JP2006010652A - Object-detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an object-detecting device for separating objects to be detected which are overlapped on an image and capable of recognizing each of the objects to be detected, in detecting the objects from the still image. <P>SOLUTION: The object-detecting device 1 is provided with cameras 10 and 11 for photographing the images; a detection part 30 for extracting a search area, containing a candidate image of a pedestrian from the images photographed by the cameras 10 and 11 and for outputting the coordinate values of the search area and the evaluation value of the search area, according to the similarity of the pedestrian candidates included in the search area with the pedestrians; a separation processing part 42 for separating the search area for each pedestrian, on the basis of the coordinate values of the search area outputted from the detection part 30; and a representative value preparing part 43 for calculating the position of the pedestrians by taking the weighted average of the coordinate values of the search area for each pedestrian separated by the separation processing part 42, according to the evaluation value. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an object detection device.

  When detecting an object (for example, a pedestrian, a bicycle, a vehicle, or the like) that is a detection target from an image captured by a camera, a plurality of detection target objects may overlap on the image. When the number of detection target objects, position information of each detection target object, and the like are necessary, it is necessary to separate each detection target object and recognize each detection target object individually.

A technique for removing such duplication of detection results is described in Patent Document 1. In the technique described in Patent Document 1, first, a difference between input image data and image data input immediately before is obtained, and a difference image is generated. Furthermore, the difference image of past m frames (m is an integer of 2 or more) is added and averaged to obtain an integrated difference image. The integrated difference image has a pixel value of 0 in a non-motion area, and a larger pixel value in a motion area. When an integrated difference image is obtained, first, the presence / absence of a motion region is checked. When there is no moving area, the number of persons is zero. When there is a motion region, for example, one person is assigned when the maximum value of the motion region width in the upper region of the integrated differential image is smaller than a certain threshold value, and two people when the maximum value is greater. When the number of persons is two, the integrated difference area is divided into two partial areas. When three or more people are detected, the number of divisions is increased.
JP 2000-222576 A

  In the technique described in Patent Document 1, when an image is divided according to the number of persons, it is necessary to acquire and process image data for m frames. Therefore, the technique described in Patent Document 1 cannot be applied when removing duplicated extraction results from still images of only one frame.

  The present invention has been made to solve the above-described problems. When an object is detected from a still image, the detection target object overlapping in the image is separated and each detection target object is recognized. It is an object to provide an object detection device capable of

  An object detection apparatus according to the present invention extracts an image area including a candidate image of a detection target object from an image picked up by an image pickup means for picking up an image, and the image picked up by the image pickup means. An extraction means for outputting an evaluation value of the image area corresponding to the degree of similarity between the detection target object candidate and the detection target object included in the image, and the detection of the image area based on the coordinate value of the image area output from the extraction means An object for calculating the position of the detection target object by taking a weighted average according to the evaluation value with respect to the coordinate value of the image area for each detection target object classified by the classification means and the classification means for classification for each object And a position calculating means.

  According to the object detection device of the present invention, for example, when a plurality of detection target object candidates are extracted for each of a plurality of detection target objects, first, they are included in the plurality of image regions based on the coordinate values of the image regions. By determining whether the detection target object candidates correspond to the same detection target object or different detection target objects, a plurality of image regions are divided for each detection target object. Then, the coordinate values of the image areas belonging to the same detection target object are weighted and averaged according to the evaluation value to calculate the position of the detection target object. Therefore, it is possible to separate the detection target objects overlapping on the image and to specify the positions of the detection target objects with high accuracy.

  An object detection apparatus according to the present invention extracts an image area including a candidate image of a detection target object from an image picked up by an image pickup means for picking up an image, and the image picked up by the image pickup means. An extraction means for outputting an evaluation value of the image area corresponding to the degree of similarity between the detection target object candidate and the detection target object included in the image, and the detection of the image area based on the coordinate value of the image area output from the extraction means Classification means for classifying each object, and object position calculation means for calculating the position of the detection target object from the coordinate value of the image area having the maximum evaluation value among the image areas classified for each detection target object by the classification means It is characterized by providing.

  When the position of the detection target object is calculated from the coordinate value of the image area having the maximum evaluation value, the processing man-hour is reduced as compared with the case where the weighted average is taken, so that the processing can be speeded up.

  In addition, when a plurality of image regions are extracted, the object detection device according to the present invention determines whether or not the number of detection target objects is a plurality based on the horizontal deviation of the plurality of extracted image regions. It is preferable that the image forming apparatus further includes an object number determination unit, and when the object number determination unit determines that there are a plurality of detection target objects, the classification unit separates the plurality of image regions for each detection target object.

  In this case, the number of detection target objects is obtained based on the horizontal deviation of the plurality of image regions. For example, when the horizontal deviation is less than or equal to a predetermined value, it is determined that there is one detection target object, and when the deviation is greater than the predetermined value, it is determined that there are a plurality of detection target objects. When there are a plurality of detection target objects, the image area is divided for each detection target object. Therefore, it is possible to efficiently separate overlapping detection objects.

  According to the present invention, when an object is detected from a still image, it is possible to separate detection target objects that overlap in the image and recognize each detection target object.

  DESCRIPTION OF EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. In the figure, the same reference numerals are used for the same or corresponding parts.

(First embodiment)
First, the overall configuration of the object detection apparatus 1 according to the first embodiment will be described with reference to FIG. FIG. 1 is a block diagram showing an overall configuration of an object detection apparatus 1 according to the first embodiment. The object detection apparatus 1 includes cameras (imaging means) 10 and 11 that capture an image of a landscape in front of a vehicle and acquire image data, and performs image processing on the image data, thereby, for example, an object such as a pedestrian or a vehicle. And an object detection electronic control device (hereinafter referred to as “object detection ECU”) 13 for calculating the position information and the like. Hereinafter, a case where a pedestrian is detected as a detection target object will be described as an example.

  Each of the camera 10 and the camera 11 is a CCD camera, for example, and is installed facing the front at the upper part of the front window of the host vehicle (for example, the back side of the rearview mirror). The optical axes of the two cameras are parallel to each other (which may be substantially parallel), and the horizontal axes of the imaging surfaces are arranged on the same line. Each of the cameras 10 and 11 captures a landscape in front of the vehicle and acquires image data. Image data acquired by the cameras 10 and 11 is output to the object detection ECU 13. In addition, the installation location of the cameras 10 and 11 is not limited to the upper part of the front window, and may be provided at any position on the vehicle body as long as it can capture an image in front of the vehicle.

  The object detection ECU 13 includes a microprocessor that performs calculations, a ROM that stores a program for causing the microprocessor to execute each process, a RAM that stores various data such as calculation results, and a backup in which the stored contents are held by a 12V battery. It is comprised by RAM etc.

  The object detection ECU 13 is a detection target in the candidate area calculation unit 20 that acquires a candidate area where a pedestrian is likely to exist from the image data acquired by the cameras 10 and 11, and the candidate area acquired by the candidate area calculation unit 20. A detection unit 30 that extracts and identifies object (pedestrian) candidates, a duplication removal unit 40 that removes duplication of pedestrian candidates extracted by the detection unit 30, and a pedestrian recognized by the duplication removal unit 40, respectively. And a detection result output unit 50 that outputs position information and the like.

  Based on the image data input from the cameras 10 and 11, the candidate area calculation unit 20 determines the position of the candidate area (see FIG. 4) on the image where a pedestrian is likely to exist and the candidates from the cameras 10 and 11. The distance to the area is calculated and output to the detection unit 30. The candidate area calculation unit 20 includes a distance calculation unit 21 and a candidate calculation unit 22.

  The distance calculation unit 21 calculates the distance from the host vehicle to the object imaged on the image based on the difference between the image data input from the camera 10 and the image data input from the camera 11. The distance to the object can be calculated based on the principle of triangulation using, for example, stereo vision image data. An example of a distance calculation method will be briefly described.

When the object is in front of the host vehicle, when the image obtained from the camera 10 and the image obtained from the camera 11 are superimposed, the object is shifted in the horizontal direction. Therefore, the most overlapping position is obtained while shifting one image pixel by pixel. The number of pixels shifted at this time is n. If the focal length of the lens constituting the camera 10 or 11 is f, the distance between the optical axes of the left and right cameras 10 and 11 is L, and the pixel pitch is d, the distance R from the vehicle to the object is Can be calculated by
R = (f · L) / (n · d) (1)

  First, the candidate calculation unit 22 determines, for example, the difference in distance between the object part and the background part, the variation in distance, and the size of the object part (for example, the width) from the distance information calculated by the distance calculation unit 21. ) Etc. are calculated. Next, based on the calculated information, a candidate area on an image in which an object having a high possibility of a pedestrian is captured is acquired. The acquired candidate area information is output to the detection unit 30.

  The detection unit 30 includes a search control unit 31 that sets a search area (see FIG. 4), a feature extraction unit 32 that extracts a feature portion from image data, and a pedestrian that extracts a pedestrian using a pedestrian pattern. A detection unit 33 is included.

  In the search control unit 31, a search region for searching for a pedestrian is set in each candidate region acquired by the candidate region calculation unit 20. The search area is set smaller than the candidate area. Further, the size of the exploration area is set according to, for example, the size of the pedestrian and the distance from the host vehicle. Further, the exploration area is sequentially shifted, and pedestrians are extracted in the entire candidate area.

  The feature extraction unit 32 and the pedestrian detection unit 33 perform extraction and identification based on pedestrian candidate patterns within the set search region. For example, techniques such as neural network, SVM (Support Vector Machine), and template matching are used to extract and identify pedestrian candidates.

  Information on the search region (image region) from which the pedestrian candidate is extracted and detection data including an evaluation value corresponding to the similarity between the pedestrian candidate and the pedestrian pattern on the image are output to the duplication removal unit 40. The That is, the detection unit 30 functions as an extraction unit. The search area information includes, for example, the distance from the host vehicle, the boundary line, and the upper left and lower right coordinate values of the search area. The evaluation value increases as the degree of similarity increases, that is, as the possibility of being a pedestrian increases.

  Here, in pattern matching in which pedestrians are extracted by comparing images captured by the cameras 10 and 11 with pedestrian patterns, the pedestrian image is not at the center of the search area, or the size varies. In consideration of the case where there is a pedestrian, the pedestrian is adjusted so that the pedestrian can be detected. Therefore, when the search area is sequentially shifted in the candidate area and pedestrians are extracted, a plurality of overlapping detection results can be output for the same pedestrian.

  The duplication removal unit 40 removes duplicate extraction results of the same pedestrian output from the detection unit 30 and separates the extraction results that are duplicated on the image from the extraction results for a plurality of pedestrians. To recognize. The duplicate removal unit 40 includes a number estimation unit 41, a classification processing unit 42, and a representative value creation unit 43.

  The number estimating unit 41 estimates the number of pedestrians included in the candidate area based on the extraction result of pedestrians in the candidate area. The number-of-people estimation unit 41 estimates the number of pedestrians based on the horizontal deviation of the coordinate values of each of the higher order data and the threshold value based on the lateral width of the pedestrian with respect to the higher order data of the evaluation value. . That is, the number estimation unit 41 functions as an object number determination unit. The estimation result of the pedestrian is output to the classification processing unit 42.

  When there are a plurality of pedestrians estimated by the number estimating unit 41, the classification processing unit 42 obtains search area information or the like from which pedestrians are extracted based on the coordinate value of the search area and the width of the pedestrian. Sort by pedestrian. That is, the classification processing unit 42 functions as a classification unit. The sorted result is output to the representative value creation unit 43.

  The representative value creation unit 43 calculates a representative value corresponding to the position of the pedestrian by taking a weighted average according to the evaluation value for the coordinate value of the search area divided for each pedestrian. is there. That is, the representative value creation unit 43 functions as an object position calculation unit. The calculated representative value is output to the detection result output unit 50.

  The detection result output part 50 converts the output result of the duplication removal part 40 into the format which a driver | operator requires, and outputs it.

  Next, the operation of the object detection device 1 will be described with reference to FIGS. FIG. 2 is a flowchart showing a processing procedure of pedestrian extraction processing by the object detection device 1. FIG. 3 is a flowchart showing a processing procedure of the duplicate removal processing by the object detection device 1. This process is repeatedly executed every predetermined time. First, the pedestrian extraction process will be described with reference to FIG.

  In step S100, one of the rectangular candidate areas acquired by the candidate area calculating unit 20 is selected. In subsequent step S102, a rectangular search region for extracting pedestrians is set in the candidate region selected in step S100. The size of this search area is set according to the size of the pedestrian and the distance from the host vehicle, for example.

  In step S104, feature extraction of the image data is performed in the search area set in step S102. In subsequent step S106, pedestrians are extracted by performing pattern matching on the image data on which the feature extraction has been performed. And detection data, such as an evaluation value according to the coordinate value and similarity of the search area | region where the pedestrian was extracted, are produced.

  In step S108, a determination is made as to whether or not the pedestrian extraction process has been completed in all search areas within the selected candidate area. If the pedestrian extraction process has been completed in all search areas, the process proceeds to step S110. On the other hand, when there is a search area where the pedestrian extraction process is not performed, the process proceeds to step S102. In step S102, the next search area is set, and the processes in steps S104 and S106 are executed. Since the processing content of steps S104 and S106 is the same as the processing content described above, description thereof is omitted here. As described above, the processes in steps S102 to S108 are repeatedly executed, so that the pedestrian extraction process is performed in all the candidate areas selected in step S100.

  In step S110, a determination is made as to whether or not pedestrian extraction processing has been performed in all candidate regions acquired by the candidate region calculation unit 20. Here, if the extraction process is completed for all candidate areas, the process temporarily exits. On the other hand, if there remains a candidate area for which extraction processing has not been performed, the processing moves to step S100. In step S100, the next candidate area is selected, and the processes in steps S102 to S108 are executed. Since the processing content of steps S102 to S108 is the same as the processing content described above, the description thereof is omitted here. As described above, the processes of steps S100 to S110 are repeatedly performed, so that the pedestrian extraction process is performed in all candidate areas acquired by the candidate area calculation unit 20.

  Next, the deduplication processing will be described with reference to FIG. First, in step S200, a candidate area to be subjected to deduplication processing is selected.

  In the subsequent step S202, the number of pedestrians in the selected candidate area is estimated by the number estimating unit 41. Here, the algorithm for estimating the number of people executed in step S202 will be described.

  First, the detection data in the selected candidate area is sorted in order from the largest evaluation value. Next, the sorted NUMBER_HIGHVAL_DATA pieces of detected data are extracted. In this embodiment, NUMBER_HIGHVAL_DATA is set to 10. That is, ten detection data with a high possibility of a pedestrian are acquired. Next, a horizontal deviation σx of the search area of the acquired detection data is obtained. A judgment reference value dw is set. Here, a value obtained by multiplying the horizontal width of the search area by a predetermined value RATIO_HUMAN_WIDTH is used as dw. In the present embodiment, the value of RATIO_HUMAN_WIDTH is set to 0.33 so that the determination reference value dw is approximately the width of the pedestrian. This is because when the pedestrians are adjacent to each other, the distance between the line segments passing through the centers of the respective pedestrians is about the width of the person. Then, the deviation σx is compared with the determination reference value dw. If σx ≦ dw, it is determined that there is one pedestrian, and if σx> dw, it is determined that there are a plurality of pedestrians.

  Next, in step S204, a determination is made as to whether or not the number of pedestrians estimated in step S202 is one. Here, when the number of pedestrians is one, the process proceeds to step S208. On the other hand, when the number of pedestrians is plural, the process proceeds to step S206.

  When it is determined that there are a plurality of pedestrians, in step S206, the classification processing unit 42 performs a classification process for classifying the detection data for each pedestrian. Here, the algorithm of the classification process executed in step S206 will be described.

  Depending on the xy coordinates of the center of the search area from which the pedestrian has been extracted, for example, an isodata (iterative self organizing data analysis techniques A) algorithm is applied to classify the search area for each pedestrian.

  A deviation σ of the distance between the cluster center and the search area belonging to the cluster is obtained. For the cluster having the largest deviation σ, if the deviation σ is larger than ½ of the criterion value dw, for example, the cluster is assumed to include a plurality of persons, and division processing is performed. In the present embodiment, the threshold value is set to dw / 2 in order to determine that they are separate pedestrians when the distance between them is more than about half of the lateral width of the pedestrian.

  Further, the distance between the cluster centers is obtained, and if the minimum distance is smaller than a predetermined threshold value dw / 4, for example, the two clusters are merged with the same cluster. This is because when the distance between the cluster centers is smaller than 1/4 of the determination reference value dw, it is determined that they are the same person.

  When the distance between the old and new cluster centers becomes 1 or less, the clustering is terminated. In the present embodiment, the clustering is finished when it reaches 0.1. This is based on the idea that the cluster center does not have to move as a pixel value. In this way, a plurality of search areas are classified for each cluster, that is, for each pedestrian.

  Next, in step S208, the representative value is calculated by the representative value creating unit 43. In step S208, by taking the weighted average according to the evaluation value for the coordinate value of the search area for each pedestrian, the search area with the highest evaluation value among the search areas classified by each pedestrian is taken as a target. The coordinates of each pedestrian, that is, the position of each pedestrian is specified.

  In subsequent step S210, a determination is made as to whether or not pedestrian classification processing has been performed in all candidate regions. Here, when the classification process has been completed for all candidate areas, the process temporarily exits. On the other hand, if there remains a candidate area for which the separation process has not been performed, the process proceeds to step S200. In step S200, the next candidate area is selected, and the processes in steps S202 to S208 are executed. Since the processing content of steps S202 to S208 is the same as the processing content described above, description thereof is omitted here. As described above, the processes of steps S200 to S210 are repeatedly executed, so that the pedestrian classification process is performed in all candidate areas.

  Thus, according to the object detection device 1, when a plurality of search areas including pedestrian images are extracted for each of a plurality of pedestrians, first, walking included in the search area based on the coordinate values of the search area. By determining whether a person corresponds to the same pedestrian or a different pedestrian, a plurality of search areas are divided for each pedestrian. Then, the coordinate values of the search areas belonging to the same pedestrian are averaged with a weight according to the evaluation value, and the position of the pedestrian is calculated. Thus, since the pedestrian which overlaps on an image can be isolate | separated, it becomes possible to recognize each pedestrian.

  Further, according to the present embodiment, the number of pedestrians in the candidate area is obtained based on the horizontal deviation σx of the plurality of search areas. And when there are a plurality of pedestrians, the search area is divided for each pedestrian. As a result, overlapping pedestrians can be separated efficiently.

  Here, an example of the extraction result of the pedestrian by the detection part 30 is shown in FIG. A broken-line rectangular frame is a candidate area, and a dashed-dotted rectangular frame is a search area. As shown in FIG. 4, a plurality of search areas are detected for one pedestrian. In addition, each search area is detected in an overlapping manner.

  An example of the duplicate removal result by the duplicate removal unit 40 is shown in FIG. As shown in FIG. 5, duplication of each search area that has been detected in duplicate is removed. From the above, the effectiveness of the present embodiment was confirmed.

(Second Embodiment)
Next, the overall configuration of the object detection apparatus 2 according to the second embodiment will be described with reference to FIG. FIG. 6 is a block diagram showing the overall configuration of the object detection apparatus 2 according to the second embodiment. In FIG. 6, the same or equivalent components as those in the first embodiment are denoted by the same reference numerals.

  The present embodiment differs from the first embodiment in that a representative value creation unit 43A is provided instead of the representative value creation unit 43. Other configurations are the same as or similar to those of the first embodiment, and thus description thereof is omitted here.

  The representative value creating unit 43A determines the position of the pedestrian from the coordinate value of the image area having the maximum evaluation value, that is, the image area including the image data having the highest similarity, among the image areas classified for each pedestrian by the classification unit. Is identified.

  According to the present embodiment, since the position of the pedestrian is calculated from the coordinate value of the image area having the maximum evaluation value, it is possible to reduce the processing man-hour compared with the case where the weighted average is taken. As a result, the processing can be speeded up.

  The weighted average is calculated using the center coordinates of the image area. Therefore, before and after performing the weighted averaging process, it is necessary to perform conversion from the upper left and lower right coordinates of the rectangular image region to the center coordinates and reverse conversion. According to the present embodiment, it is not necessary to perform conversion and reverse conversion from the upper left and lower right coordinates of the rectangular image region to the center coordinates, and the detection data can be output as a representative value as it is.

(Third embodiment)
Next, the overall configuration of the object detection apparatus 3 according to the third embodiment will be described with reference to FIG. FIG. 7 is a block diagram showing the overall configuration of the object detection device 3 according to the third embodiment. In FIG. 7, the same or equivalent components as those in the first embodiment are denoted by the same reference numerals.

  This embodiment is different from the first embodiment in that a separation processing unit 42A is provided instead of the separation processing unit 42. Other configurations are the same as or similar to those of the first embodiment, and thus description thereof is omitted here.

  The classification processing unit 42A performs clustering on the basis of detection data obtained by adding one or both of the distance z and the evaluation value v to the coordinate values x and y of the center of the search area. It is something to sort into.

However, each of the coordinate values x, y, the distance z, and the evaluation value v needs to be normalized because the range, variance, and the like that the value can take are different. The coordinate values x, y and the distance z can be normalized by the following equation, for example.
x _n = (x−m _x ) / σ _x (2)
y _n = (y−m _y ) / σ _y (3)
z _n = (z−m _z ) / σ _z (4)
Here, m is an average value, and σ is a standard deviation.

For the evaluation value v, for example, by using a logistic function as shown in the following equation, the domain [0, ∞) of the evaluation value v is normalized to the range of the value range [0,1) of f. be able to.
f (x) = 1 / (1 + e ^−kv ) (5)
Here, k is a parameter that determines the change of f according to the output value of v. For example, k = 3.

  Note that the clustering process is the same as or similar to the process in step S206, and thus the description thereof is omitted here.

  According to the present embodiment, since the clustering is performed based on the data obtained by adding the distance z and the evaluation value v to the coordinate values x and y of the center of the search area, it is possible to improve the accuracy of the classification process. .

(Fourth embodiment)
Next, the overall configuration of the object detection device 4 according to the fourth embodiment will be described with reference to FIG. FIG. 8 is a block diagram illustrating an overall configuration of the object detection device 4 according to the fourth embodiment. In FIG. 8, the same or equivalent components as those in the first embodiment are denoted by the same reference numerals.

  This embodiment is different from the first embodiment in that a distance measuring device 12 that measures the distance from the host vehicle is provided instead of the camera 11, and the distance calculation unit 21 and the candidate calculation unit 22 are replaced. The sensor coordinate system candidate calculation unit 23 and the image coordinate system candidate correspondence unit 24 are provided. Other configurations are the same as or similar to those of the first embodiment, and thus description thereof is omitted here.

  The distance measuring device 12 measures a distance from the host vehicle to an object such as a pedestrian ahead, and for example, a millimeter wave radar or a laser radar is preferably used.

  From the distance information to the object measured by the distance measuring device 12, the sensor coordinate system candidate calculation unit 23, for example, the difference in distance between the object part and the background part, the variation in distance, and the size of the object part (for example, (Width) and the like are calculated. Next, based on the calculated information, a candidate area on an image in which an object having a high possibility of a pedestrian is captured is acquired. The acquired candidate area information is output to the detection unit 30.

  The image coordinate system candidate corresponding unit 24 performs processing for taking correspondence between the sensor coordinate system and the camera coordinate system.

  According to the object detection device 4, since the distance from the own vehicle to the pedestrian can be directly measured by the distance measurement device 12, a distance calculation process using a stereo camera becomes unnecessary.

(Fifth embodiment)
Next, the overall configuration of the object detection apparatus 5 according to the fifth embodiment will be described with reference to FIG. FIG. 9 is a block diagram showing the overall configuration of the object detection apparatus 5 according to the fifth embodiment. In FIG. 9, the same or equivalent components as those in the first embodiment are denoted by the same reference numerals.

  This embodiment is different from the first embodiment in that the number-of-people estimation unit 41 is not provided.

  The operation of the object detection device 3 will be described with reference to FIG. FIG. 10 is a flowchart illustrating the processing procedure of the duplicate removal processing by the object detection device 5.

  In step S300, a candidate area to be subjected to deduplication processing is selected. In subsequent step S302, a classification process is performed in which the detection data is classified for each pedestrian. Since the classification process in step S302 is the same as the process in step S206, description thereof is omitted here.

  Subsequently, in step S304, a representative value is calculated. In step S304, by taking the weighted average according to the evaluation value for the coordinate value of the image area for each pedestrian, the detection data with the highest evaluation value among the detection data classified for each pedestrian is used as a target. The coordinates of each pedestrian, that is, the position of each pedestrian is specified.

  In subsequent step S306, a determination is made as to whether or not pedestrian classification processing has been performed in all candidate regions. Here, when the classification process has been completed for all candidate areas, the process temporarily exits. On the other hand, if there remains a candidate area for which the separation process has not been performed, the process proceeds to step S300. In step S300, the next candidate area is selected, and the processes in steps S302 and S304 are executed. Since the processing content of steps S302 and S304 is the same as the processing content described above, description thereof is omitted here. As described above, the processes of steps S300 to S306 are repeatedly executed, so that the pedestrian classification process is performed in all candidate areas.

  In the present embodiment, direct clustering is performed without performing the number of people estimation process. According to the present embodiment, since the number of persons estimation process can be omitted, the processing speed can be further increased.

  Although the embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications can be made. For example, the function sharing of each processing unit of the object detection ECU 13 constituting the object detection device 1 is not limited to the above embodiment.

  Moreover, although the case where the object detection apparatus 1 was mounted in the vehicle was demonstrated in description of the said embodiment, the object detection apparatus 1 can also be mounted in a robot etc., for example. Furthermore, it can also be used for indoor detection and monitoring of people.

It is a block diagram which shows the whole structure of the object detection apparatus which concerns on 1st Embodiment. It is a flowchart which shows the process sequence of the pedestrian candidate extraction process by the object detection apparatus which concerns on 1st Embodiment. It is a flowchart which shows the process sequence of the duplication removal process by the object detection apparatus which concerns on 1st Embodiment. It is a figure which shows an example of the pedestrian extraction result by a detection part. It is a figure which shows an example of the duplication removal result by a duplication removal part. It is a block diagram which shows the whole structure of the object detection apparatus which concerns on 2nd Embodiment. It is a block diagram which shows the whole structure of the object detection apparatus which concerns on 3rd Embodiment. It is a block diagram which shows the whole structure of the object detection apparatus which concerns on 4th Embodiment. It is a block diagram which shows the whole structure of the object detection apparatus which concerns on 5th Embodiment. It is a flowchart which shows the process sequence of the duplication removal process by the object detection apparatus which concerns on 5th Embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Object detection apparatus, 10, 11 ... Camera, 12 ... Distance measuring device, 13 ... Object detection ECU, 20 ... Candidate area | region calculation part, 30 ... Detection part, 40 ... Duplication removal part, 41 ... Number-of-person estimation part, 42 ... Classification processing unit, 43 ... representative value creation unit, 50 ... detection result output unit.

Claims (3)

An imaging means for capturing an image;
An image region including a candidate image of the detection target object is extracted from the image captured by the imaging unit, and the coordinate value of the image region and the detection target object candidate included in the image region and the detection target object are extracted. Extraction means for outputting an evaluation value of the image area according to the similarity,
Sorting means for classifying the image area for each detection target object based on the coordinate values of the image area output from the extraction means;
Object position calculating means for calculating the position of the detection target object by taking a weighted average according to the evaluation value for the coordinate value of the image region for each detection target object classified by the classification means; An object detection apparatus comprising: An imaging means for capturing an image;
An image region including a candidate image of the detection target object is extracted from the image captured by the imaging unit, and the coordinate value of the image region and the detection target object candidate included in the image region and the detection target object are extracted. Extraction means for outputting an evaluation value of the image area according to the similarity,
Sorting means for classifying the image area for each detection target object based on the coordinate values of the image area output from the extraction means;
Object position calculation means for calculating the position of the detection target object from the coordinate value of the image area having the maximum evaluation value among the image areas classified by the classification means for each detection target object. An object detection device characterized by. When a plurality of the image areas are extracted, further comprising an object number determination means for determining whether or not the number of the detection target objects is a plurality based on a horizontal deviation of the plurality of extracted image areas,
The said classification | category means classify | categorizes the said several image area | region for every said detection target object, when it is judged by the said object number determination means that the number of the said detection target objects is plural. Or the object detection apparatus of 2.

Images