performs the processing of steps S51 to S55 on the pedestrian candidate region to which the next label is attached, while performing the above steps on all pedestrian candidate regions included in the first pedestrian candidate region extraction image. When it is determined that the processes in S51 to S55 have been executed (“Yes” in the determination process in step S56), the image processing unit 51 proceeds to the subsequent process in step S60 (see FIG. 2). In this way, the image processing unit 51 extracts the first pedestrian candidate area extraction image, which is an image from which a pedestrian candidate area having a high possibility of having captured a road sign is excluded. Converted to an image.

  In the road sign exclusion process S50 described above, when a figure similar to the road sign is detected in the outer shape of the pedestrian candidate area, the average luminance value in the inscribed rectangle is high and the luminance in the inscribed rectangle is high. It was decided whether or not the variance value was low. Thereby, since it is reduced that the pedestrian candidate area | region where the pedestrian was imaged is excluded from the pedestrian candidate area | region contained in the said 1st pedestrian candidate area | region extraction image, the pedestrian is not detected but it does not detect Can be suppressed. The road sign exclusion process S50 corresponds to the road sign exclusion means described in the claims.

  When such a road sign excluding process is executed, the image processing unit 51, as the subsequent process of step S60, may have captured the utility pole reflector from the pedestrian candidate area included in the second pedestrian candidate area extracted image. The utility pole reflector exclusion process for excluding a high pedestrian candidate area is executed.

  Hereinafter, with reference to FIG. 5 and FIG. 6, an outline of the utility pole reflector exclusion process that is particularly effective for excluding utility poles located at a short distance in the utility pole reflector exclusion process will be described. FIG. 5A is a part of an infrared image captured by the infrared camera 10, and a utility pole including a reflector located at a short distance from the infrared camera 10 is captured. Since the reflector provided on the utility pole is often subjected to surface treatment that easily reflects infrared rays, in the infrared image obtained by imaging with the infrared camera 10, the pedestrian candidate in which the utility pole provided with the reflector is imaged The luminance average value of the region is high, and the luminance dispersion value is low. That is, the pedestrian candidate area where the utility pole with the reflector is imaged becomes bright and uniform. For this reason, as described above, the prior art assumes that a pedestrian candidate region having a high luminance average value and a low luminance dispersion value is not a pedestrian candidate region in which a pedestrian is imaged.

  However, as already described in the column of the problem, the above assumption may not be satisfied. For example, when the distance between the infrared camera 10 and the utility pole is close enough to capture even a striped pattern applied to the reflector, the average luminance value of the pedestrian candidate area in which the utility pole is imaged is high, and the luminance dispersion thereof The value may be high. In such a case, it becomes impossible to exclude the pedestrian candidate area in which the utility pole including the reflector is imaged from the pedestrian candidate area included in the second pedestrian candidate area extraction image. was there.

  Therefore, in the present embodiment, the image processing unit 51 has a ratio of a specific direction to the pedestrian candidate area included in the second pedestrian candidate area extracted image in the edge direction included in the pedestrian candidate area. It was decided whether or not would exceed a predetermined ratio.

  Specifically, the image processing unit 51 extracts the edge included in the infrared image shown in FIG. 5A, and acquires the binary image shown in FIG. This binary image is a binary image from which edges included in an infrared image (see FIG. 5A) are extracted. Next, the image processing unit 51 creates a histogram shown in FIG. 6 for each pedestrian candidate region based on the binary image shown in FIG. This histogram is a histogram showing the relationship between the edge direction included in each pedestrian candidate region and its frequency. Here, the image processing unit 51, based on the histogram, among the edge directions included in each pedestrian candidate area, the edge direction along the stripe formed on the utility pole reflector (hereinafter also referred to as a specific edge direction). It is determined whether or not the ratio of the arrow A in FIG. 5B exceeds a predetermined ratio (for example, “15%”). When it is determined that the ratio exceeds the predetermined ratio, it can be determined that the pedestrian candidate area is likely to be a pedestrian candidate area obtained by imaging a utility pole located at a short distance. On the other hand, when it is determined that the ratio is equal to or less than the predetermined ratio, it is possible to determine that the possibility that the pedestrian candidate area is a pedestrian candidate area obtained by imaging a utility pole located at a short distance is low. When the image processing unit 51 determines that the pedestrian candidate area is likely to be a pedestrian candidate area obtained by imaging a utility pole located at a short distance, the pedestrian candidate area is determined as the second pedestrian candidate area. Exclude from the extracted image. In the binary image example shown in FIG. 5B, since the ratio of edges in a specific direction is equal to or greater than a predetermined ratio, this pedestrian candidate area is a pedestrian candidate area obtained by imaging a utility pole located at a short distance. It is determined that the possibility is high, and the pedestrian candidate area included in the second pedestrian candidate area extraction image is excluded.

  Hereinafter, with reference to FIG. 7 (a) and (b), an outline | summary is demonstrated about the utility pole reflector exclusion process effective in excluding especially the utility pole located in a long distance among the utility pole reflector exclusion processes. FIG. 7A is a part of an infrared image captured by the infrared camera 10, and a utility pole including a reflector located at a long distance from the infrared camera 10 is captured. As can be seen from the comparison between FIG. 7A and FIG. 5A, when the utility pole including the reflector is located at a long distance from the infrared camera 10, the stripe pattern applied to the reflector is as follows. Since it will be crushed, it is difficult to extract a striped pattern as an edge.

  Therefore, in the present embodiment, the image processing unit 51, for the pedestrian candidate area included in the second pedestrian candidate area extracted image, the area of the pedestrian candidate area relative to the circumscribed rectangle area of the pedestrian candidate area. It was decided to determine whether or not the labeling density, which is the percentage of

  Specifically, the image processing unit 51 calculates the labeling density of the pedestrian candidate area with respect to the infrared image shown in FIG. Here, the labeling density is the ratio of the area Sb of the pedestrian candidate area to the area Sa of the circumscribed rectangle of the pedestrian candidate area (see FIG. 7B). When the utility pole is an imaged pedestrian candidate area, the outer shape of the pedestrian candidate area is close to a rectangle. Therefore, regardless of the distance from the infrared camera 10, the circumscribed rectangle area Sa of the pedestrian candidate area is the area of the pedestrian candidate area. It becomes close to Sb and the labeling density is close to “1”. On the other hand, when the pedestrian is a pedestrian candidate area in which the pedestrian is imaged, the outer shape thereof is not a rectangle. Therefore, regardless of the distance from the infrared camera 10, the circumscribed rectangle area Sa of the pedestrian candidate area is the pedestrian candidate. The area becomes larger than the area Sb of the region, and the labeling density becomes smaller than “1”. Therefore, the image processing unit 51 determines whether the labeling density of each pedestrian candidate exceeds a predetermined density (for example, “0.75”). When it is determined that the density exceeds the predetermined density, it is possible to determine that the pedestrian candidate area is likely to be a pedestrian candidate area in which the utility pole is imaged, while it is determined that the pedestrian candidate area is equal to or lower than the predetermined density. In this case, it is possible to determine that the possibility that the pedestrian candidate area is a pedestrian candidate area in which the utility pole is imaged is low. When it is determined that there is a high possibility that the utility pole is an imaged pedestrian candidate area, the image processing unit 51 excludes the pedestrian candidate area from the second pedestrian candidate area extracted image. In the binary image example shown in FIG. 7B, since the labeling density is determined to be higher than the predetermined density, this pedestrian candidate area may be a pedestrian candidate area obtained by imaging a utility pole located at a long distance. Therefore, it is excluded from the pedestrian candidate area included in the second pedestrian candidate area extracted image.

  Next, with reference to FIG. 8, the process sequence of a utility pole reflector exclusion process is demonstrated concretely. In addition, FIG. 8 is a flowchart which shows the process sequence of a utility pole reflector exclusion process.

  The second pedestrian candidate area extraction image includes a plurality of pedestrian candidate areas, and a pedestrian candidate area that is highly likely to have been imaged with a road sign is already excluded from the plurality of pedestrian candidate areas. ing.

  When the image processing unit 51 proceeds to the utility pole reflector exclusion process S60, first, as the process of step S61, an edge included in the pedestrian candidate area labeled with the smallest number among these pedestrian candidate areas is selected. Extract (see FIG. 5B). When the edge included in this pedestrian candidate area is extracted, the image processing unit 51 creates a histogram of the edge direction included in this pedestrian candidate area and its frequency as the processing of the subsequent step S62 (see FIG. 6). Subsequently, as a determination process in step S63, it is determined whether or not the ratio of the edge direction (specific edge direction) along the stripe formed on the utility pole reflector out of the edge directions included in the pedestrian candidate area exceeds a predetermined ratio. to decide.

  Here, when it is determined that the ratio of the specific edge direction exceeds the predetermined ratio (“Yes” in the determination process in step S63), there is a high possibility that the utility pole is located in a short distance and is a pedestrian candidate area. It is possible to judge. Therefore, the image processing unit 51 excludes the pedestrian area from the second pedestrian candidate area extracted image as the subsequent process of step S64. Then, the image processing unit 51 proceeds to the determination process in the subsequent step S65.

  On the other hand, when it is determined in the determination process of the previous step S63 that the ratio of the specific edge direction is less than the predetermined ratio (“No” in the determination process of step S63), the walking in which the electric pole located at a short distance is imaged. It is possible to determine that the possibility of the candidate area is low. Then, the image processing unit 51 calculates a labeling density that is a ratio of the area Sb of the pedestrian candidate area to the area Sa of the circumscribed rectangle of the pedestrian candidate area as the process of subsequent step S66, and then continues the determination process of step S67. As described above, it is determined whether or not the labeling density of the pedestrian candidate area exceeds a predetermined density.

  Here, when it is determined that the labeling density exceeds the predetermined density (“Yes” in the determination process in step S67), the pedestrian candidate area is a pedestrian candidate area in which a utility pole located at a long distance is imaged. It is possible to determine that the possibility is high. Therefore, the image processing unit 51 excludes this pedestrian area from the second pedestrian candidate area extraction image as the process of the previous step S64, and proceeds to the subsequent determination process of step S65.

  On the other hand, when it is determined that the labeling density is equal to or lower than the predetermined density (“No” in the determination process in step S67), the pedestrian candidate area may be a pedestrian candidate area obtained by imaging a utility pole located at a long distance. Can be determined to be low. Therefore, the image processing unit 51 proceeds to the subsequent determination process of step S65 without excluding this pedestrian area from the second pedestrian candidate area extraction image.

  When the process of step S64 is executed, or when “No” is determined in the determination process of step S67, the image processing unit 51 applies the second pedestrian candidate area extraction image as the determination process of step S65. It is determined whether or not the processing in steps S61 to 67 has been executed for all the included pedestrian candidate regions.

  When it is determined that the processes in steps S61 to 67 are not executed for all the pedestrian candidate areas included in the second pedestrian candidate area extraction image (“No” in the determination process in step S65), image processing is performed. The unit 51 performs the process of steps S61 to S67 on the pedestrian candidate area labeled with the next label, while performing the above steps on all pedestrian candidate areas included in the second pedestrian candidate area extraction image. When it is determined that the processing of S61 to 67 has been executed (“Yes” in the determination processing of step S65), the image processing unit 51 proceeds to the processing of the subsequent step S70 (see FIG. 2). In this manner, the second pedestrian candidate extracted image is converted into a third pedestrian candidate region extracted image that is an image from which a pedestrian candidate region having a high possibility that the utility pole has been imaged is excluded by the image processing unit 51. Converted.

  And CPU41 detects the pedestrian who exists in the periphery of a vehicle based on a 3rd pedestrian candidate area | region extraction image. The CPU 41 corresponds to pedestrian detection means described in the claims.

  In the utility pole reflector removal processing S60 described above, the image processing unit 51 performs an edge direction included in the determination target pedestrian area with respect to the pedestrian candidate area included in the second pedestrian candidate area extraction image. It is determined whether or not the ratio in a specific direction exceeds a predetermined ratio. Here, when determining that the image processing unit 51 exceeds the predetermined ratio, the image processing unit 51 determines that the pedestrian candidate area is likely to be a pedestrian candidate area obtained by imaging a utility pole located at a short distance, while the predetermined ratio or less. When it is determined that the pedestrian candidate area is, it is determined that the possibility that the pedestrian candidate area is a pedestrian candidate area obtained by imaging a utility pole located at a short distance is low. When the image processing unit 51 determines that the pedestrian candidate area is likely to be a pedestrian candidate area obtained by imaging a utility pole located at a short distance, the pedestrian candidate area is determined as the second pedestrian candidate area. Excluded from the extracted image. Thereby, since it becomes possible to exclude the pedestrian candidate area in which the utility pole including the reflector is imaged from the pedestrian candidate area included in the second pedestrian candidate area extraction image, the utility pole including the reflector is provided. It is possible to suppress the occurrence of false detection of detecting a pedestrian as a pedestrian.

  In the utility pole reflector exclusion process S60 described above, the image processing unit 51 determines the labeling density of the pedestrian candidate area to be determined for the pedestrian candidate area included in the second pedestrian candidate area extracted image. It is determined whether or not the density is equal to or higher than a predetermined density. Here, when determining that the image processing unit 51 exceeds the predetermined density, the image processing unit 51 determines that the pedestrian candidate region is likely to be a pedestrian candidate region in which the utility pole is imaged, and determines that the pedestrian candidate region is below the predetermined density. In this case, it is determined that the possibility that the pedestrian candidate area is a pedestrian candidate area in which the utility pole is imaged is low. Then, when the image processing unit 51 determines that there is a high possibility that the power pole is a pedestrian candidate region in which the power pole is imaged, the image processing unit 51 excludes the pedestrian candidate region from the second pedestrian candidate region extraction image. Thereby, since it becomes possible to exclude the pedestrian candidate area in which the utility pole including the reflector is imaged from the pedestrian candidate area included in the second pedestrian candidate area extraction image, the utility pole including the reflector is provided. It is possible to suppress the occurrence of false detection of detecting a pedestrian as a pedestrian. In addition, this road sign exclusion process S60 is equivalent to the utility pole reflector exclusion means described in the claims.

DESCRIPTION OF SYMBOLS 1 ... Vehicle periphery monitoring system, 10 ... Infrared camera, 20 ... Floodlight, 21 ... Switch relay, 30 ... Pedestrian detection apparatus, 41 ... CPU, 42 ... ROM, 43 ... RAM, 51 ... Image processing means, 52 ... VRAM

Claims (9)

A high-brightness region extraction image generation unit that generates a high-brightness region extraction image that is an image obtained by extracting a high-brightness region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle;
Pedestrian candidate area extraction image generation means for generating a pedestrian candidate area extraction image, which is an image obtained by extracting a pedestrian candidate area from the high luminance area extraction image, based on an aspect ratio of a circumscribed rectangle circumscribing the high luminance area When,
The utility pole reflector exclusion which excludes the pedestrian candidate area determined to be a pedestrian candidate area in which the utility pole reflector is imaged based on the edge direction included in the pedestrian candidate area from the pedestrian candidate area extraction image Means,
Pedestrian detection for detecting pedestrians present in the vicinity of a vehicle based on the pedestrian candidate area extraction image after the pedestrian candidate area in which the utility pole reflector is imaged is excluded by the utility pole reflector exclusion means And a pedestrian detection device. A high-brightness region extraction image generation unit that generates a high-brightness region extraction image that is an image obtained by extracting a high-brightness region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle;
Pedestrian candidate area extraction image generation means for generating a pedestrian candidate area extraction image, which is an image obtained by extracting a pedestrian candidate area from the high luminance area extraction image, based on an aspect ratio of a circumscribed rectangle circumscribing the high luminance area When,
Based on a labeling density that is a ratio of the area of the pedestrian candidate area to the area of the circumscribed rectangle of the pedestrian candidate area, a pedestrian candidate area determined to be a pedestrian candidate area in which a utility pole reflector is imaged , Utility pole reflector exclusion means for excluding from the pedestrian candidate area extraction image,
Pedestrian detection for detecting pedestrians present in the vicinity of a vehicle based on the pedestrian candidate area extraction image after the pedestrian candidate area from which the utility pole reflector is imaged is excluded by the utility pole reflector exclusion means And a pedestrian detection device. A high-brightness region extraction image generation unit that generates a high-brightness region extraction image that is an image obtained by extracting a high-brightness region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle;
Pedestrian candidate area extraction image generation means for generating a pedestrian candidate area extraction image, which is an image obtained by extracting a pedestrian candidate area from the high luminance area extraction image, based on an aspect ratio of a circumscribed rectangle circumscribing the high luminance area When,
Exclude from the pedestrian candidate area extracted image the pedestrian candidate area determined to be a pedestrian candidate area in which the utility pole reflector is imaged based on the edge direction included in the pedestrian candidate area; and
Based on a labeling density that is a ratio of the area of the pedestrian candidate area to an area of a circumscribed rectangle of the pedestrian candidate area, a pedestrian candidate area determined to be a pedestrian candidate area in which a utility pole reflector is imaged. , Utility pole reflector exclusion means for excluding from the pedestrian candidate area extraction image,
Pedestrian detection for detecting pedestrians present in the vicinity of a vehicle based on the pedestrian candidate area extraction image after the pedestrian candidate area in which the utility pole reflector is imaged is excluded by the utility pole reflector exclusion means And a pedestrian detection device. In the pedestrian detection device according to claim 1 or 3,
The utility pole reflector exclusion means determines whether or not a ratio of a specific edge direction among edge directions included in the pedestrian candidate area is a predetermined ratio or more, and determines that the ratio is a predetermined ratio or more. The pedestrian detection device is characterized by determining that the pedestrian candidate region is a pedestrian candidate region obtained by imaging the utility pole reflector. In the pedestrian detection device according to claim 4,
The pedestrian detection apparatus according to claim 1, wherein the specific edge direction is a direction along a stripe formed on the utility pole reflector. In the pedestrian detection device according to claim 2 or 3,
The electric pole reflector exclusion means determines whether the labeling density is equal to or higher than a predetermined density, and based on determining that the labeling density is equal to or higher than the predetermined density, the pedestrian candidate area walks the electric pole reflector. A pedestrian detection device, characterized in that it is determined to be a pedestrian candidate area. In the pedestrian detection device according to claim 6,
The said predetermined density is set to the density higher than the labeling density of the pedestrian located in the periphery of a vehicle, The pedestrian detection apparatus characterized by the above-mentioned. In the pedestrian detection apparatus as described in any one of Claims 1-7,
Based on the outer shape of the pedestrian candidate area, the average luminance value in the inscribed rectangle inscribed in the outer shape, and the luminance variance value in the inscribed rectangle inscribed in the outer shape, road signs installed on the road A pedestrian detection apparatus, further comprising a road sign excluding unit for excluding a pedestrian candidate area determined to be an imaged pedestrian candidate area from the pedestrian candidate area extraction image. A high-brightness region extraction image generation unit that generates a high-brightness region extraction image that is an image obtained by extracting a high-brightness region having a luminance equal to or higher than a predetermined luminance from an infrared image captured by an in-vehicle imaging device that images the periphery of the vehicle;
Pedestrian candidate area extraction image generation for generating a pedestrian candidate area extraction image that is an image obtained by extracting a pedestrian candidate area from the high luminance area extraction image based on an aspect ratio of a circumscribed rectangle circumscribing the high luminance area Means,
Based on the outer shape of the candidate area for the pedestrian, the average luminance value in the inscribed rectangle inscribed in the outer shape, and the luminance variance value in the inscribed rectangle inscribed in the outer shape, a road sign installed on the road Road sign excluding means for excluding a pedestrian candidate area determined to be an imaged pedestrian candidate area from the pedestrian candidate area extraction image;
A pedestrian detection device comprising: pedestrian detection means for detecting a pedestrian existing around a vehicle based on the pedestrian candidate region extraction image.