JP6570219B2 - Object detection device - Google Patents

Description

  The present invention relates to an object detection apparatus that extracts a feature amount from an image by image processing and detects an object on the image based on the extracted feature amount.

  2. Description of the Related Art Conventionally, for example, a technique for detecting a person or an object by extracting the edge direction and strength as a feature amount in a local region of an image, such as using an HOG feature amount (see Patent Document 1). ing.

  For example, when human detection is performed using the HOG feature value, the vertical edge component is important because human vertical features include many vertical edge components. However, for images of structures such as pillars that are shown together as images in addition to human images, there are many vertical edges in the same way, and when extracting components in local areas of the image, human body components There was a very high possibility that the components of structures such as pillars would be mistaken.

JP 2014-81826 A

  The present invention has been made in view of the above points, and in image detection of a target object (including a human body), confusion between information related to a target object extracted from an image and information related to an object other than the target is made. It is an object of the present invention to provide an object detection apparatus that can suppress and detect a certain target object as an image.

An object detection apparatus for achieving the above object removes a linear component detection unit that detects a linear component from an image to be detected and a linear component detected by the linear component detection unit based on an image having a time difference. A linear component removing unit that determines a power linear component, a feature amount calculating unit that calculates a feature amount based on a linear component that remains without being removed by the linear component removing unit, and a calculation result in the feature amount calculating unit A detection processing unit that performs detection processing.

  In the object detection device, by determining a linear component to be removed based on images having a time difference, the linear component for a target object such as a human body and the linear component for other objects are removed. Can be removed. In other words, information related to the target object and information related to the object other than the target are separated, and a high-precision result can be expected in the calculation by the feature amount calculation unit, and the target object can be detected more reliably.

  In a specific aspect of the present invention, a long-time exposure image creation unit that creates a long-time exposure image as an image having a time difference is provided. In this case, if the target object moves within a certain time, such as a human body, this information can be distinguished from the information of the non-moving object.

  In another specific aspect of the present invention, the linear component removal unit extracts a linear component whose position does not change between a plurality of images having a time difference as an object to be removed. In this case, as a linear component for an object other than a target object such as a human body, a linear component of an object that does not change its position, that is, an object that does not move (stationary object) can be regarded as a removal target.

  In yet another aspect of the present invention, the angle information is extracted from the straight line component detected by the straight line component detection unit, and the straight line component to be processed in the straight line component removal unit is selected based on the angle information. An information selection unit is provided. In this case, as a pre-stage of component removal in the linear component removal unit, components that cannot be determined as component removal can be removed in advance, and the amount of component removal processed in the linear component removal unit can be suppressed.

  In yet another aspect of the present invention, the angle information selection unit extracts a straight line component having a vertical component that satisfies a threshold set for an angle as a straight line component to be processed in the straight line component removal unit. In this case, for example, when it is important to extract a straight line component in the vertical direction as in the detection of a human body image, only the straight line component in the vertical direction is extracted in advance, and the component removal process is performed by the straight line component removing unit. Can be done.

  In still another aspect of the present invention, the linear component detector is a line segment detector that performs line segment detection. In this case, an appropriate linear component can be extracted.

  In still another aspect of the present invention, the feature amount calculation unit calculates a HOG feature amount as the feature amount. In this case, by calculating the HOG feature value in a state in which the information on the target object is left and the information on the object other than the target can be removed, it is possible to expect the calculation of the HOG feature value with high accuracy, and the target object can be detected more reliably. Can be anything.

It is a block diagram which illustrates notionally an example of the structure of the object detection apparatus of this embodiment. (A) is a figure which shows notionally that a straight line component is detected from an image of a human body, and (B) is a figure that conceptually shows a state of detecting a straight line component from an image other than the human body as a comparative example. is there. It is a block diagram which illustrates notionally the structure of a memory | storage device among object detection apparatuses. (A) is a figure which shows an example of the imaged image, (B) is a figure which shows a mode that the linear component was extracted from the image of (A). (A) is a figure which shows notionally extracting image gradient data about a linear component in order to calculate a HOG feature-value as a feature-value by an object detection apparatus, (B) is shown to (A). It is a figure which shows the gradient direction histogram about the extracted image gradient data. It is a flowchart for demonstrating the process of a human body image detection using the object detection apparatus of this embodiment. (A) And (B) is a figure for demonstrating notionally the selection process of the linear component used as the object of the production process of a line segment long exposure image. It is a figure which shows an example of the produced line segment long exposure image. 7 is a flowchart for explaining an example of a line segment long exposure image creation process in FIG. 6. 7 is a flowchart for explaining an example of a HOG feature amount calculation process in FIG. 6.

  As shown in FIG. 1, the object detection device 100 according to the present invention includes an information processing unit 80 and an imaging device 90. The information processing unit 80 includes a CPU 10, an image information receiving unit 20, a storage device 30, a display device 40, an input device 50, a bus 70, and the like.

  The object detection device 100 is a device that operates the CPU 10 of the information processing unit 80 and performs image detection on a target object from an image captured by an imaging device 90 installed as a fixed camera on the road, for example. . In the present embodiment, a case where a human body is captured as a target object will be described. That is, the object detection apparatus 100 according to the present embodiment is an apparatus that can determine whether or not a human body image exists in an image to be processed, and thus is an apparatus that performs human body detection. If various objects other than a human body are detected, the object detection device 100 functions as an object detection device.

  Among the information processing units 80 constituting the object detection device 100, the CPU 10 exchanges data with the image information receiving unit 20, the storage device 30, the display device 40, and the input device 50 via the bus 70. It is possible.

  The imaging device 90 can include a solid-state imaging device such as a CMOS or CCD, and an image detected by the solid-state imaging device is output to the information processing unit 80 as a digital image signal, for example. Here, as described above, for example, a person or the like that is fixedly installed on the road and passes through the road is set as an imaging target (see, for example, FIG. 4A).

  The information processing unit 80 acquires the image information transmitted from the imaging device 90 and, from the acquired captured image, for example, as shown in FIG. A line segment L1 is extracted from the captured image, and it is determined from the characteristics of each linear component L1 whether or not a human body image exists as an image portion of the captured image.

  Hereinafter, referring back to FIG. 1, each unit configuring the information processing unit 80 will be specifically described. The CPU 10 performs various controls related to image processing. In particular, the CPU 10 stores the image information captured from the imaging device 90 by the image information receiving unit 20 in the storage device 30 and performs various image processing based on the image information stored in the storage device 30. That is, it is a control unit that performs control for various processes. In particular, in the present embodiment, the information processing unit 80 extracts a linear component from the image information acquired by the imaging device 90, finds a feature amount (HOG feature amount) from the extracted linear component, and relates to the feature amount. It is determined whether or not a human body image exists by analyzing a learning sample by a learning algorithm. As a detection method, for example, human detection using a method called HOG feature amount and Boosting is known.

  The image information receiving unit 20 can communicate with the imaging device 90 by wire or wirelessly, and receives image information acquired by the imaging device 90 according to the control of the CPU 10.

  The storage device 30 includes a program storage unit having a program area for storing various programs for operating the object detection apparatus 100, and a data storage unit having a data area for temporarily storing input instructions, input data, processing results, and the like. Is provided. That is, the storage device 30 includes various programs such as a program for extracting a linear component from image information and a program for calculating a feature amount from the linear component, and stores various information. In particular, in the present embodiment, the storage device 30 includes a program for removing data that should be excluded when calculating the feature value. The feature amount includes a program for calculating a HOG feature amount, which is a feature amount obtained by histogramating the gradient direction in the local region. Details of the storage unit and the like of the storage device 30 will be described later with reference to the block diagram shown in FIG.

  The display device 40 includes a display drive circuit that generates a drive signal based on data input from the CPU 10, an image display unit that performs necessary display based on the drive signal input from the display drive circuit, and the like. Necessary display is performed based on the command signal from.

  The input device 50 is composed of a keyboard or the like, and outputs a command signal reflecting the intention of an operator who operates the object detection device 100 to the CPU 10. That is, the CPU 10 can read a predetermined program and data from the storage device 30 based on an operator instruction via the input device 50 and execute various processes based on these programs and data.

  Here, when detecting whether or not the image is a human body image as described above, it is possible to calculate the feature amount with respect to the linear component in the vertical direction among the linear components extending in various directions in the secondary image. It is generally known that it is important. For example, as shown in FIG. 2A, if only the component relating to the human body can be extracted as the component in the vertical direction from the linear component L1 of the image P1 (the image P1 is actually a human body image), the feature amount for that component is calculated. By doing so, it can be estimated whether the image P1 is a human body image. However, in actual processing, there is a linear component LL1 that is not a human body image even if the image includes a linear component in the vertical direction, such as an image P2 shown in FIG. In the example of FIG. 2B, a vertical linear component LL1 is extracted from the columnar image P2. Even if the feature amount is calculated in a state where the data of the linear component LL1 and the data of the linear component L1 of the human body image are mixed for such a non-human body image, it is determined whether the image is a human body image. It will not always be accurate. In the present embodiment, in order to avoid such a situation, image information other than the image information of the target object can be removed in advance in the image processing, so that the feature amount can be calculated with high accuracy, and surely. Target object can be detected.

  Hereinafter, the configuration of the storage device 30 that stores various programs for performing object detection processing by image processing and stores various data will be described with reference to FIG. 3.

  FIG. 3 is a block diagram for conceptually explaining the configuration of the storage device 30 in the information processing unit 80 configuring the object detection device 100. As illustrated, the storage device 30 includes a program storage unit PM having a program area for storing various programs and a data storage unit DM having a data area for storing various data.

  Among the components of the storage device 30, the program storage unit PM includes an image capture program CP for capturing a captured image and a line for extracting a linear component (line segment) from the captured image that is raw image data. A segment program LS, a line segment long exposure image creation program LE for creating a line segment long exposure image (hereinafter also simply referred to as a long exposure image) based on the linear components extracted by the line segment program LS, Features of an extracted line segment angle setting program AS for extracting a vertical component to be drawn in a long-time exposure image out of straight line components (line segments), and a straight line component selected based on the long-time exposure image HOG feature quantity calculation program FC for calculating the HOG feature quantity as a quantity, and the calculated HO And a human body image detection program GD for detecting the presence of a human body image based on the feature quantity.

  The HOG feature quantity calculation program FC includes image gradient data SC that is necessary for calculating the HOG feature quantity and image gradient data that is data to be excluded in the calculation, in addition to the gradient direction histogram calculation program HC. An image gradient data deletion program DP to be deleted is included.

  In addition to the above, the program storage unit PM includes various programs necessary for image processing and the like. For example, it is assumed that various programs such as a program for cutting out a partial image of a captured image and a normalization processing program necessary for calculating the HOG feature value are included.

  Among the components of the storage device 30, the data storage unit DM includes a captured image data storage unit CD that stores captured image data captured by the image capture program CP, and a linear component (line) extracted by the line segment program LS. Line segment long-exposure image data that stores data of a long-exposure image created based on a line segment long-exposure image creation program LE It has a storage unit BD, a HOG feature amount data storage unit FD that stores various data calculated by the HOG feature amount calculation program FC, and a detection data storage unit DD that stores detection results in the human body image detection program GD.

  Among the above, the HOG feature amount data storage unit FD has been selected based on the image gradient data storage unit SD that stores image gradient data as data for calculating the HOG feature amount and the image gradient data deletion program DP. It includes an image gradient data selection result storage unit CR for storing results, and a gradient direction histogram data storage unit HD for storing gradient direction histogram data indicating HOG feature values.

  In the program storage unit PM, the image capture program CP captures a captured image PA as shown in FIG. 4A, for example, and the image data is stored in the captured image data storage unit CD.

  The line segment program LS is a program for extracting, for example, a line segment (straight line component) as seen in the image PB shown in FIG. 4B from the captured image PA in FIG. Data on (straight line component) is stored in the straight line component (line segment) data storage unit LD. In other words, the CPU 10 functions as a line segment detector that is a linear component detection unit that detects a line segment by reading the line segment program LS.

  The line segment long-exposure image creation program LE is an image including line segments (straight line components) stored in the data storage unit LD, and is a long-exposure image (by overlapping a plurality of images having a time difference with each other). This is a program for creating (see FIG. 8). That is, the CPU 10 reads the line segment long exposure image creation program LE and functions as a long exposure image creation unit. The created long exposure image is stored in the line segment long exposure image data storage unit BD. Details of the long-exposure image creation process will be described later with reference to FIG.

ここで、Ｌ（ｘ，ｙ）は、各画素での輝度を示す。また、図５（Ｂ）は、上記勾配強度ｍ（ｘ，ｙ）を予め定められた勾配方向にヒストグラム化したものを示す一例である。本実施形態では以上のようなＨＯＧ特徴量を算出するために、ＨＯＧ特徴量算出プログラムＦＣが画像勾配算出プログラムＳＣ及び勾配方向ヒストグラム算出プログラムＨＣを有するものとなっている。この上で、さらに、ＨＯＧ特徴量算出プログラムＦＣは、算出において除外すべきデータとなる画像勾配データを削除する画像勾配データ削除プログラムＤＰを含んでいる。具体的には、画像勾配データ削除プログラムＤＰは、ＨＯＧ特徴量の算出において、ラインセグメントの長時間露光画像でデータのある部分を削除する処理を行っている。これにより、以外の物体についての直線成分以外の情報についての直線成分に関するデータを除去することになる。つまり、ＣＰＵ１０は、記憶装置３０から画像勾配データ削除プログラムを読み出して、除去すべき直線成分を除去することで、直線成分除去部として機能する。 For general calculation of the HOG feature amount, for example, the explanation in detail in the cited reference 1 and the like, and the explanation in the known literature etc. regarding the HOG feature amount will be repeated. For example, the image gradient calculation program SC in the HOG feature amount calculation program FC calculates image gradient data necessary for calculating the HOG feature amount, and the rectangular region shown as an example in FIG. For the divided block image G1, calculation processing such as calculating the gradient intensity m (x, y) and gradient direction θ (x, y) of the luminance is performed for each pixel PX constituting each block image G1. The definitions of the gradient strength m (x, y) and the gradient direction θ (x, y) are as follows.

Here, L (x, y) indicates the luminance at each pixel. FIG. 5B is an example showing a histogram of the gradient strength m (x, y) in a predetermined gradient direction. In this embodiment, in order to calculate the HOG feature value as described above, the HOG feature value calculation program FC includes an image gradient calculation program SC and a gradient direction histogram calculation program HC. In addition, the HOG feature amount calculation program FC further includes an image gradient data deletion program DP that deletes image gradient data as data to be excluded in the calculation. Specifically, the image gradient data deletion program DP performs a process of deleting a portion of data in the long exposure image of the line segment in the calculation of the HOG feature amount. As a result, data relating to the straight line component for information other than the straight line component for other objects is removed. That is, the CPU 10 functions as a linear component removal unit by reading the image gradient data deletion program from the storage device 30 and removing the linear component to be removed.

  Hereinafter, a human body image detection process using the object detection apparatus 100 according to the present embodiment will be described with reference to the flowchart of FIG. 6 and the like. First, the CPU 10 acquires image information from the imaging device 90 via the image information receiving unit 20, reads the image capture program CP from the storage device 30 as appropriate, and captures the captured image, as shown in FIG. An input image to be exemplified is acquired (step S1). Next, the CPU 10 reads the line segment program LS from the storage device 30 and extracts image data including a line segment (straight line component) as seen in the image PB illustrated in FIG. 4B (step S2). ). Next, the CPU 10 reads the line segment long exposure image creation program LE from the storage device 30, and creates a long exposure image of the line segment based on the image data extracted in step S2 (step S3). Details of the image processing in step S3 will be described later with reference to FIG. 7 and the like. In step S3, data that enables separation of information related to the target object and information related to the object other than the target is acquired. Next, the CPU 10 reads the HOG feature value calculation program FC from the storage device 30 based on the straight line component (line segment) included in the long-exposure image of the line segment, and performs a process of calculating the HOG feature value (step). S4). In step S4, unnecessary data that becomes noise can be deleted using the result of step S3. Next, the CPU 10 determines whether or not a human body image exists in the captured image data to be detected based on the HOG feature calculated in step S4 (step S5). Next, the CPU 10 confirms whether or not there is an instruction to end detection of the human body image (step S6). If not confirmed (step S6: No), the operations of steps S1 to S6 are repeated. If there is an instruction to end detection (step S6: Yes), the process ends.

  Hereinafter, with reference to the flowchart of FIG. 9 in addition to FIGS. 7 and 8, an example of the line segment long-time exposure image creation process (the process of step S3) among the various processes shown in FIG. 6 will be described. Here, the long-exposure image is obtained by superimposing (adding luminance) image data for several seconds obtained by superimposing images with the time difference on each other, for example, by capturing the same part with a fixed camera. Means an image to be played. Here, as an example, it is assumed that image data including the straight line component (line segment) extracted in step S2 is superimposed on a plurality of captured images (data corresponding to several seconds of imaging time) captured at the same location. In this case, for example, with respect to the linear component drawn in the superimposed image, an object that does not move over time (stationary object) is more clearly revealed (see FIG. 8).

  Further, for the straight line component (line segment), the horizontal component is deleted in advance, and only the vertical component is extracted. Specifically, as shown in FIGS. 7A and 7B, whether the component is a vertical component or a horizontal component is determined based on the angle α set as the threshold value. Here, as an example, the vertical direction is 0 °, and ± 20 ° is the boundary. That is, as shown in FIG. 7A, the vertical component is a linear component LE1 within ± 20 ° in the vertical direction, that is, in an angle range from 0 ° to 20 ° or from 160 ° to less than 180 °. As shown in FIG. 7B, a linear component LE2 in the other angle range is set as a horizontal component. In the present embodiment, as described above, the purpose is to detect a human body image, and in detecting a human body image, the vertical component is important. Therefore, the vertical direction as shown in FIG. 7A is determined. Only the linear component LE1 to be processed is handled, and the linear component LE2 determined to be in the horizontal direction as shown in FIG. 7B is excluded. For this reason, the linear component acquired as a result has only the linear component of a vertical direction so that it may illustrate in FIG.

  A program for determining whether the component is a vertical component or a horizontal component as described above is stored in the storage device 30 as the extracted line segment angle setting program AS. In other words, the CPU 10 functions as an angle information selection unit that reads the extracted line segment angle setting program AS, extracts angle information about the straight line component (line segment), and selects the straight line component based on the angle information. It has become. In particular, in the present embodiment, a straight line component having a vertical component that satisfies the threshold set for the angle α is extracted as a straight line component to be processed in the HOG feature value in step S4, which is the next stage process, and the horizontal direction component is extracted. The linear component is previously removed in the step S3.

  Hereinafter, an example of a line segment long exposure image creation process will be described in detail with reference to the flowchart of FIG. Here, as an example, in step S2, a line segment (straight line component) in the image is drawn on the image, and the drawn images are overlapped to create a long exposure image as shown in FIG. To do.

  First, in step S2, when a plurality of captured images (data corresponding to several seconds of imaging time) taken at the same or substantially the same location and with a time difference are extracted, the CPU 10 extracts a line segment long-time exposure image creation program. LE is read out and various processes for creating a long exposure image are performed. Specifically, first, it is confirmed whether or not the line drawing process has been completed for all of these captured images (step S101). If not completed (step S101: No), the image ( That is, an undrawn image) is appropriately selected (step S102), and information extracted as a line segment (straight line component) in the image in step S2 is read (step S103). Of the finite number of line segments (straight line components) read out in step S103, the first processing target is appropriately selected (step S104), and all line segments (straight line components) are determined to be drawn. It is confirmed whether the processing image to be completed has been completed (step S105). If it is determined in step S105 that the processing has not ended (step S105: No), the CPU 10 reads the extracted line segment angle setting program AS from the storage device 30, and the inclination of the line segment (linear component) to be processed. It is determined whether the angle is within a certain range, that is, whether the angle is in the vertical direction or the horizontal direction (step S106). If it is determined in step S106 that the angle is within a certain range (step S106: Yes), the CPU 10 stores the line segment (straight line component) to be processed as a drawing target (step S107), and the angle is within the certain range. If it is determined that it is not within (step S106: No), the CPU 10 excludes the line segment (straight line component) to be processed from the drawing target (step S108), and the line segment (straight line component) to be processed next. Is selected (step S109). CPU10 repeats the operation | movement from the said step S106 to step S109 until it completes a process about all the line segments (straight line component) (step S105: YES). CPU10 will draw all the line segments memorize | stored in step S107, if the process about all the line segments (straight line component) of the finite lake contained in the image of a process target is complete | finished (step S110). After the process of step S110, it is confirmed whether there is a captured image to be processed next (step S101), and until it is determined in step S101 that the line drawing process has been completed for all captured images (step S101: Yes), the operations from step S101 to step S110 are repeated. When the CPU 10 determines that the line drawing process has been completed for all the captured images (step S101: Yes), the CPU 10 reads the image overlay program included in the line segment long exposure image creation program LE and reads the line. The respective drawn images are superimposed (step S111). By the process of step S111, the line segment long exposure image illustrated in FIG. 8 is created, and the process of step S4 ends. The straight line component (line segment) drawn in the line segment long exposure image acquired by the above processing is not particularly changed in position in the image, that is, out of the straight line component of an object (stationary object) that does not move with time. The component in the vertical direction is extracted. Therefore, when the detection target is a moving (moving) human body such as the human body as in the case of the present embodiment, the information about the linear component extracted as described above is used as the next stage. By removing at the time of calculating the feature amount, which is a process, it is possible to reduce unnecessary information that becomes noise, so that a highly accurate result can be expected, and the target object can be detected more reliably.

  Hereinafter, an example of the HOG feature amount calculation process (the process of step S4) will be described with reference to the flowchart of FIG. First, the CPU 10 reads an image gradient calculation program from the HOG feature amount calculation program FC of the storage device 30 and locally blocks the input image as described with reference to FIG. From the image, a calculation process related to the luminance gradient is performed (step S201). Next, the CPU 10 reads the image gradient data deletion program DP and deletes the data of the linear component (line segment) for the line segment long-time exposure image acquired in step S4 described above (step S202). The CPU 10 reads the gradient direction histogram calculation program HC and creates a gradient direction histogram as described with reference to FIG. 5B and the like (step S203).

  In the human detection process in step S5 shown in FIG. 6, the CPU 10 reads the human body image detection program GD and, based on the gradient direction histogram created in step S203 as described above, the person who is the detection target object. Human detection by image processing is performed as a detection processing unit that performs detection processing of (human body image).

  As described above, in the object detection apparatus 100 according to the present embodiment, the CPU 10 determines a linear component to be removed based on an image having a time difference created based on the line segment long exposure image creation program LE. The CPU 10 discriminates the straight line component for the target object such as a human body and the straight line component for the other object, and the CPU 10 reads the image gradient data deletion program DP as a straight line component removal unit, and the straight line to be removed. Remove ingredients. That is, the information about the target object and the information about the object other than the target are separated. Thereby, in the calculation of the HOG feature value by the HOG feature value calculation program FC which is the feature value calculation unit, a highly accurate result can be expected, and the detection of the target object can be made more reliable.

  The present invention is not limited to the above embodiments or examples, and can be implemented in various modes without departing from the scope of the invention.

  In the above description, the object to be detected by the object detection device 100 is a person (human body image). However, the present invention is not limited to this, and many moving (moving) objects can be detected. For example, a vehicle can be a detection target. In addition, when a person is a detection target, it is important to extract a vertical linear component as a HOG feature, so that only a vertical component is extracted and then those other than those derived from the human body are removed. However, when detecting the vehicle, the linear component in the horizontal direction may be extracted for analysis of the feature amount.

  In the above description, for example, the description in the flowchart of FIG. 9 shows an example in which a line segment (straight line component) in the image is drawn on the image, but the processing procedure is limited to the case of FIG. Various methods can be selected. Furthermore, as a purpose, it is only necessary to acquire data related to a line segment (straight line component) to be finally removed. If this can be acquired, a method of acquiring only data is not limited to a drawing. Good.

  Further, although the imaging device 90 is a fixedly installed camera, for example, a camera that moves somewhat may be used as long as linear component data that can distinguish a stationary object from a moving object can be acquired.

  Although the exposure time is about several seconds, other image data having a time difference may be used as long as the target linear component data can be acquired.

  Furthermore, in the area that is the target of the imaging range of the imaging device 90, the image data of the area that is determined to be unnecessary and can be deleted in advance is deleted in advance even if a linear component or the like is extracted. It is good.

  In the above description, the image gradient data that is to be excluded by the image gradient data deletion program DP is deleted. However, the timing for deleting the data to be deleted is at various timings for the convenience of image processing. Do what you do.

  In the above description, the HOG feature amount is analyzed and analyzed as a feature amount detection method. However, a feature amount other than the HOG feature amount may be used.

  The configurations, shapes, sizes, and arrangement relationships described in the above embodiments are merely schematically shown to the extent that the present invention can be understood and implemented. Therefore, the present invention is not limited to the described embodiments, and can be variously modified without departing from the scope of the technical idea shown in the claims.

  DESCRIPTION OF SYMBOLS 20 ... Image information reception part 30 ... Memory | storage device 40 ... Display apparatus 50 ... Input device 70 ... Bus | bath 80 ... Information processing part 90 ... Imaging device 100 ... Object detection apparatus AS ... Extraction line segment angle setting Program: BD: Line segment long exposure image data storage unit, CD: Captured image data storage unit, CP: Image capture program, CR: Image gradient data selection result storage unit, DD: Detection data storage unit, DM: Data storage unit DP ... image gradient data deletion program, FC ... feature amount calculation program, FD ... feature amount data storage unit, GD ... human body image detection program, HC ... gradient direction histogram calculation program, HD ... gradient direction histogram data storage unit, LD ... Data storage unit, LE ... Line segment long exposure image creation program, LS ... Line segment program, PM ... Program storage unit, SC ... Image gradient calculation program, SD ... Image gradient data storage unit, L1, LE1, LE2, LL1 ... Linear component (line segment), G1 ... Block image, P1, P2 ... Image, PA ... Image, PB ... Image, PX ... Pixel, α ... Angle

Claims (6)

A linear component detector for detecting a linear component from the detection target image;
And a linear component removing unit that determines as a linear component to be remove those which do not change in position of the linear components that are sorted based on the extracted angle information from the line component detected by the linear component detection section,
A feature amount calculation unit that calculates a feature amount based on a linear component that remains without being removed in the linear component removal unit;
An object detection apparatus comprising: a detection processing unit that performs an object detection process from a calculation result of the feature amount calculation unit.   The object detection apparatus according to claim 1, wherein the linear component removal unit determines a linear component to be removed based on an image having a time difference.   The object detection apparatus according to claim 2, further comprising a long-time exposure image creation unit that creates a long-time exposure image as the image having a time difference. An angle information selection unit for selecting a linear component to be processed in the linear component removal unit based on the angle information extracted for the linear component detected by the linear component detection unit;
The angle information sorting unit, a line component having a longitudinal component which satisfies the set threshold for the angle, is extracted as a linear component to be processed in said linear component removal unit, one of the Claims 1 to 3 one The object detection device according to item. The linear component detection section is a line segment detector that performs line detection, object detection apparatus according to any one of claims 1 to 4. The object detection device according to any one of claims 1 to 5 , wherein the feature amount calculation unit calculates a HOG feature amount as the feature amount.

Images