JP2006078261A - Object detector - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device capable of efficiently detecting an object such as obstacles by reducing a processing load. <P>SOLUTION: From a photographed image, characteristic points are detected (step S102) and the characteristic point data groups from the past to the present are stored in a memory (step S104). The characteristic points of which the characteristic quantity is over a specific value in the present characteristic point data groups are marked as detection object candidate points (step S105), the past characteristic point data groups of the detection object candidate points are extracted and tracking processing is performed for the detection object candidate points (steps S106, S107). Based on the image position from the past to the present and the history of optical flow of the detection object candidate points obtained as the results of the tracking processing, obstacles are detected (step S109 to S113). <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an apparatus for detecting an object, and more particularly to an object detection apparatus that is mounted on a vehicle and is suitable for detecting surrounding obstacles.

  A method of detecting a front obstacle by extracting and tracking a moving object (moving body) from an image captured by a camera and determining whether the object is an obstacle by the movement is known. (For example, Patent Document 1).

JP 2001-91246 A

  In the obstacle detection method disclosed in Patent Document 1, all moving objects extracted from an image are tracked and their movements are examined to determine whether or not the obstacle is an obstacle. However, since objects that are not obstacles are actually extracted as moving objects, it is necessary to perform tracking processing for moving objects that are not obstacles, and there is a problem that the processing load is large. Therefore, there is a need for an obstacle detection method that can reduce a processing load and efficiently detect a target object such as an obstacle.

  An object detection apparatus according to the present invention is a data storage / holding unit that stores and holds time-series data groups from the past to the present, and a current data group stored in the data storage / holding unit for detecting a target object. Past valid data extraction means for extracting valid current valid data, and past valid data corresponding to the current valid data extracted by the valid data extraction means from the past data group stored in the data storage holding means It comprises effective data extraction means, and object detection means for detecting a target object based on current valid data and past valid data.

  According to the present invention, a time-series data group from the past to the present is stored and held, current valid data effective for detecting the target object is extracted from the current data group, and the current valid data is supported. The past valid data to be extracted is extracted from the past data group. The target object is detected based on the current valid data and the past valid data thus extracted. As a result, data representing objects other than the target object can be prevented from being extracted as valid data, and the processing load is reduced by performing tracking processing of the target object based on the extracted valid data. The target object can be detected efficiently.

  FIG. 1 shows the configuration of an obstacle detection apparatus according to an embodiment of the present invention. This obstacle detection device is used by being mounted on a vehicle, detects an obstacle ahead of the host vehicle based on a captured image of a camera, and issues a warning to the driver. The obstacle detection apparatus includes a camera 1, an image processing IC 2, a frame memory 3, a microcomputer 4, and an alarm device 5.

  The camera 1 is installed at the front of the host vehicle so as to capture the front of the host vehicle, and outputs the acquired captured image data to the image processing IC 2 at a predetermined frame rate for each image frame. As the camera 1, for example, a CCD camera or the like is used.

  The image processing IC 2 extracts the vertical edge (vertical edge) and the horizontal edge (horizontal edge) in the captured image based on the captured image data from the camera 1, and determines the intersection of the vertical edge and the horizontal edge. Ask. For this edge extraction, a well-known image filter called a Sobel filter is used. After the edge intersection is extracted from the captured image, an edge intersection having a luminance value larger than a predetermined luminance value determined in advance by the microcomputer 4 from the edge intersection (hereinafter, such edge intersection is referred to as a feature point). ) Is detected. Then, the position of each detected feature point on the image (image position) is specified, and an optical flow indicating the moving speed of each feature point in the image is calculated.

  With the above processing, the image position data indicating the image position of any feature point is collected for each feature point (image position data group), and the size of the optical flow of any feature point is The collected optical flow data for each feature point (optical flow data group) is output from the image processing IC 2 to the frame memory 3 for each frame rate. Hereinafter, these data groups output from the image processing IC 2 are collectively referred to as feature point data groups.

  The frame memory 3 stores and holds the feature point data group output from the image processing IC 2 for each image frame. The frame memory 3 stores and holds feature point data groups for a predetermined number of image frames determined according to the memory capacity. That is, the frame memory 3 stores from the feature point data group based on the captured image acquired a predetermined time ago to the feature point data group based on the latest (current) captured image in units of image frames. In the following description, the feature point data group based on the current captured image is referred to as the current feature point data group, and the feature point data group based on the captured image at the previous time point is referred to as the past feature point data group.

  The microcomputer 4 reads the feature point data group stored in the frame memory 3 and detects an obstacle based on the feature point data group by the method described below. When an obstacle is detected in the microcomputer 4, a signal for operating the alarm device 5 is output from the microcomputer 4, and an alarm sound is output in the alarm device 5. In this way, the presence of an obstacle is notified to the driver of the host vehicle.

  Processing contents when the obstacle is detected in the microcomputer 4 will be described. In the microcomputer 4, based on the current feature point data group read from the frame memory 3, an amount (hereinafter referred to as a feature amount) indicating a degree of obstacle-likeness to be detected from each feature point is determined in advance. Feature points that are greater than or equal to a predetermined value are extracted. The feature points thus extracted are referred to as detection target candidate points. As will be described later, for example, the size of the optical flow or the amount of change in the edge size is used as the feature amount. Then, data indicating the current image position of the detection target candidate point and the size of the optical flow is extracted from the current feature point data group stored in the frame memory 3. In this way, current valid data effective for detecting an obstacle is extracted.

  Further, data indicating the past image position of the detection target candidate point and the size of the optical flow is extracted from the past feature point data group stored in the frame memory 3. In this manner, past valid data effective for detecting an obstacle is extracted. Based on the current and past valid data extracted in this way, the detection target candidate point is traced back from the current captured image to the past captured image, so that the image position of the detection target candidate point from the past to the present Optical flow history is required.

  The detection target candidate points are tracked as follows. First, the image position and optical flow of each feature point one frame before are obtained from the past feature point data group acquired in the previous image frame. The image position and optical flow one frame before the detection target candidate point are obtained by selecting the closest detection target candidate point to the current image position and optical flow. Next, a detection target candidate first obtained from the image position and optical flow two frames before each feature point obtained from the past feature point data group acquired in the image frame two frames before the present. By selecting an image position closest to the image position and the optical flow one frame before the point, the image position and the optical flow two frames before the detection target candidate point are obtained.

  By performing the above-described processing for a predetermined number of frames for each image frame based on the past feature point data group, the feature point data group at a point closer to the present is traced back to the past and stored in the frame memory 3. The image position and the optical flow of the detection target candidate point are respectively obtained from the feature point data group at each past time point. In this manner, the image position and optical flow history of the detection target candidate points from the past to the present are obtained.

  In addition, after performing processing on a plurality of image frames in the tracking processing described above, it is possible to statistically process changes in the image position and the optical flow of the detection target candidate points between the plurality of frames, so that the past arbitrary It is possible to predict the image position and the size of the optical flow at the time. Therefore, the image position and the optical flow of the detection target candidate point may be obtained using such a predicted value.

  Furthermore, when the predicted value as described above is used, or when the optical flow is small and the change in the image position of the detection target candidate point is small between image frames, the image frame to be tracked is thinned out. Also good. That is, when obtaining the image position of the detection target candidate point and the history of the optical flow, instead of obtaining one image frame as described above, a plurality of image frames may be obtained. At this time, among the feature point data groups stored in the frame memory 3 in the past, the past feature point data groups that have not been thinned out are traced back to the past in order from the feature point data group at the closest time point, The image position of the detection target candidate point and the optical flow are respectively obtained.

  As described above, for each of the detection target candidate points extracted based on the current feature point data group, by performing tracking processing based on the past feature point data group, from the past of each detection target candidate point The image position and optical flow history up to now can be obtained. Then, using the image position of each detection target candidate point and the optical flow history, the detection target candidate points are clustered with those representing the same object, and those representing the obstacle are detected from the cluster. In this way, the obstacle detection process is performed.

  As described above, the present apparatus extracts detection target candidate points based on the current feature point data group, and traces the detection target candidate points from the current captured image to the past captured image. In this way, it is possible to extract the detection target candidate point by capturing the obstacle at a position closer to the own vehicle than in the case of tracking in the captured image after extracting the detection target candidate point as usual. it can. As a result, it is possible to set a higher threshold for the feature amount when extracting the detection target candidate points. With respect to this point, an example in which the magnitude of the optical flow is used as a feature value will be described with reference to FIG. 2, and an example in which the change amount of the edge size is used as a feature value will be described with reference to FIG. Hereinafter, description will be given with reference to FIG.

FIG. 2 shows the positional relationship between the imaging surface of the camera 1 and the obstacle A when extracting the detection target candidate points using the size of the optical flow as the feature amount, and (a) shows the position at time T1. The relationship (b) shows the positional relationship at time T2. Note that (a) and (b) show the positional relationship when the camera 1 and the obstacle A are viewed from above. (A) as in the vertical position the distance to the obstacle A from the image plane L _A, the distance transverse position W _A from the visual axis direction of the camera 1 to the obstacle _A, the focal length of the camera 1 f respectively In terms of representation, the lateral image position X _A1 of the feature point representing the obstacle A in the captured image at time T1 is geometrically represented by the following equation (1).
_{X A1 = W A · f /} (L A -f) (1)

Further, if the distance traveled by the host vehicle between time T1 and time T2 is dL, and using this distance, the distance from the imaging surface to the obstacle _A is expressed as the vertical position L _A -dL as shown in (b). The horizontal image position _XA2 of the feature point representing the obstacle A in the captured image at time T2 is represented by the following equation (2).
_{X A2 = W A · f /} (L A -dL-f) (2)

From the above equations (1) and (2), the movement amount of the image position of the feature point between time T1 and time T2, that is, the optical flow Vimg of the feature point at time T2 is expressed by the following equation (3). Can do.
Vimg = X _A2 -X _A1
= W _A · f · (dL / L _A )
_{/ [(1-dL / L} A -f / L A) · (1-f / L A)] (3)

Since the focal length f is sufficiently small compared to the vertical position L _A, Equation (3) can be approximated as the following equation (4).
Vimg≈W _A · f · (dL / L _A ) / (1-dL / L _A ) (4)

The threshold value of the optical flow for extracting the detection target candidate point from the captured image at time T2 is set such that the feature point representing the obstacle A is extracted as the detection target candidate point. According to Expression (4), (dL / L _A ) is larger, that is, the value of the optical flow Vimg of the feature point increases as the distance between the obstacle A and the vehicle amount gets closer, and (dL / L _A ) becomes more It can be seen that the value of the optical flow Vimg of the feature point decreases as the distance is smaller, that is, the distance between the obstacle A and the vehicle amount increases. Accordingly, since the optical flow threshold value can be set higher as the distance from the host vehicle to the obstacle A is shorter, the feature value threshold value when extracting the detection target candidate point can be set higher. As a result, it becomes difficult to erroneously extract feature points representing a slow object such as a background as detection target candidate points, so that it is possible to reduce the load of tracking processing performed on each extracted detection target candidate point.

Next, it demonstrates using FIG. FIG. 3 shows the positional relationship between the imaging surface of the camera 1 and the obstacle B in the case where the detection target candidate point is extracted using the amount of change in the edge size as a feature amount, as in FIG. Represents the positional relationship at time T1, and (b) represents the positional relationship at time T2. As in (a), the distance from the imaging surface to the obstacle B is the vertical position L _B , the distance from the visual axis direction of the camera 1 to the both end points of the obstacle B is the horizontal positions W _B1 and W _B2 , When the focal length is represented by f, the horizontal image positions X _B1 and X _B2 of the two feature points respectively representing the two end points of the obstacle B in the captured image at time T1 are expressed by the following equation (5). expressed.
X _B1 = W _B1 · f / (L _B −f)
X _B2 = W _B2 · f / (L _B −f) (5)

The equation (5), the width U ₁ between the feature points in the captured image at the time T1 is expressed by the following equation (6).
U ₁ = X _B2 −X _B1
= (W _B2 −W _B1 ) · f / (L _B −f) (6)

Further, if the distance traveled by the host vehicle between time T1 and time T2 is dL, and using this, the distance from the imaging surface to the obstacle _B is expressed as the vertical position L _B -dL as shown in (b). The horizontal image positions X _B3 and X _B4 of the two feature points respectively representing the end points of the obstacle B in the captured image at time T2 are expressed by the following equation (7).
X _B3 = W _B1 · f / (L _B −dL−f)
_{X B4 = W B2 · f /} (L B -dL-f) (7)

By equation (7), the width U ₂ between the feature points in the captured image at time T2 is represented by the following equation (8).
U ₂ = X _B3 -X _B4
= (W _B2 −W _B1 ) · f / (L _B −dL−f) (8)

From the above equations (6) and (8), the change amount dU of the width between feature points between time T1 and time T2 can be expressed by the following equation (9). Note that the change amount dU represents the change amount of the edge size in the captured image.
dU = U ₂ −U ₁
= (W _B2 -W _B1 ) · f · (dL / L _B )
_{/ [(1-dL / L} B -f / L B) · (1-f / L B)] (9)

As for formula (3), since the focal length f sufficiently small compared to the vertical position L _B, formula (9) can be approximated as the following equation (10).
_{dU ≒ (W B2 -W B1)} · f · (dL / L B) / (1-dL / L B) (10)

  Two feature points representing the obstacle B are extracted as detection target candidate points as the threshold value of the change amount of the edge size for extracting the detection target candidate points from the captured image at time T2. A correct value is set. As with the equation (4), according to the equation (10), as the distance between the obstacle B and the own vehicle amount gets closer, the value of the edge size change amount, that is, the width change amount dU between the feature points increases. It can be seen that as the distance between the obstacle B and the vehicle amount increases, the value of the change amount dU of the width between the feature points decreases. Therefore, even when the amount of change in the edge size is used as the feature amount, the threshold value for extracting the detection target candidate point can be set higher as in the case where the size of the optical flow is used as the feature amount. .

  FIG. 4 shows a flowchart of processing executed when an obstacle is detected as described above. This flowchart is executed for each image frame of the captured image for each frame rate. In step S100, image data captured by the camera 1 is input to the image processing IC2. In step S101, the image processing IC 2 extracts vertical edges and horizontal edges from the captured image data input from the camera 1 in step S100, and obtains an edge intersection.

  In step S102, the image processing IC 2 detects a feature point from the edge intersection obtained in step S101. At this time, as described above, an edge intersection having a luminance value larger than a predetermined luminance value determined in advance is detected as a feature point. In step S103, the optical flow of each feature point detected in step S102 is calculated. The calculated optical flow size and image position (x, y coordinate values) of each feature point are output from the image processing IC 2 to the frame memory 3 as the aforementioned feature point data group. In step S104, the feature point data group output from the image processing IC 2 is stored and held in the frame memory 3. By repeating this process for each image frame, a feature point data group for a predetermined number of image frames is stored and held in the frame memory 3.

  In step S105, the microcomputer 4 reads the current feature point data group from the frame memory 3, and based on the read current feature point data group, the feature amount of the current captured image is equal to or greater than a predetermined value. It is determined whether there is a point, that is, a detection target candidate point. If there is a detection target candidate point whose feature value is equal to or greater than a predetermined value among the feature points whose feature point data group is stored in the frame memory 3, the detection target candidate point is selected from the feature points. Extract and proceed to step S106. On the other hand, when it determines with there being no detection target candidate point in step S105, the processing flow of FIG. 4 is complete | finished. In this case, detection target candidate points are not extracted, and no obstacle is detected.

  In steps S106 and S107, the tracking process as described above is performed for each detection target candidate point extracted in step S105. In step S106, a past feature point data group is read from the frame memory 3 for one image frame. In step S107, each detection target candidate point extracted in step S105 is tracked based on the past feature point data group read in step S106, and the image position and optical flow in the past captured image are obtained. By repeatedly performing such tracking processing for a predetermined number of frames, as described above, one image frame or a plurality of image frames when the image frame to be tracked is thinned out from the current captured image. The detection target candidate points are traced back to the captured image. As a result, the image position and optical flow history of the detection target candidate points from the past to the present are obtained.

  In step S108, it is determined whether or not the detection target candidate point tracking process in steps S106 and S107 has been executed a predetermined number of repetitions or more. The detection target candidate point tracking process is performed for the same number of frames as the number of repetitions. If the number of executions of the tracking process is equal to or greater than the predetermined number of repetitions, the process proceeds to step S109. If the number of repetitions is less than the number of repetitions, the process returns to step S106 and the above tracking process is repeated.

  In step S109, the history of the optical flow of each detection target candidate point obtained by the tracking processing in steps S106 and S107 is statistically processed, whereby the image speed of each detection target candidate point between the past and the present is determined. calculate. The image speed here is obtained by removing an optical flow component generated by a factor other than the movement of an obstacle from the history of the optical flow. This includes, for example, a pitching fluctuation component superimposed by the pitching behavior of the host vehicle. Statistical processing methods for removing such optical flow components include, for example, pitching frequency component removal processing for removing pitching fluctuation components and smoothing processing for removing noise components. Detailed description is omitted.

  In step S110, clustering of the detection target candidate points is performed based on the image position of each detection target candidate point and the image speed of each detection target candidate point calculated in step S109. At this time, detection target candidate points whose image speed difference is equal to or smaller than a predetermined value and whose image position is within a predetermined range are clustered as one cluster representing the same object. Each cluster generated by clustering the detection target candidate points for each object is hereinafter referred to as a detection target candidate cluster.

  In step S111, for each detection target candidate cluster clustered in step S110, a distance in real space from the own vehicle to the object represented by the detection target candidate cluster is calculated. This distance is obtained by obtaining an angle with respect to the visual axis direction of the camera 1 based on the image position of the detection target candidate cluster, and using the known height of the camera 1 with respect to the ground and the angle in the visual axis direction, a triangulation method. Can be calculated. Note that the image positions of the detection target candidate clusters are obtained by averaging the image positions of the detection target candidate points clustered as representing the same object. In step S112, the size in real space of the object represented by each detection target candidate cluster is calculated based on the image size of each detection target candidate cluster and the distance in real space calculated in step S111.

  In step S113, it is determined whether or not the object represented by each detection target candidate cluster is an obstacle. The object represented by the detection target candidate cluster when the size in the real space calculated in step S112 is in a predetermined size range and the image speed is in a predetermined speed range. Is determined to be an obstacle. If any detection target candidate cluster satisfies this determination condition, the process proceeds to step S114, where the alarm device 5 is turned on and an alarm sound is output in step S114. After executing step S114, the flowchart of FIG. 4 is terminated. On the other hand, if there is no detection target candidate cluster that satisfies the above determination condition, step S114 is not executed and the flowchart of FIG. As described above, the obstacle is detected in the present apparatus.

According to the embodiment described above, the following operational effects can be obtained.
(1) A time-series data group from the past to the present is stored and held, current valid data effective for detecting an obstacle as a target object is extracted from the current data group, and the current valid data is extracted. Corresponding past valid data is extracted from the past data group. An obstacle is detected based on the current valid data and the past valid data thus extracted. As a result, data representing objects other than the obstacle can be prevented from being extracted as valid data, and the processing load is reduced by performing the obstacle tracking process based on the extracted valid data. Obstacles can be detected efficiently.

  Specifically, a feature point data group related to feature points extracted from the captured image is output from the image processing IC 2 to the frame memory 3 at predetermined intervals, and the feature point data group from the past to the present is stored in the frame memory 3. Store and hold (step S104). In the microcomputer 4, a feature point having a feature amount equal to or larger than a predetermined value in the latest feature point data group is extracted as a detection target candidate point (step S105), and the current image position and optical of the detection target candidate point are extracted. Data indicating the flow is extracted from the current feature point data group. Thereafter, the detection target candidate point is tracked by extracting data indicating the past image position and optical flow of the detection target candidate point from the past feature point data group (steps S106 and S107). An obstacle is detected based on the past and present image positions of the detection target candidate points obtained as a result of the tracking process and the optical flow history (steps S109 to S113).

(2) Feature point data groups at a plurality of past times are stored and held in the frame memory 3 (step S104), and feature points at each past time point are traced back in order from the feature point data group at a time point closer to the present time. Data indicating the image position of the detection target candidate point and the optical flow are extracted from the data group (steps S106 and S107). Since it did in this way, the past effective data effective in order to detect an obstacle from the data of each past time can be extracted reliably.

(3) In the processes of steps S106 and S107, feature point data at a plurality of past points stored in the frame memory 3 are thinned out, and past feature point data groups that have not been thinned out are past points closer to the present time. The feature point data group can be traced back to the past in order. In this way, it is possible to further reduce the processing load by reducing the number of tracking processing targets.

(4) As the feature point data group, the image position data group based on the image position data of each feature point extracted by the image processing IC 2 and the optical flow data group based on the optical flow data of each feature point are transferred from the image processing IC 2 to the frame memory 3. The data is output and stored in the frame memory 3. Since it did in this way, only the data required for the tracking process can be stored and held as a feature point data group, and the memory capacity can be saved.

(5) Using the amount of change in the size of the edge intersection obtained based on the image position of each feature point and the optical flow size of each feature point as a feature amount, feature points that are greater than or equal to a predetermined value are detected. Extract as candidate points. Then, the image position of the extracted detection target candidate point and the size of the optical flow are extracted from the feature point data group composed of the current image position data group and the optical flow data group. Since it did in this way, the present and past effective data can be easily extracted from the feature point data group.

  In the above-described embodiment, feature points are extracted from the captured image of the camera 1, and a feature point data group indicating the image position and optical flow of each feature point is stored and held for each image frame. An example in which an obstacle is detected using a data group acquired and stored in time series has been described. However, the present invention is not limited to this content, and can be applied to the case where a target object is detected using various data groups acquired and stored in time series. For example, the method of the present invention may be applied when a data group of distances and directions to a target object is acquired in time series by a laser radar and the target object is detected using the data group.

  Further, in the above-described embodiment, feature points whose feature amounts such as optical flow size and edge size change amount are greater than or equal to a predetermined value are extracted as detection target candidate points. It may be within a predetermined range. That is, feature points whose feature amounts are included in a predetermined range, that is, feature points whose feature amounts satisfy a predetermined condition can be extracted as detection target candidate points.

  In the above-described embodiment, an example has been described in which the camera 1 is used to image the front of the host vehicle and an obstacle existing in front of the host vehicle is detected based on the captured image. However, the present invention is not limited to this content, and when other directions around the host vehicle, for example, the front or side are imaged, and an obstacle existing in that direction is detected based on the captured image. Is also applicable.

  In the above embodiment, the obstacle detection device that is mounted on a vehicle and detects an obstacle around the vehicle has been described as an example. However, the present invention is not limited to this. The present invention is not necessarily installed in a vehicle, and the present invention can be applied to an object detection device that detects various objects.

  In the above embodiment, the imaging means is realized by the camera 1, the feature point extraction means is realized by the image processing IC 2, the data storage holding means is realized by the frame memory 3, and the other means are realized by the processing of the microcomputer 4. This is merely an example, and each component is not limited to the above embodiment as long as the characteristics of the present invention are not impaired.

It is a figure which shows the structure of the obstruction detection apparatus by one Embodiment of this invention. It is a figure which shows the positional relationship of a camera imaging surface and an obstruction in the case of extracting a detection target candidate point using the magnitude | size of an optical flow as a feature-value. It is a figure which shows the positional relationship of a camera imaging surface and an obstruction in the case of extracting a detection target candidate point using the variation | change_quantity of the magnitude | size of an edge as a feature-value. It is a flowchart of the process performed when an obstruction is detected.

Explanation of symbols

1: Camera 2: Image processing IC
3: Frame memory 4: Microcomputer 5: Alarm

Claims (8)

Data storage and holding means for storing and holding time-series data groups from the past to the present;
Current effective data extraction means for extracting current effective data effective for detecting a target object from the current data group stored in the data storage holding means;
Past valid data extraction means for extracting past valid data corresponding to current valid data extracted by the valid data extraction means from past data groups stored in the data storage holding means;
An object detection apparatus comprising: object detection means for detecting the target object based on the current valid data and past valid data. The object detection device according to claim 1.
The data storage and holding means stores and holds a data group of a plurality of past points in time,
The past valid data extracting means extracts each of the past valid data from the data group at a plurality of past time points sequentially from the data group at a time point closer to the present time. The object detection apparatus according to claim 2, wherein
A data thinning means for thinning out a plurality of past data points stored in the data storage holding means;
The past valid data extracting unit traces a past data group that has not been thinned out by the data thinning unit in order from a data group at a time point closer to the present. In the object detection apparatus in any one of Claims 1-3,
Imaging means for acquiring a captured image around the host vehicle;
A feature point extracting unit that extracts a plurality of feature points having a predetermined feature from the captured image acquired by the imaging unit and outputs a data group (feature point data group) related to each feature point at a predetermined period; Prepared,
The data storage / holding means stores and holds the feature point data group output by the feature point extracting means at predetermined intervals in a time series from the past to the present,
The current valid data extracting unit satisfies a predetermined condition in which a feature amount representing the degree of the target object is determined as the current valid data from the current feature point data group stored in the data storage holding unit. Extract data about feature points (candidate points for detection)
The past valid data extracting unit extracts data relating to the detection target candidate point as the past valid data from the past feature point data group stored in the data storage holding unit. The object detection apparatus according to claim 4.
The feature point extracting means extracts a plurality of edge intersections as the feature points from the captured image, and a data group (image position data group) indicating the positions of the edge intersections on the image as a predetermined feature point data group. Output every cycle,
The object storage device stores and holds the image position data group in time series from the past to the present. The object detection apparatus according to claim 5.
The present effective data extraction means uses, as the detection target candidate points, feature points that are included in a predetermined range in which a change amount of the edge size obtained based on the image position data group is the detection target An object detection apparatus that extracts an image position of a candidate point from a current image position data group. The object detection apparatus according to claim 4.
The feature point extraction means outputs a data group (optical flow data group) indicating the magnitude of the optical flow representing the moving speed of the feature points in the captured image as the feature point data group at predetermined intervals. ,
The data storage and holding means stores and holds the optical flow data group in a time series from the past to the present. The object detection device according to claim 7.
The present effective data extraction means uses a feature point included in a predetermined range in which the size of the optical flow is determined in advance as the detection target candidate point, and determines the optical flow size of the detection target candidate point as the current optical data An object detection apparatus characterized by extracting from a flow data group.

Images