JP2006209334A - Position detection apparatus, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a position detection apparatus capable of detecting the two-dimensional position of an object of varying height such as a human through a simple detection process using one photographing means. <P>SOLUTION: A videocamera 11 takes a photo of a person to obtain a two-dimensional image. A projecting part 12 extracts a person region from the two-dimensional image obtained and projects the person region onto a plurality of horizontal planes within a three-dimensional space. An integrating part 13 calculates an integrated value by integrating the mappings of the person projected onto the horizontal planes. A detection part 14 detects a horizontal position within the three-dimensional space where the peak of the integrated value calculated is situated as the position of the person, and detects as the height of the person the height of the highest horizontal plane where the mapping of the person exists at the horizontal position within the three-dimensional space where the peak is situated. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a position detection apparatus and a position detection method for detecting a three-dimensional position of an object, for example, a human, using a single photographing unit.

  In recent years, image monitoring systems have been installed in the city due to advances in image acquisition technology and networking. Such image surveillance systems will be used for various important purposes in the near future, monitoring people with safety problems, providing useful information to pedestrians, especially the elderly and children, etc. The quality of life can be improved by analyzing the data provided by these services.

The image monitoring system as described above needs to detect and track the position of a person. For example, it has been reported that a three-dimensional position of a person is tracked using stereo technology using a plurality of cameras. (See Non-Patent Document 1). When a person is tracked using this stereo technology, the three-dimensional position of each person can be accurately detected even when the person overlaps.
A. Mittal et al., “M2T Tracker: A Multi-View Approach to Segmenting and Tracking People in a Cluttered Scene”, Computer Vision International Journal (International Journal of Computer Vision), 2003, Vol. 51 (3), p. 189-p. 203

  However, when a plurality of cameras are used as described above, it is necessary to process a large number of image data, which increases the calculation cost. In addition, a plurality of cameras are required for one shooting range. When shooting a wide range, a large number of cameras are required, which increases the cost of the apparatus. For this reason, it is conceivable to detect the position of a human by projecting an image photographed by one camera onto a virtual plane arranged at a predetermined height in the three-dimensional space. Therefore, it is difficult to detect an accurate two-dimensional position when the height of the measurement object does not match the assumed height of the plane. On the other hand, since the height of the human subject to be measured by the present invention is generally different for each individual, the influence of the problem that the measurable height is fixed is serious.

  An object of the present invention is to provide a position detection apparatus, a position detection method, and position detection that can detect two-dimensional positions of objects having various heights such as human beings by a simple detection process using a single photographing means. Is to provide a program.

  The position detection device according to the present invention includes a photographing unit that captures a target and acquires a two-dimensional image including the target, and extracts a target region from the two-dimensional image acquired by the photographing unit. Projecting means for projecting the object region onto a plurality of horizontal planes set in advance in the three-dimensional space at predetermined intervals, and integrating the mapping of the target object projected onto each horizontal plane by the projecting means, the integral value of the mapping is obtained. An integrating means for calculating, and a detecting means for detecting the horizontal position in the three-dimensional space where the peak of the integrated value calculated by the integrating means is located as the position of the object.

  In the position detection apparatus according to the present invention, a two-dimensional image including the target object is acquired by shooting the target object with one imaging unit, and the target object region extracted from the acquired two-dimensional image is a three-dimensional space. Are projected onto a plurality of preset horizontal planes at predetermined intervals, and the integral of the projection is calculated by integrating the projection of the object projected onto each horizontal plane, and the peak of the calculated integral value is located A horizontal position in the three-dimensional space is detected as the position of the object.

  In this way, assuming that the object is upright, the position of the peak of the integrated value is detected as the horizontal position of the object, and the height of the highest horizontal plane where the mapping of the object exists at this peak position is detected. Since the height is detected as the height of the object, it is possible to detect the two-dimensional position of an object having various heights such as a human by using only an image photographed by one photographing means. Moreover, since only the image photographed by one photographing means is used, the detection process can be simplified.

  Preferably, the detecting means detects a horizontal position in the three-dimensional space where the convolution value of the approximate filter approximating the three-dimensional shape of the object and the integral value is the maximum as the position of the object.

  In this case, only the target object can be extracted by the approximate filter, and robust detection can be performed against disturbance caused by other objects.

  Preferably, the detection means detects the height of the highest horizontal plane where the mapping of the object exists at a horizontal position in the three-dimensional space where the peak is located as the height of the object.

  In this case, since the height of the highest horizontal plane where the mapping of the object exists at the horizontal position in the three-dimensional space where the peak is located can be detected as the height of the object, the position of the top of the three-dimensional object It is possible to detect (a two-dimensional coordinate where the three-dimensional object exists on the map and its height) as the three-dimensional position of the object.

  When a plurality of peaks are present in the integration value calculated by the integration unit, the detection unit is positioned for each peak in which the interval between the height of the valley between adjacent peaks and the peak height is equal to or greater than a predetermined interval. The horizontal position in the three-dimensional space is detected as the position of the object, and the height of the highest horizontal plane where the mapping of the object exists at the horizontal position in the three-dimensional space where the peak is located is defined as the height of the object. It is preferable to detect.

  In this case, since it can be determined that each peak having an interval between the height of the valley between adjacent peaks and the height of the peak is a predetermined interval or more is due to the object, a plurality of objects are close to each other However, the three-dimensional position of each object can be detected with high accuracy.

  When the interval between the height of the valley between the adjacent peaks and the peak height is less than the predetermined interval, the detection unit detects the first object and the second object that is separated from the first object from the imaging unit. It is determined that the object overlaps, and the horizontal position in the three-dimensional space where the peak equal to or higher than the reference height is detected as the position of the first object, and the height of the first object is determined for a predetermined time. It is preferred to use the previously detected height.

  In this case, even if the first and second objects are very close to each other and overlap on the photographed image, the three-dimensional position of the first object located on the photographing means side can be detected with high accuracy. .

  The detection means is detected a predetermined time ago as a straight line connecting the photographing center position of the photographing means in the three-dimensional space and the three-dimensional position of the peak higher than the reference height in the three-dimensional space, and the height of the second object. The horizontal position in the three-dimensional space where the point where the horizontal plane located at the height intersects is located is detected as the position of the second object, and the height of the second object is detected a predetermined time ago. It is preferable to use the height.

  In this case, even if the first and second objects are very close to each other and overlap each other on the photographed image, the three-dimensional position of the second object that is farther from the first object from the photographing means is highly accurate. Can be detected.

  The position detection method according to the present invention includes a first step of capturing an object using one image capturing unit to acquire a two-dimensional image including the object, and an object region from the acquired two-dimensional image. A second step of extracting and projecting the object region onto a plurality of preset horizontal planes at predetermined intervals in a three-dimensional space, and integrating the mapping of the target object projected onto each horizontal plane to A third step of calculating an integral value, and detecting a horizontal position in the three-dimensional space where the peak of the calculated integral value is located as the position of the object, and at the horizontal position in the three-dimensional space where the peak is located Detecting the height of the highest horizontal plane where the mapping of the object exists as the height of the object.

  The position detection program according to the present invention extracts a target area from a two-dimensional image acquired by one imaging unit that captures a target and acquires a two-dimensional image including the target. Projection means for projecting onto a plurality of horizontal planes set in advance in the three-dimensional space at predetermined intervals, and integration means for calculating the integral value of the map by integrating the mapping of the object projected onto each horizontal plane by the projection means The computer is caused to function as detection means for detecting the horizontal position in the three-dimensional space where the peak of the integration value calculated by the integration means is located as the position of the object.

  According to the present invention, since the position of the peak of the integrated value is detected as the horizontal position of the object on the assumption that the object is upright, only the image photographed by one photographing means is used. It is possible to detect a two-dimensional position of an object having various heights, such as a human being, and to simplify the detection process.

  Hereinafter, a position detection device according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a position detection apparatus according to an embodiment of the present invention.

  The position detection apparatus shown in FIG. 1 includes a single video camera 11, a projection unit 12, an integration unit 13, a detection unit 14, and a tracking unit 15. The projection unit 12, the integration unit 13, the detection unit 14, and the tracking unit 15 are an input device, a ROM (read only memory), a CPU (central processing unit), a RAM (random access memory), and an image I / F (interface) unit. In addition, a CPU or the like executes a position detection program for performing each process described later using a computer including an external storage device or the like. Note that the configuration example of the projection unit 12, the integration unit 13, the detection unit 14, and the tracking unit 15 is not particularly limited to this example, and each block may be configured by dedicated hardware, or may be included in some blocks or blocks. Various modifications such as configuring only a part of processing with dedicated hardware are possible.

  The video camera 11 is composed of a normal surveillance camera or the like, and is attached to a predetermined position in the photographing space, for example, a corner of the ceiling. The video camera 11 is calibrated by a calibration method described later, captures a human being as a target, acquires a two-dimensional image including the human, and outputs the two-dimensional image to the projection unit 12. As the video camera 11, an existing surveillance camera may be used.

  The projecting unit 12 extracts a human area as a target area from the acquired two-dimensional image, and extracts the extracted human area in a three-dimensional space at a plurality of horizontal planes (the height of each horizontal plane is preset). Project onto (known). A known image processing method such as background difference can be used to separate the background area (non-human area) and the person area (human silhouette) in the two-dimensional image.

  Here, a calibration method of the video camera 11 will be described. In order to project a person area image onto a plurality of horizontal planes set in the three-dimensional space, 2 between the captured image of the video camera 11 and two reference planes (horizontal planes) set in the three-dimensional space. A dimensional projective transformation matrix is required. First, it is assumed that one of the two reference planes is positioned at the height of the ground (height 0) and the other plane is positioned at the height Yh. Four calibration bars that are taller than the height are installed perpendicularly to the ground at predetermined intervals (so that the four calibration bars form a cube). Colored markers are provided on the head and bottom of each calibration bar, and two-dimensional projective transformation matrices H0 and H1 are calculated by observing four markers on each reference plane. Here, assuming that the height of a horizontal plane located between two reference planes is Yn, the two-dimensional projective transformation matrix Hn of this horizontal plane is two-dimensional as shown in the following equation (1). It can be estimated by interpolation using the projective transformation matrices H0 and H1.

Hn = ((Yh−Yn) H0 + (Yn) H1) / Yh (1)
As described above, by accurately measuring the three-dimensional position of the calibration bar at each installation position, conversion between a plurality of horizontal planes and the three-dimensional world coordinate system can be easily performed. Can be simplified.

  The integration unit 13 integrates the human map projected on each horizontal plane, calculates an integral value of the map, and outputs it to the detection unit 14. The detection unit 14 detects the horizontal position in the three-dimensional space where the calculated peak of the integrated value is located as a human position, and a mapping of the person region exists at the horizontal position in the three-dimensional space where the peak is located. The height of the highest horizontal plane to be detected is detected as the height of the person. The tracking unit 15 tracks a person using the detected three-dimensional position, and stores the three-dimensional position (horizontal position and height) of the person at each observation time.

  When the video camera 11 is configured by a movable camera that can move in the pan direction and the tilt direction, the tracking unit 15 may control the shooting position of the video camera 11 according to the tracking result. In this case, by performing camera calibration at the shooting position, the object can always be shot at the optimum observation position, and the three-dimensional position can be detected with high accuracy.

  Specifically, the detection unit 14 detects the horizontal position in the three-dimensional space where the convolution value between the approximate filter that approximates the three-dimensional shape of the target object and the integral value is the maximum as the position of the target object. The human three-dimensional position is detected by detecting the height of the highest horizontal plane where the human mapping exists at the position of the object as the human height.

  In addition, the detection unit 14 captures images of a plurality of adjacent humans, and when there are a plurality of peaks in the integrated value, the interval between the valley height between adjacent peaks and the peak height is equal to or greater than a predetermined interval, for example, 1 m For each peak, the horizontal position in the three-dimensional space where the peak is located is detected as a human position, and the highest horizontal plane where the human mapping exists at the horizontal position in the three-dimensional space where the peak is located. The height is detected as the height of the object.

  Further, the detection unit 14 is configured so that two or more humans are overlapped when a plurality of humans are very close to each other and overlap each other on the captured image, and the interval between the valley height between adjacent peaks and the peak height is less than a predetermined interval. And the horizontal position in the three-dimensional space where the peak of the reference height or more, for example, 2.5 m or more is located, is detected as the position of the person closest to the video camera 11, and The height detected for a predetermined time before, for example, one or several frames before, is read from the tracking unit 15 as the height, and the read height is used as the current height.

  At this time, the detection unit 14 reads, from the tracking unit 15, the height detected as a height of a person far from the video camera 11 for a predetermined time, for example, one or several frames before, and the read height is currently As well as a straight line connecting the camera center position of the video camera 11 in the three-dimensional space and the three-dimensional position of the peak higher than the reference height in the three-dimensional space, and the height read from the tracking unit 15. The horizontal position in the three-dimensional space where the intersection with the horizontal plane is located is detected as the human position.

  As described above, in the present embodiment, not a single horizontal plane but a plurality of horizontal planes are set in the three-dimensional space, and only a monocular image photographed by the calibrated video camera 11 is used. When estimating a three-dimensional position using only this monocular image, that is, two-dimensional information, the number of dimensions is insufficient by one dimension. To compensate for this, it is assumed that a human is standing upright on a plurality of horizontal planes. A human region, which is a human silhouette, is projected onto the image, and the three-dimensional position (horizontal position and height) of the human is detected using a peak of an integral value obtained by integrating the projection region along the vertical axis. Humans of various heights can be tracked.

  In the present embodiment, the video camera 11 corresponds to an example of a photographing unit, the projection unit 12 corresponds to an example of a projection unit, the integration unit 13 corresponds to an example of an integration unit, and the detection unit 14 includes a detection unit. It corresponds to an example.

  Next, normal three-dimensional position detection processing for detecting the three-dimensional position of a person using an image obtained by photographing one person will be described. FIG. 2 is a principle diagram for explaining normal three-dimensional position detection processing by the position detection apparatus shown in FIG.

  As shown in FIG. 2, a plurality of horizontal planes HP are set in a three-dimensional space, and the two-dimensional coordinates of each horizontal plane HP are defined by the X axis and the Y axis, the height direction is the Y axis, A certain human OB is shown in a bar shape in the drawing for easy explanation. First, the projection unit 12 changes the height Yn (0 ≦ Yn ≦ Yh) to project the person area S divided from the two-dimensional image captured by the video camera 11 onto the horizontal plane n (each horizontal plane HP). To do. This projection result P (Yn) can be expressed by the following formula (2) using the height Yn.

P (Yn) = HnS (2)
Here, assuming that the human OB is standing upright, the projected human region PR (shaded portion in the figure) always includes a certain point (X, Z) on the horizontal plane n where the actual human exists. The three-dimensional position of the person can be estimated by integrating (merging) all the horizontal planes n.

  The integration unit 13 calculates an integral value Itg (X, Z) of the person region P (Y) that is a mapping of the object along the vertical axis Y at the point (X, Z) using the following equation (3). To do.

  Further, in order to detect the peak of the integral value, a three-dimensional convex filter that is an approximation filter that approximates a human shape is used. For example, a cylindrical filter Cvx (X, Z) represented by the following formula (4) Where r is the radius of the bottom of the cylinder. The approximate filter is not particularly limited to this example, and various filters can be used according to the three-dimensional shape of the object. Also, the method for detecting the peak of the integral value is not particularly limited to the following method, and various methods can be used. For example, the peak is detected by differentiating the integral value represented by the above formula (3). A method (a method using a differential filter) may be used.

  The detection unit 14 calculates a convolution value between the convex filter Cvx (X, Z) and the integral value Itg (X, Z), and obtains a coordinate value (Xm, Zm) that maximizes the convolution value as a human horizontal position. And the maximum height of the person area at the horizontal position (Xm, Zm) of the person (the height of the highest horizontal plane where the human mapping exists) is detected as the human height Ym. In the example shown in FIG. 2, the convolution value CV is calculated, the coordinate value (Xm, Zm) where the peak PI is located is detected as the human horizontal position, and the coordinate value (Xm, Zm) where the peak PI is located. The maximum height of the person area is detected as the human height Ym.

  During calculation of the above integration processing, division errors such as small holes and cracks in the image are smoothed, and it is not necessary to completely divide each region, and only the human is extracted by the convex filter Cvx (X, Z). It is possible to perform robust detection against disturbances caused by other objects.

  Next, a plurality of three-dimensional position detection processing for detecting the three-dimensional position of each person using images obtained by photographing a plurality of persons will be described. First, when a plurality of persons are photographed in an input image and each person area can be divided, as a first plurality of three-dimensional position detection processing, the projection unit 12 converts the two-dimensional image into divided areas including only one person area. The integration unit 13 and the detection unit 14 detect the three-dimensional positions of all humans by repeating the normal three-dimensional position detection process for one person for each divided region.

  However, for example, when a plurality of photographed people are holding hands, a part of the two person areas are connected and the two person areas are weakly connected, so the area can be divided as described above. Can not. In this case, as the second plurality of three-dimensional position detection processes, the above-described normal three-dimensional position detection process is executed with a connected region in which a part of a plurality of person regions is connected as one region.

  That is, similarly to the above, the projecting unit 12 projects a connected region in which a part of a plurality of person regions is connected as a single region onto a plurality of horizontal planes, and the integrating unit 13 integrates an integrated value Itg (Xn of the projected region. , Zn), and the detection unit 14 calculates a convolution value of the convex filter Cvx (Xn, Zn) and the integral value Itg (Xn, Zn).

  Next, the detection unit 14 compares the interval between the height of the valley located between the peaks of adjacent convolution values and the height of the peak of the convolution value with a predetermined interval stored in advance, and a peak that is equal to or greater than the predetermined interval. Is detected. Next, the detection unit 14 compares the detected peak of the convolution value with a predetermined value stored in advance, for example, 1.5 m, and extracts a peak greater than or equal to the predetermined value. Next, the detection unit 14 detects the coordinate point (Xn, Zn) where the peak is located for each extracted peak as the horizontal position of one person, and the maximum height of the person region at the horizontal position (Xn, Zn). Is detected as a human height Yn. The above processing is applied to the entire three-dimensional space, and a plurality of human three-dimensional positions are sequentially detected. The height of the peak and the height of the valley are detected in the same manner as the detection of the three-dimensional position of a human, and the coordinate point where the peak and the valley are located is detected as the horizontal position of the peak and the valley. The maximum height of the person area at the horizontal position can be detected as the peak height.

  With the above processing, it is possible to determine that each peak whose interval between the height of the valley between the peaks of adjacent convolution values and the peak height of the convolution values is equal to or greater than a predetermined interval is caused by humans. Even in the state of being close to each other, the three-dimensional position of each person can be detected with high accuracy.

  Next, when there is another person on the straight line connecting the camera center of the video camera 11 and the person, the person areas of the two humans greatly overlap and the two person areas are strongly connected. Since the interval between the height of the valleys located between the peaks and the height of the peak of the convolution value is less than a predetermined interval, and it is difficult to divide the connected region by the filter processing, the second plurality of The three-dimensional position detection process cannot be applied. In this case, a third plurality of three-dimensional position detection processes for detecting overlapping human positions using the positions of the humans already detected several frames before are executed. If there are two or more people in the same projection area, the following processing is repeated by the number of people.

  First, similarly to the above, the projection unit 12 projects a connected region including two people onto a plurality of horizontal planes, and the integration unit 13 calculates an integration value Itg (Xn, Zn) of the projection region. The detection unit 14 calculates a convolution value of the convex filter Cvx (Xn, Zn) and the integral value Itg (Xn, Zn), and maximizes the person area at the maximum coordinate values (X0 (t), Z0 (t)). The height Y0 (t) ′ is detected (where t is a time parameter).

  Next, the detection unit 14 compares the detected height Y0 (t) ′ with a reference value Yr (for example, 2.5 m) stored in advance, and the height Y0 (t) ′ is the reference value Yr. In the above case, it is determined that the person closer to the video camera 11 is in the horizontal position (X0 (t), Z0 (t)), and the same person (the one closer to the video camera 11) detected using the previous frame. Is read from the tracking unit 15 and the read height Y0 (t-1) is used as the current height Y0 (t).

  Through the above processing, the detection unit 14 acquires the horizontal position (X0 (t), Z0 (t)) and the height Y0 (t−1) as the three-dimensional position of the person closer to the video camera 11. Therefore, even if two humans are very close to each other and overlap each other on the captured image, the three-dimensional position of the human who is closer to the video camera 11 can be detected with high accuracy.

  Next, the detection unit 14 reads the height Y1 (t−1) of the same person detected using the previous frame as a person far from the video camera 11 from the tracking unit 15, and reads the height Y1 ( t-1) is used as the current height Y1 (t) of the person far from the video camera 11, and the detected coordinate values (X0 (t), Y0 (t) ′, Z0 (t)) and a straight line connecting the camera center (Xc, Yc, Zc) of the video camera 11 are set, and the coordinate value of the intersection of the horizontal plane located at the read height Y1 (t-1) and this line (X1 (t), Z1 (t)) is detected as the position of a person far from the video camera 11.

  Through the above processing, the detection unit 14 acquires the horizontal position (X1 (t), Z1 (t)) and the height Y1 (t-1) as the three-dimensional position of the person far from the video camera 11. Therefore, even if two people are very close to each other and overlap each other on the captured image, the three-dimensional position of the person far from the video camera 11 can be detected with high accuracy.

  As the above same person prediction process, for example, the tracking unit 15 stores the three-dimensional position of the person detected in each frame, and uses the three-dimensional position to move the person (for example, constant velocity). The current three-dimensional position is predicted by performing speed prediction and / or acceleration prediction with respect to (movement), and the person whose predicted three-dimensional position is closest to the detected current horizontal position is the person at the current horizontal position. It can be performed by determining that they are the same person. In this case, since human 3D position prediction using speed prediction and / or acceleration prediction is directly performed in the 3D space, it is not affected by geometric non-linearity due to projection, and the speed prediction and Acceleration prediction can be performed and / or the same person can be determined with high accuracy.

  Next, the result of actually detecting the three-dimensional position of a human using the above position detection device will be described. FIG. 3 is a schematic diagram for explaining an imaging space, and FIG. 4 is a diagram illustrating an example of a detection result by the position detection device illustrated in FIG.

  As shown in FIG. 3, the video camera 11 is mounted at a position where the height of the center of the camera is 2.6 m, tilted 68 degrees in the horizontal plane with respect to the 2.5 × 11 m shooting space (corridor), and the shooting space is Arranged to look down, a human moving in the shooting space was photographed and 100 images of 320 × 240 pixels were acquired at 1 second intervals. Twenty-two landmarks are arranged in the photographing space, and the three-dimensional positions of all the points are measured using a three-dimensional laser measuring device. Using the three-dimensional position information and the two-dimensional position information of the photographed image, The video camera 11 was calibrated. The spatial resolution of the captured image depends on the distance, and the spatial resolution when the object located at the end of the corridor is captured is about 56 mm, and the captured object is standing in front of the video camera 11 The spatial resolution of was about 17 mm.

  As a result of detecting the three-dimensional position of a human under the above measurement conditions, the result shown in FIG. 4 was obtained. First, (a) in FIG. 4 is a scene where one person is walking, (b) is a scene where one person is bending, and (c) is a scene where two persons pass each other (the person area is weakly combined). (D) and (d) respectively show a scene where two people overlap in a captured image (a state in which person areas are strongly coupled). 4 shows the photographed image, the middle part shows the person area extracted from the photographed image, the lower part shows the result of integrating the mapping onto the horizontal plane of the photographing space, and the upper and lower squares and circles are detected. The position of the recorded peak, that is, the position of a human being is shown. FIG. 4 shows that one or two people can be accurately detected in the above four types of scenes, and can be accurately tracked even if a plurality of people overlap.

  As described above, since the present invention requires only one video camera as an input sensor, it is possible to easily detect the three-dimensional position of a human using a surveillance camera that is already widely installed. Since the detection result is described in a three-dimensional coordinate system, it is suitably used when sharing the same contents as other sensors (for example, GPS, RFID) or navigating the robot via a network. be able to.

  In the above description, a human is described as an example of an object to be detected. However, the present invention can be similarly applied to other animals, other moving objects, and the like, and similar effects can be obtained.

It is a block diagram which shows the structure of the position detection apparatus by one embodiment of this invention. It is a principle figure for demonstrating the normal three-dimensional position detection process by the position detection apparatus shown in FIG. It is a schematic diagram for demonstrating imaging | photography space. It is a figure which shows an example of the detection result by the position detection apparatus shown in FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Video camera 12 Projection part 13 Integration part 14 Detection part 15 Tracking part

Claims (8)

A photographing means for photographing the object and obtaining a two-dimensional image including the object;
Projection means for extracting an object region from the two-dimensional image acquired by the photographing means and projecting the object region onto a plurality of horizontal planes set in advance in the three-dimensional space at predetermined intervals;
Integrating means for integrating the mapping of the object projected on each horizontal plane by the projection means to calculate an integral value of the mapping;
A position detection apparatus comprising: a detection unit configured to detect a horizontal position in the three-dimensional space where the peak of the integration value calculated by the integration unit is located as a position of an object.   The detection means detects a horizontal position in the three-dimensional space where a convolution value of an approximate filter approximating a three-dimensional shape of the object and the integral value is maximum as the position of the object. Item 1. The position detection device according to Item 1.   The detection means detects the height of the highest horizontal plane where a mapping of the object exists at a horizontal position in the three-dimensional space where the peak is located, as the height of the object. 2. The position detection device according to 2.   When there are a plurality of peaks in the integration value calculated by the integration unit, the detection unit detects the peak for each peak having a predetermined gap or more between a valley height and a peak height between adjacent peaks. The horizontal position in the three-dimensional space where the position is located is detected as the position of the object, and the height of the highest horizontal plane where the mapping of the object exists at the horizontal position in the three-dimensional space where the peak is located is the object. The position detection device according to claim 3, wherein the position detection device detects the height of the position detection device.   The detection means has a first object and a first object that is farther from the imaging means than the first object when an interval between a valley height between adjacent peaks and a peak height is less than a predetermined interval. 2, the horizontal position in the three-dimensional space where the peak higher than the reference height is located is detected as the position of the first object, and the height of the first object is determined. 5. The position detecting device according to claim 3, wherein a height detected before a predetermined time is used.   The detection means is predetermined as a height of a second object and a straight line connecting a photographing center position of the photographing means in the three-dimensional space and a three-dimensional position of a peak higher than the reference height in the three-dimensional space. The horizontal position in the three-dimensional space where the point where the horizontal plane located at the height detected before time intersects is detected as the position of the second object, and the height of the second object is predetermined. 6. The position detecting device according to claim 5, wherein a height detected before time is used. A first step of capturing an object using one imaging means to obtain a two-dimensional image including the object;
A second step of extracting an object region from the acquired two-dimensional image and projecting the object region onto a plurality of predetermined horizontal planes at predetermined intervals in the three-dimensional space;
A third step of calculating an integral value of the mapping by integrating the mapping of the object projected on each horizontal plane;
And a step of detecting a horizontal position in the three-dimensional space where the peak of the calculated integral value is located as the position of the object. A target area is extracted from a two-dimensional image acquired by one imaging unit that captures a target and acquires a two-dimensional image including the target, and the target area is previously stored in a three-dimensional space at predetermined intervals. Projection means for projecting onto a plurality of set horizontal planes;
Integrating means for integrating the mapping of the object projected on each horizontal plane by the projection means to calculate an integral value of the mapping;
A position detection program for causing a computer to function as detection means for detecting a horizontal position in the three-dimensional space where the peak of the integration value calculated by the integration means is located as the position of an object.

Images