JP2022097190A - Object recognition device and object recognition method - Google Patents

Abstract

To provide an object recognition device and an object recognition method that can suppress reduction in determination accuracy of a correlation.SOLUTION: An object recognition device D includes: a clustering section 22 that obtains observation data which is object state information including a size and a position of an object by clustering a detection point obtained by detecting the object around its own device in a predetermined detection range; a prediction section 28 that obtains prediction data of the object state information on the basis of past observation data; a correlation section 25 that determines whether the object represented by the observation data and the object represented by the prediction data are identical by correlation processing between the observation data and the prediction data; and a first size correction section 24 that corrects at least one of the size of the object in the observation data and the size of the object in the prediction data so that the size of the object in the observation data and the size of the object in the prediction data match before the correlation processing.SELECTED DRAWING: Figure 1

Description

The present invention relates to an object recognition device and an object recognition method for recognizing an object existing around the own machine.

In recent years, there has been a demand for recognition of the surrounding situation due to demands for safety technology and automatic driving technology. In particular, in the case of collision avoidance or lane change in a moving body such as a vehicle or a robot, it is important to recognize the situation on the side of the moving body. For this reason, the moving body is often equipped with an object recognition device (object detection device) that detects an object in order to recognize the surrounding situation. Techniques for detecting and recognizing such an object are disclosed in, for example, Patent Document 1 and Patent Document 2.

The radar device disclosed in Patent Document 1 derives the observation data of the target including at least one of the data of the distance, the relative speed and the angle from the target based on the reflected wave by the target. The unit, the prediction unit that generates the prediction data that predicts the observation data, the setting unit that sets the prediction range based on the prediction data, and the observation data derived this time by the derivation unit are within the prediction range. In this case, it is provided with a determination unit for determining that the observation data previously derived by the derivation unit and the observation data derived this time are observation data of the same target that are continuous in time. In the detection of such an object, the identity of the object is determined at each detection timing by determining the correlation between the observation data and the prediction data.

The object detection device disclosed in Patent Document 2 is an object detection device mounted on a vehicle and detecting an object around the vehicle, and indicates a capture point obtained by detecting an object around the vehicle. A movement direction calculation means for calculating the movement direction of each of the capture points using a signal, a frame corresponding to the shape of the object to be detected in advance, and a travel reference direction in the frame as the travel direction assumed for the object. Is provided, and among the capture points, a determination means for determining a capture point existing in the frame in which the traveling reference direction is aligned with the movement direction is provided as a capture point of the same object. In the detection of such an object, a plurality of supplementary points in a single object are combined into one by so-called grouping (clustering), but in the object detection device disclosed in Patent Document 2, the object is described. A frame is set according to the shape of the above, and clustering is performed by determining whether or not the frame is within this frame.

Japanese Unexamined Patent Publication No. 2018-063130 JP-A-2010-156567

By the way, a detector (sensor) for detecting (sensing) an object, such as a radar device, detects an object within a predetermined detection range defined in advance. Due to the performance limit of the detector, even if an object exists within the detection range, a detection point may be generated in a part of the object and the detection point in the rest of the object may leak. For example, when the detector is a radar device, this phenomenon is likely to occur near the end of the detection angle. The size (frame size, object size (object size)) of the frame (cluster frame) generated as the detection area of the object by clustering based on the detection point in the part of the object is the part of the object. Since it is the detection point of, it will be different from the size of the object detected at the previous timing. Therefore, when determining the correlation between the object detected at the previous detection timing and the object detected at the current detection timing based on the size, the determination accuracy of the correlation is lowered.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an object recognition device and an object recognition method capable of suppressing a decrease in correlation determination accuracy.

As a result of various studies, the present inventor has found that the above object can be achieved by the following invention. That is, the object recognition device according to one aspect of the present invention has the size of the object and the size of the object by clustering one or a plurality of detection points obtained by detecting the object around the own machine within a predetermined detection range. A clustering unit that obtains observation data that is object state information including position, a prediction unit that obtains prediction data that predicts the object state information based on past observation data, an object represented by the observation data, and the prediction data. A correlation unit that determines whether or not the object represented by is the same by the correlation processing between the observation data and the prediction data, and an object in the observation data before the correlation processing is performed by the correlation unit. It is provided with a size correction unit that corrects at least one of the size of the object in the observation data and the size of the object in the prediction data so that the size of the object and the size of the object in the prediction data match. Preferably, in the above-mentioned object recognition device, the detection point is obtained by a radar device that detects an object in a predetermined detection range based on the reflected wave of the transmitted wave. Preferably, in the above-mentioned object recognition device, the detection point is obtained by an object detection device that detects an object in a predetermined detection range based on a pair of images captured by a stereo camera.

Such an object detection device is such that the size of the object in the observation data and the size of the object in the prediction data match so that the size of the object in the observation data and the size of the object in the prediction data match before performing the correlation processing. Since at least one of the sizes is corrected, it is possible to suppress a decrease in the determination accuracy of the correlation due to the difference in size.

In another aspect, in the above-mentioned object recognition device, the size of the object is the size of a frame surrounding the area where the object exists, and the size correction unit is the closest to the own machine in the frame. Correct the contact positions so that they are the same before and after the correction.

Since such an object recognition device corrects the positions of the recent contacts having high detection accuracy so that they are the same before and after the correction, the position accuracy can be maintained or its deterioration can be suppressed after the correction.

In another aspect, in these above-mentioned object recognition devices, the size of the object is the size of a rectangular frame surrounding the area where the object exists, and the size correction unit has a predetermined first coordinate axis and the first coordinate axis. When the second coordinate axis orthogonal to one coordinate axis and the coordinate system whose coordinate origin is the center position of the own machine are set and the sides of the rectangular frame intersect with the first coordinate axis or the second coordinate axis, the above The ratio of the first length of the first partial side extending to one quadrant to the second length of the second partial side extending to the other quadrant at the closest tangent to the own machine at the intersecting side is before and after the correction. Correct so that they are the same. Preferably, in the above-mentioned object recognition device, the first coordinate axis extends along the viewing direction, which is the center line of the detection range. Preferably, in the above-mentioned object recognition device, the first coordinate axis extends along the traveling direction of the own machine when moving.

Such an object recognition device corrects the ratio of each length at each partial side of each quadrant in the nearest edge, which tends to have high detection accuracy, so that it is the same before and after the correction. The position accuracy can be maintained or its deterioration can be suppressed after the correction.

In another aspect, in these above-mentioned object recognition devices, the size of the object is the size of a rectangular frame surrounding the area where the object exists, and in the size correction unit, the rectangular frame of the prediction data is detected. If it exists beyond the range, it is within the third length of the third partial side extending beyond the detection range and within the detection range at the second closest edge closest to the own machine at the side intersecting the detection range. The size of the object in the observed data is corrected so that the ratio of the extended fourth partial side to the fourth length is the same in the predicted data and the corrected observation data.

If the rectangular frame of the prediction data exists beyond the detection range, the size of the object in the observation data may be missing in the portion outside the detection range. In the object recognition device, the ratio of each length in each partial side of the detection range in the nearest edge of the rectangular frame of the prediction data is the same in the prediction data and the corrected observation data. In addition, since the size of the object in the observation data is corrected, the defect can be compensated, and the position accuracy can be maintained or its deterioration can be suppressed after the correction.

In another aspect, in the above-mentioned object recognition device, one of a plurality of predefined sizes is used based on the size of the object in the observation data obtained by the clustering unit before the correction is performed by the size correction unit. The size of the object in the observation data is selected so that the selected size is the size of the object in the observation data and the position obtained from the selected size is the position of the object in the observation data. And further includes a replacement that replaces the position.

Since such an object recognition device replaces the size selected from a plurality of predefined sizes with the size of the object in the observation data, it is possible to suppress the fluctuation of the size due to the detection omission of the detection point.

The object recognition method according to another aspect of the present invention is the size of the object and the size of the object by clustering one or more detection points obtained by detecting the object around the own machine within a predetermined detection range. A clustering process for obtaining observation data which is object state information including a position, a prediction process for obtaining prediction data for predicting the object state information based on past observation data, and an object represented by the observation data and the prediction data. The observation data before performing the correlation processing in the correlation processing step of determining whether or not the object represented by is the same or not by the correlation processing between the observation data and the prediction data, and the correlation processing step. The size correction step of correcting at least one of the size of the object in the observation data and the size of the object in the prediction data is provided so that the size of the object in the prediction data and the size of the object in the prediction data match.

In such an object detection method, the size of the object in the observation data and the size of the object in the prediction data are matched so that the size of the object in the observation data and the size of the object in the prediction data match before the correlation processing is performed. Since at least one of the sizes is corrected, it is possible to suppress a decrease in the determination accuracy of the correlation due to the difference in size.

The object recognition device and the object recognition method according to the present invention can suppress a decrease in the accuracy of determining the correlation.

It is a block diagram which shows the structure of the object recognition apparatus in the case which includes the 1st to 5th detection part in an embodiment. It is a figure for demonstrating the 1st to 5th detection part attached to a vehicle. It is a figure for demonstrating the size replacement process. It is a figure for demonstrating the 1st and 2nd size correction processing. It is a flowchart which shows the operation of the said object recognition apparatus regarding the generation and use of recognition data. FIG. 5 is a flowchart showing the operation of the object recognition device regarding the first size correction process, the correlation process, and the size reset shown in FIG. FIG. 5 is a flowchart showing the operation of the object recognition device regarding the second size correction process and the generation of recognition data shown in FIG. It is a figure for demonstrating the effect of the object recognition apparatus.

Hereinafter, one or more embodiments of the present invention will be described with reference to the drawings. However, the scope of the invention is not limited to the disclosed embodiments. It should be noted that the configurations with the same reference numerals in the respective drawings indicate the same configurations, and the description thereof will be omitted as appropriate. In the present specification, when they are generically referred to, they are indicated by reference numerals without subscripts, and when they refer to individual configurations, they are indicated by reference numerals with subscripts.

The object recognition device in the embodiment is a device for recognizing an object existing around the own machine (an object visible from the own machine within a predetermined range). This object recognition device is object state information including the size and position of the object by clustering one or a plurality of detection points obtained by detecting an object around the own machine within a predetermined detection range. The clustering unit that obtains the observation data, the prediction unit that obtains the prediction data that predicts the object state information based on the past observation data, and the object represented by the observation data and the object represented by the prediction data are the same. The correlation unit that determines whether or not the data is based on the correlation processing between the observation data and the prediction data, and the size of the object in the observation data and the object in the prediction data before the correlation processing is performed by the correlation unit. A size correction unit (first size correction unit) for correcting at least one of the size of the object in the observation data and the size of the object in the prediction data is provided so as to match the size of. Then, in the present embodiment, this object recognition device is a recognition unit that obtains recognition data that is object state information as a recognition result based on the observation data obtained by the clustering unit and the prediction data obtained by the prediction unit. And, before the recognition data is obtained by the recognition unit, the size of the object in the observation data and the size of the object in the prediction data so that the size of the object in the observation data and the size of the object in the prediction data match. A size correction unit (second size correction unit) for correcting at least one of the two is further provided. Such an object recognition device may be used alone or may be mounted on another device (may be incorporated). The other device may or may not be a moving body such as a vehicle or a robot. Here, as an example, an object recognition device that further includes a detection unit for obtaining the detection point and is mounted on a vehicle (an example of a moving body) will be described more specifically.

FIG. 1 is a block diagram showing a configuration of an object recognition device when the first to fifth detection units are provided in the embodiment. FIG. 2 is a diagram for explaining the first to fifth detection units attached to the vehicle. FIG. 3 is a diagram for explaining the size replacement process. FIG. 3A shows a cluster frame before replacement, and FIG. 3B shows a cluster frame after replacement. FIG. 4 is a diagram for explaining the first and second size correction processes. 4A and 4B show the case where the cluster frame to be corrected exists in one quadrant, and FIGS. 4C and 4D show the case where the cluster frame to be corrected exists over a plurality of quadrants. FIG. 4F shows a case where the cluster frame of the predicted data exists beyond the detection range of the detection unit. 4A, 4C and 4E show before the correction, and FIGS. 4B, 4D and 4F show after the correction.

The object recognition device D in the embodiment includes, for example, a control processing unit 2 and a storage unit 3, as shown in FIG. In the example shown in FIG. 1, the object recognition device D further includes five first to fifth detection units 1-1 to 1-5 for obtaining detection points.

The first to fifth detection units 1-1 to 1-5 are connected to the control processing unit 2, respectively, and detect an object around the object recognition device D within a predetermined detection range according to the control of the control processing unit 2. Then, the detection result is output to the control processing unit 2 as a detection point. The moving body is a device capable of changing the position of its own machine, such as a vehicle or a robot. The vehicle is, for example, a parts carrier vehicle, an automobile, a railroad vehicle, or the like in a factory. In the present embodiment, as described above, the first to fifth detection units 1-1 to 1-5 are mounted on the vehicle VC which is an example of the moving body. These first to fifth detection units 1-1 to 1-5 are shown in FIG. 2, for example, so that an object around the vehicle VC (another object different from the own vehicle VC) can be detected in a wider range. As such, it is arranged in front of the vehicle VC and at each corner. The first detection unit 1-1 is provided at a substantially central position in the front portion of the vehicle VC, and is an object around the vehicle VC in a predetermined first detection range RF centered on the first direction PR which is the traveling direction of the vehicle VC. Is detected. The second detection unit 1-2 is provided at the front end on one side (for example, the right front corner) of the vehicle VC, and is a predetermined second centering on the second direction FR diagonally forward from the vehicle width direction on the one side. The detection range RRF detects an object around the vehicle VC. The third detection unit 1-3 is provided at the rear end of the vehicle VC on one side (for example, the right rear corner), and is a predetermined third detection unit centered on a third direction BR diagonally rearward from the vehicle width direction on the one side. 3 Detection range RRB detects an object around the vehicle VC. The fourth detection unit 1-4 is provided at the front end (for example, the left front corner) on the other side of the vehicle VC, and is a predetermined fourth centering on the fourth direction FL diagonally forward from the vehicle width direction on the other side. The detection range RLF detects an object around the vehicle VC. The fifth detection unit 1-5 is provided at the rear end of the other side of the vehicle VC (for example, the left rear corner), and is a predetermined fifth detection unit centered on a fifth direction BL diagonally rearward from the vehicle width direction of the other side. 5 Detection range RLB detects objects around the vehicle VC. Here, in the present embodiment, since the object recognition device D is mounted on the vehicle VC, the periphery of the vehicle VC (moving body) is the periphery of the object recognition device D (around the own machine) in the present embodiment. Means, and corresponds to one example. Similarly, the center position of the vehicle VC (moving body) described later means the center position of the object recognition device D (center position of the own machine) in the present embodiment, and corresponds to an example thereof.

Each of the first to fifth detection units 1-1 to 1-5 detects the object by transmitting a predetermined transmission wave and receiving the reflected wave of the transmission wave reflected by the object, respectively. It is a radar device that measures the position of the object (the relative position (relative direction, relative distance) of the object with respect to the own vehicle VC) by measuring the direction of the object and the distance to the object. The transmitted wave may be, for example, an electromagnetic wave in the millimeter wave band, an infrared laser beam, or an ultrasonic wave. These first to fifth detection units 1-1 to 1-5 transmit, for example, transmission waves in the millimeter wave band while scanning them in the first to fifth detection ranges RF, RRF, RRB, RLF, and RLB, respectively. The object is obtained by determining the direction of the object and the distance to the object based on the transmitting unit, the receiving unit that receives the reflected wave reflected by the transmitted wave, and the transmitted wave and the reflected wave. It is provided with a signal processing unit that obtains the position of the above as a detection point. The signal processing unit obtains the direction of the object from the transmission direction of the transmission wave in the scanning of the detection range, and based on the time difference between the transmission timing of the transmission wave and the reception timing of the reflected wave, reaches the object. Find the distance (TOF (Time-Of-Flight) method). The first to fifth detection units 1-1 to 1-5 are not limited to such a configuration, but include an appropriate configuration, for example, a plurality of receiving antennas, and the reflected wave received by the plurality of receiving antennas. It may be a device that obtains the direction of the object from the phase difference. Further, in the present embodiment, the radar device as an example of the first to fifth detection units 1-1 to 1-5 also has a relative velocity of the object with respect to the own vehicle VC based on the Doppler shift of the reflected wave with respect to the transmitted wave. It is obtained and output to the control processing unit 2.

Since the object has a predetermined size, one or a plurality of reflected waves can be obtained from each position (each location) corresponding to the spatial resolution of the detection unit 1 in one object. The detection units 1-1 to 1-5 each detect one or a plurality of detection points from one object. These first to fifth detection units 1-1 to 1-5 control processing units for one or more detection points (relative position (relative direction, relative distance) and relative speed in the object) detected in this way, respectively. Output to 2. The first to fifth detection units 1-1 to 1-5 are not limited to radar devices, but are within a predetermined detection range based on a pair of images captured by another device, for example, a stereo camera. It may be an object detection device that detects an object.

The storage unit 3 is a circuit connected to the control processing unit 2 and stores various predetermined programs and various predetermined data under the control of the control processing unit 2. The various predetermined programs include, for example, a control program that controls each part 1 and 3 of the object recognition device D according to the function of each part, and first to fifth detection ranges RF, RRF, RRB, and RLF. By clustering one or more detection points obtained by detecting an object around the vehicle VC (moving body) with RLB, clustering to obtain observation data which is object state information including the size and position of the object. Whether or not the object represented by the observation data and the object represented by the prediction data are the same as the program, the prediction program that obtains the prediction data that predicts the object state information based on the past observation data, and the prediction data. The size of the object in the observation data and the size of the object in the prediction data match before the correlation process for determining the observation data and the prediction data is performed or the correlation processing is performed by the correlation program. The first size correction program that corrects at least one of the size of the object in the observation data and the size of the object in the prediction data, and the observation data obtained by the clustering program and the prediction obtained by the prediction program. Based on the data, the size of the object in the observation data and the size of the object in the prediction data are determined before the recognition program for obtaining the recognition data which is the object state information as the recognition result and the recognition data in the recognition program. A second size correction program that corrects at least one of the size of the object in the observation data and the size of the object in the prediction data so as to match, and the clustering program before correction by the first size correction program. Based on the size of the object in the observation data obtained in the above, one size is selected from a plurality of predefined sizes, the selected size is used as the size of the object in the observation data, and the selected size is used. Recognition data that performs predetermined processing using the replacement program that replaces the size and position of the object in the observation data and the recognition data obtained by the recognition program so that the obtained position is the position of the object in the observation data. Control processing programs such as usage programs are included. The various predetermined data include, for example, data necessary for executing each program such as a plurality of predefined sizes used for replacing the size of an object. The storage unit 3 includes, for example, a ROM (Read Only Memory) which is a non-volatile storage element, an EEPROM (Electrically Erasable Programmable Read Only Memory) which is a rewritable non-volatile storage element, and the like. The storage unit 3 includes a RAM (Random Access Memory) or the like that serves as a working memory of the so-called control processing unit 2 that stores data or the like generated during the execution of the predetermined program.

The control processing unit 2 is a circuit for controlling each unit 1 and 3 of the object recognition device D according to the function of each unit and recognizing an object existing around the vehicle VC. The control processing unit 2 includes, for example, a CPU (Central Processing Unit) and its peripheral circuits. When the control processing program is executed, the control processing unit 2 includes a control unit 21, a clustering unit 22, a replacement unit 23, a first size correction unit 24, a correlation unit 25, a second size correction unit 26, and a recognition unit 27. , The prediction unit 28 and the recognition data utilization unit 29 are functionally provided.

The control unit 21 controls each of the units 1 and 3 of the object recognition device D according to the function of each unit, and controls the entire control of the object recognition device D.

The clustering unit 22 is obtained by detecting an object around the vehicle VC in the first to fifth detection ranges RF, RRF, RRB, RLF, and RLB by the first to fifth detection units 1-1 to 1-5. By clustering one or a plurality of detection points, observation data which is object state information including the size and position of the object is obtained. In the present embodiment, the object state information includes not only the size and position (relative direction, relative distance) of the object, but also the relative velocity of the object. As described above, since the object has a predetermined size, one or a plurality of reflected waves can be obtained from each position (each location) according to the spatial resolution of the detection unit 1 in one object. The first to fifth detection units 1-1 to 1-5 each detect one or a plurality of detection points from one object. Therefore, the clustering unit 22 clusters (groups) the detection points detected by the first to fifth detection units 1-1 to 1-5, and the clusters (groups) are the detection points obtained from one object. Defined as a set of. As the clustering method, various methods such as the shortest distance method, the group average method, and Ward's method are known and can be used. Since the detection points output from each of the first to fifth detection units 1-1 to 1-5 are based on the respective arrangement positions of the respective detection units 1-1 to 1-5, they are detected. Before clustering points, they are represented in one unified coordinate system, and the position (relative direction, relative distance) and relative velocity of the object are also represented in this coordinate system. The clustering unit 22 generates a frame (cluster frame) surrounding the detection points belonging to the cluster in the cluster as a frame surrounding the area in which the object associated with the cluster exists, and is a representative of the detection points belonging to the cluster. Find the point. The size of the object associated with the cluster is represented by the size of the generated cluster frame (frame size), the position of the object associated with the cluster is represented by the position of the representative point, and the velocity is , Represented by the speed of the representative point. The representative points are, for example, the center position of the cluster frame, the position of the center of gravity, and the like. In the present embodiment, the representative point is the center position of the cluster frame. The cluster frame may have an arbitrary shape such as a circle or a fan shape, which is set in advance, but in the present embodiment, it is a rectangle and is a rectangular frame (rectangular frame). Here, when the object is located substantially directly in front of or substantially directly on the side surface of the detection unit 1, the plurality of detection points from the object are arranged in a line segment, so that the rectangle as the cluster frame generated by the clustering unit 22 is formed. , Shall include line segments. The cluster frame of such a line segment is also replaced with a rectangle by the replacement unit 23, as will be described later. In the present embodiment, since the cluster frame is a rectangular frame, the central position as a representative point is an intersection of diagonal lines.

The replacement unit 23 selects one size from a plurality of predefined sizes based on the size of the object in the observation data obtained by the clustering unit 22 before the correction is performed by the first size correction unit 24. It replaces the size and position of the object in the observation data so that the selected size is the size of the object in the observation data and the position obtained from the selected size is the position of the object in the observation data. ..

More specifically, for example, the size is divided into four first to fourth size divisions XS, S, M, L. The first size category is a category assuming that the object detected by the detection unit 1 is, for example, a pedestrian, and the cluster frame (first cluster frame) of the first size category has a length of 1.0 [m. ] Is a rectangular frame having a width of 1.0 [m] (1.0 [m] × 1.0 [m]). The second size classification is a classification assuming that the object detected by the detection unit 1 is, for example, a bicycle, and the cluster frame (second cluster frame) of the second size classification has a length of 2.0 [m]. It is a rectangular frame having a width of 1.0 [m] (2.0 [m] × 1.0 [m]). The third size classification is a classification assuming that the object detected by the detection unit 1 is, for example, an ordinary vehicle, and the cluster frame (third cluster frame) of the third size classification has a length of 5.0 [m. ] Is a rectangular frame having a width of 1.8 [m] (5.0 [m] × 1.8 [m]). The fourth size classification is a classification assuming that the object detected by the detection unit 1 is, for example, a large vehicle, and the cluster frame (fourth cluster frame) of the fourth size classification has a length of 10 [m]. It is a rectangular frame having a width of 2.0 [m] (10 [m] × 2.0 [m]). In this example, the size is divided into four, but the number of divisions is arbitrary.

Such first to fourth size divisions and their first to fourth cluster frames are defined in advance and stored in the storage unit 3, and the replacement unit 23 first determines the object in the observation data obtained by the clustering unit 22. It has at least one of the length and width in the rectangular frame of the cluster associated with the object, that is, at least one of the length and width in the cluster frame of the larger size division, and is the smallest in size. Select the size classification of the cluster frame from the first to fourth size classifications. More specifically, for example, the replacement unit 23 has a length and width larger than the length and width of the rectangular frame of the cluster associated with the object obtained by the clustering unit 22, and has the smallest cluster frame. The size category of is selected from the first to fourth size categories. In one example, when the rectangular frame of the cluster obtained by the clustering unit 22 is 1.2 [m] × 0.8 [m], the second size category S is selected, and the rectangular frame of the cluster obtained by the clustering unit 22 is selected. Is replaced with a second cluster frame of 2.0 [m] × 1.0 [m]. Alternatively, for example, the replacement unit 23 has a length larger than the length in the rectangular frame of the cluster associated with the object obtained by the clustering unit 22, and the size division of the cluster frame having the smallest size is the first. Or select from the 4th size category. In one example, when the rectangular frame of the cluster obtained by the clustering unit 22 is 4.4 [m] × 0 [m], the third size division M is selected, and the rectangular frame of the cluster obtained by the clustering unit 22 is It is replaced with a third cluster frame of 5.0 [m] × 1.8 [m]. Alternatively, for example, the replacement unit 23 has a width larger than the width in the rectangular frame of the cluster associated with the object obtained by the clustering unit 22, and determines the size classification of the cluster frame having the smallest size. Select from the 4th size category. In one example, when the rectangular frame of the cluster obtained by the clustering unit 22 is 1.0 [m] × 0.8 [m], the first size division XS is selected, and the rectangular frame of the cluster obtained by the clustering unit 22 is selected. Is replaced with a first cluster frame of 1.0 [m] × 1.0 [m]. Next, the replacement unit 23 has the cluster frame of the selected size division so that the positions of the most recent contacts closest to the vehicle VC in the rectangular frame of the cluster obtained by the clustering unit 22 are the same before and after the replacement. To set. Then, the replacement unit 23 obtains the center position of the set cluster frame as the position of the representative point. As a result, the size and position of the object in the observation data are set so that the selected size is the size of the object in the observation data and the position obtained from the selected size is the position of the object in the observation data. Will be replaced. In the present embodiment, the replacement unit 23 may obtain the speed of the representative point and replace the speed, but since the speed does not substantially change before and after the replacement of the cluster frame, the replacement of the speed may be omitted. .. By replacing the cluster frame FR with the recent contact having a relatively high detection accuracy as a reference, the position accuracy can be maintained or its deterioration can be suppressed before and after the replacement.

For example, the rectangular frame FRb1 of the cluster obtained by the clustering unit 22 shown in FIG. 3A is replaced with the second cluster frame FRa1 of the second size category S selected from the first to fourth size categories shown in FIG. 3B. The replaced second cluster frame FRa1 has a point such that the position of the latest contact CS _b1 closest to the vehicle VC in the rectangular frame FRb1 of the cluster obtained by the clustering unit 22 is the same before and after the replacement. The center position CP _a1 of the second cluster frame FRa1 set in CS _a1 (= CS _b1 ) and set in this way is obtained. As a result, the observation data obtained by the clustering unit 22 is replaced from the rectangular frame FRb1 shown in FIG. 3A and its center position CP _b1 to the second cluster frame FRa1 shown in FIG. 3B and its center position CP _a1 .

Alternatively, for example, the rectangular frame FRb2 (cluster frame of the line segment) of the cluster obtained by the clustering unit 22 shown in FIG. 3A is the third size category M selected from the first to fourth size categories shown in FIG. 3B. In the replaced third cluster frame FRa2, the position of the nearest contact CS _b2 closest to the vehicle VC in the rectangular frame FRb2 of the cluster obtained by the clustering unit 22 is before and after the replacement. Then, the center position CP _a2 of the third cluster frame FRa2 set in this way is obtained, which is set at the point CS _a2 (= CS _b2 ) so as to be the same.

Here, as shown in FIG. 3, a coordinate system in which a predetermined first coordinate axis X, a second coordinate axis Y orthogonal to the first coordinate axis X, and a center position of the vehicle VC are coordinate origins (0, 0). When the side of the rectangular frame of the cluster obtained by the clustering unit 22 intersects with the first coordinate axis X or the second coordinate axis Y, at the closest tangent to the vehicle VC at the intersecting side. , The ratio of the first length of the first partial side extending to one quadrant to the second length of the second partial side extending to the other quadrant is replaced so as to be the same before and after the replacement. The first coordinate axis X extends, for example, along the viewing direction which is the center line of the detection range (for example, in the case of only the first detection unit 1-1, the viewing direction is the center of the detection range RF. Line PR). Alternatively, for example, the first coordinate axis X extends along the traveling direction of the own machine when moving. In the present embodiment, since it is mounted on the vehicle VC (moving body), the first coordinate axis X extends along the traveling direction of the vehicle VC.

The sides S ₁ S ₂ of the rectangular frame FRb2 shown in FIG. 3A intersect the second coordinate axis Y at the recent contact CS _b2 on the second coordinate axis Y, and the rectangular frame FRb2 is the vehicle at the intersecting sides. At the nearest edge S ₁ S ₂ closest to the VC, the first length L1 of the first partial side CS _b2 S ₁ extending to one quadrant and the second length L2 of the second partial side CS _b2 S ₂ extending to the other quadrant. It is replaced with the cluster frame FRa2 so that the ratio (L1: L2) with and is the same before and after the replacement. That is, in the nearest tangent S'1 S'2 closest to the vehicle VC in the cluster frame _FRa2 , the first length L' ₁ of the _first partial side CS _a2 S'1 extending to one quadrant and the other quadrant. The second length L' ₂ of the extending second partial side CS _{a2 S'2} is the length given by L1: L2 = L'1: L'2, respectively.

As a result, the observation data obtained by the clustering unit 22 is replaced from the rectangular frame FRb2 shown in FIG. 3A and its center position CP _b2 to the third cluster frame FRa2 shown in FIG. 3B and its center position CP _a2 .

In these above-mentioned cases, each side of the cluster frame is parallel before and after the replacement. That is, the traveling direction of the object represented by the cluster frame is maintained before and after the replacement.

Before the correlation processing is performed by the correlation unit 25, the first size correction unit 24 determines the size of the object in the observation data and the prediction so that the size of the object in the observation data and the size of the object in the prediction data match. It corrects at least one of the sizes of objects in the data. As described above, the difference between the size of the object in the observation data and the size of the object in the prediction data is inconvenient in the correlation processing, so the correction target for correcting the size is the size of the object in the observation data and the size of the object in the prediction data. Either is fine.

More specifically, first, the first size correction unit 24 corrects the position of the nearest contact point closest to the vehicle VC (moving body) in the cluster frame so that it is the same before and after the correction. ..

In one example, as shown in FIG. 4A, in the case of the cluster frame FRb3 of the object in the observation data and the cluster frame FRb4 of the object in the prediction data, here, for convenience, when correcting the size of the object in the observation data, FIG. 4B. As shown in, the cluster frame FRb3 has the recent contact CS _a3 (= recent contact CS _b3 ) and has the cluster frame FRb4 so that the positions of the recent contacts CS _b3 in the cluster frame FRb3 match before and after the correction. It is corrected to the cluster frame FRa3 having a size. As a result, the position of the object in the observation data is corrected from the center position CP _b3 of the cluster frame FRb3 to the center position CP _a3 of the cluster frame FRa3. Since the cluster frame FRa4 of the object in the prediction data shown in FIG. 4B is uncorrected, it is the cluster frame FRb4 of the object in the prediction data shown in FIG. 4A (cluster frame FRb4 = cluster frame FRa4, center position CP of the cluster frame FRb4). _b4 = center position CP _a4 of the cluster frame FRa4).

Second, when a predetermined first coordinate axis X, a second coordinate axis Y orthogonal to the first coordinate axis X, and a coordinate system having the center position of the vehicle VC as the coordinate origin (0, 0) are set. When the side of the cluster frame intersects the first coordinate axis X or the second coordinate axis Y, the first size correction unit 24 extends to one quadrant at the closest tangent to the vehicle VC at the intersecting side. The ratio of the first length of one partial side to the second length of the second partial side extending to the other quadrant is corrected so as to be the same before and after the correction. Similar to the case of FIG. 3, the first coordinate axis X extends, for example, along the viewing direction which is the center line of the detection range. Alternatively, for example, the first coordinate axis X extends along the traveling direction of the own machine when moving. In the present embodiment, since it is mounted on the vehicle VC (moving body), the first coordinate axis X extends along the traveling direction of the vehicle VC. In this case as well, the first size correction unit 24 corrects the position of the nearest contact point closest to the vehicle VC in the cluster frame so that it is the same before and after the correction.

In one example, as shown in FIG. 4C, in the case of the cluster frame FRb5 of the object in the observation data and the cluster frame FRb6 of the object in the prediction data, each of the cluster frame FRb5 and the cluster frame FRb6 intersects the second coordinate axis Y (in this example,). (Orthogonal). Here, for convenience, when correcting the size of the object in the observation data, as shown in FIG. _4C , in the nearest quadrant S3 S4 at the side intersecting the second coordinate axis Y in the cluster frame _FRb5 , one quadrant ( In this example, the first length a1 of the _first partial side CP _{b5 S3} extending to the first _quadrant ) and the second length a2 of the second partial side CP _{b5 S4} extending to the other quadrant (the fourth quadrant in this example). As shown in FIG. 4D, the cluster frame FRb5 extends to one quadrant (the first quadrant in this example) so that the ratio (a1: a2) to and (a1: a2) is the same before and after the correction. The ratio of the first length _a'1 of _a5 S'3 to the second length a'2 of the second partial side CP _a5 S'4 extending to the other quadrant ( _fourth quadrant in this example) (a'1: 1. a'2) is the ratio (a1: a2), (a'1: a'2 = a1: a2), and has the latest contact CS _a5 (= recent contact CS _b5 ) on the second coordinate axis Y. And, it is corrected to the cluster frame FRa5 having the size of the cluster frame FRb6. As a result, the position of the object in the observation data is corrected from the center position CP _b5 of the cluster frame FRb5 to the center position CP _a5 of the cluster frame FRa5. Since the cluster frame FRa6 of the object in the prediction data shown in FIG. 4D is uncorrected, it is the cluster frame FRb6 of the object in the prediction data shown in FIG. 4C (cluster frame FRb6 = cluster frame FRa6, center position CP of the cluster frame FRb6). _b6 = center position CP _a6 of the cluster frame FRa6). In the above, the cluster frame FRb5 and the cluster frame FRb6 each intersect the second coordinate axis Y (orthogonal in this example), but the side of the cluster frame to be corrected is the first coordinate axis X or the second coordinate axis Y. When they intersect, the above process is applied, and the sides of the cluster frame that are not subject to correction may or may not intersect the first coordinate axis X or the second coordinate axis Y.

Third, when the cluster frame of the object in the prediction data exists beyond the detection range of the detection unit 1, the first size correction unit 24 is the closest to the vehicle VC (moving body) at the side intersecting the detection range. In the second closest edge, the ratio of the third length of the third partial side extending beyond the detection range to the fourth length of the fourth partial side extending within the detection range is the predicted data and the corrected said. The size of the object in the observation data is corrected so that it becomes the same as the observation data. In this case as well, the first size correction unit 24 corrects the position of the nearest contact point closest to the vehicle VC in the cluster frame so that it is the same before and after the correction, but the closest contact intersects the detection range. Since the nearest tangent side is extended starting from the intersection of the sides, the position of the recent contact point in the orthogonal direction orthogonal to the extension direction of the closest tangent side is corrected so as to be the same before and after the correction.

In one example, as shown in FIG. 4E, in the case of the cluster frame FRb7 of the object in the observation data and the cluster frame FRb8 of the object in the prediction data, the cluster frame FRb8 of the object in the prediction data is the detection range of the third detection unit 1-3. It exists beyond the RRB. In this case, it is desirable to correct the size of the object in the observation data. As shown in FIG. 4E, in the second closest edge S5 S6 at the side intersecting the detection range _RRB of the third detection unit 1-3 in the cluster frame _FRb8 , the third partial side extending beyond the detection range RRB. The ratio (b1: b2) of the third length b1 of the BP ₁ S ₅ to the fourth length b2 of the fourth partial side BP ₁ S ₆ extending within the detection range RRB is the predicted data and the corrected said. As shown in FIG. 4F, the cluster frame _FRb7 is located on the second closest edge S7 _S8 at the side intersecting the detection range RRB of the third detection unit 1-3 so as to be the same as the observation data. The third length a'1 of the third partial side BP ₂ S ₇ extending beyond the detection range RRB and the fourth length a'2 of the fourth partial side BP ₂ S ₈ extending within the detection range RRB. The ratio (a'1: a'2) is the ratio (b1: b2) (a'1: a'2 = b1: b2), and the second contact CS _b7 of the cluster frame FRb7. The second closest edge S ₇ S ₈ is held at a position orthogonal to the extension direction of the closest edge S ₇ S ₈ (in this example, the direction of the second coordinate axis Y), and the size of the cluster frame FRb 8 is set. It is corrected to the cluster frame FRa7 to have. As a result, the position of the object in the observation data is corrected from the center position CP _b7 of the cluster frame FRb7 to the center position CP _a7 of the cluster frame FRa7. As described above, the recent contact CS _b7 of the cluster frame FRb7 shown in FIG. 4E has the same position in the recent contact in the orthogonal direction orthogonal to the extension direction of the nearest contact before and after the correction. As a result of maintaining the coordinate values of the second coordinate axis Y before and after the correction, in the cluster frame FRa7 shown in FIG. 4F, the recent contact CS _a7 (point directly beside the vehicle VC) on the second coordinate axis Y. It has become. Since the cluster frame FRa8 of the object in the prediction data shown in FIG. 4F is uncorrected, it is the cluster frame FRb8 of the object in the prediction data shown in FIG. 4E (cluster frame FRb8 = cluster frame FRa8, center position CP of the cluster frame FRb8). _b8 = center position CP _a8 of the cluster frame FRa8).

In the first to third cases described above, each side of the cluster frame is parallel before and after the correction. That is, the traveling direction of the object represented by the cluster frame is maintained before and after the correction.

The correlation unit 25 determines whether or not the object represented by the observation data and the object represented by the prediction data are the same by the correlation processing between the observation data and the prediction data.

More specifically, for example, the correlation unit 25 is centered on the position of the object in the prediction data (the center position of the cluster frame FR as a representative point in the present embodiment) after the correction by the first size correction unit 24. The correlation process is performed depending on whether or not the position of the object in the observation data (the center position of the cluster frame FR as a representative point in the present embodiment) exists within a predetermined range (correlation determination range) determined in advance. Run. When the position (representative point) of the object in the observation data exists within the correlation determination range, the correlation unit 25 determines that the object represented by the observation data and the object represented by the prediction data are the same. If the position (representative point) of the object in the observation data does not exist within the correlation determination range, it is determined that the object represented by the observation data and the object represented by the prediction data are not the same. .. If it is determined that the object represented by the observation data is not the same as all the objects represented by the prediction data, the correlation unit 25 treats the object represented by this uncorrelated observation data as a newly recognized object. .. On the other hand, the correlation unit 25 continuously deviates from the surroundings of the vehicle VC an object represented by the predicted data which is not determined to be the same as any of the objects represented by the observation data for a predetermined number of times. Treat as an object. The correlation determination range is appropriately set according to the specifications of the object recognition device D for determining the identity, and is set, for example, to the size of the object in the prediction data (the size of the cluster frame in the present embodiment). In the case of such correlation processing, it is desirable to maintain the position of the recent contact point before and after the correction in order to appropriately move the representative point before and after the correction.

Alternatively, for example, the correlation unit 25 is the area of the overlapping portion between the object in the prediction data and the object in the observation data after the correction by the first size correction unit 24 (in this embodiment, the cluster frame of the object in the prediction data and the observation data. The correlation process is executed depending on whether or not the area of the overlapping portion of the object with the cluster frame in the above is equal to or greater than a predetermined threshold value (correlation determination threshold value). The correlation unit 25 determines that the object represented by the observation data and the object represented by the prediction data are the same when it is equal to or higher than the correlation determination threshold, and when it is less than the correlation determination threshold, the correlation unit 25 determines that the object is the same. , It is determined that the object represented by the observation data and the object represented by the prediction data are not the same. The correlation determination threshold value is appropriately set according to the specifications of the object recognition device D for determining the identity. In the case of such correlation processing, since the area of the overlapping portion is evaluated, it is not always necessary to maintain the position of the recent contact point before and after the correction.

Then, after the completion of such the correlation processing, the correlation unit 25 returns the size of the object corrected by the first size correction unit 24 to before the correction. In the present embodiment, as described above, the size of the object in the observation data (cluster frame FR) is corrected, so that the size of the object in the observation data (cluster frame FR) is returned before the correction. That is, the size replaced by the replacement unit 23 is the size of the object in the observation data.

The second size correction unit 26 sets the size of the object in the observation data and the prediction data so that the size of the object in the observation data and the size of the object in the prediction data match before the recognition data is obtained by the recognition unit 27. It corrects at least one of the sizes of objects in.

A more specific correction method is the same as that of the first size correction unit 24. That is, the second size correction unit 26 corrects the position of the nearest contact point from the vehicle VC (moving body) in the cluster frame so that it is the same before and after the correction. When a predetermined first coordinate axis X, a second coordinate axis Y orthogonal to the first coordinate axis X, and a coordinate system in which the center position of the vehicle VC is the coordinate origin (0, 0) are set, the sides of the cluster frame are set. When crosses the first coordinate axis X or the second coordinate axis Y, the second size correction unit 26 is a first partial side extending in one quadrant at the nearest closest side to the vehicle VC at the intersecting side. The ratio of the first length to the second length of the second partial side extending to the other quadrant is corrected so as to be the same before and after the correction. When the cluster frame of the object in the prediction data exists beyond the detection range of the detection unit 1, the second size correction unit 26 is the closest to the vehicle VC (moving body) at the side intersecting the detection range. In the side, the ratio of the third length of the third partial side extending beyond the detection range to the fourth length of the fourth partial side extending within the detection range is the ratio between the predicted data and the corrected observation data. , Correct the size of the object in the observation data so that they are the same.

The recognition unit 27 obtains recognition data, which is object state information as a recognition result, based on the observation data obtained by the clustering unit 22 and the prediction data obtained by the prediction unit 28. The recognition unit 27 obtains the velocity of the object in the recognition data based on the position of the object in the observation data and the position and velocity of the object in the prediction data.

The prediction unit 28 obtains prediction data for predicting object state information based on past observation data.

More specifically, the recognition unit 27 and the prediction unit 28 obtain prediction data based on past observation data by using a so-called Kalman filter or the like, obtain the prediction data in the clustering unit 22, and replace the observation data in the replacement unit 23. And the recognition data is obtained based on the prediction data obtained by the prediction unit 28. More specifically, in the present embodiment, the repetitive object is detected in the sampling cycle T, and the repetitive recognition data is obtained in the sampling cycle T. At the time tk of this repeated sampling timing, the position of the object in the observation data (center position of the cluster frame) is x _ok , the position of the object in the prediction data is x _{pk ,} and the position of the object in the recognition data is x _sk . When the velocity of the object in the prediction data is v _pk , the velocity of the object in the recognition data is v _sk , the predetermined first parameter is α, and the predetermined second parameter is β, the object in the recognition data The position x _sk and the velocity v _sk are obtained by the following equations 1 and 2, and the position x _pk and the velocity v _pk of the object in the prediction data are obtained by the following equations 3 and 4. The first and second parameters α and β are gains for the prediction dependence of the position and velocity in the so-called α-β filter, and are appropriately set.
Equation 1; x _sk = x _pk + α × (x _ok -x _pk )
Equation 2; v _sk = v _pk + β / T × (x _ok -x _pk )
Equation 3; x _pk = x _sk-1 + T × v _sk-1
Equation 4; v _pk = v _sk-1

On the other hand, when the size of the object in the prediction data is _spk and the size of the object in the recognition data is _sk at the time tk, the prediction unit 28 _{sets the size spk of} the object in the prediction data by the following equation 5. Ask.
Equation 5; s _pk = s _sk-1

Then, in the present embodiment, the recognition unit 27 obtains the position of the object in the recognition data as described above, and then further corrects the size and prediction of the object in the observation data before the correction by the second size correction unit 26. The larger size of the object size in the data is selected and the size of the object in the recognition data is replaced so that the selected size is the size of the object in the recognition data. The smaller of the size of the object in the observation data and the size of the object in the prediction data may leak the detection point and become that size, so replace it with the larger size of both as described above. This makes it possible to bring the size of the object in the recognition data closer to the true size. As a result, the accuracy of collision determination and the like in the recognition data utilization unit 29, which will be described later, is improved. The replacement of the size of the object is processed in the same manner as the replacement of the replacement portion 23 described above with reference to FIG. That is, the replacement unit 23 sets the size of the object in the observation data obtained by the clustering unit 22 to the size of the size division selected as described above, and before and after the replacement, the position of the latest contact point and each partial side in each quadrant. It was replaced so that the ratio of each length would be the same, but here, the recognition unit 27 of the object in the observation data before the size of the object in the observation data is corrected by the second size correction unit 26. Replace with the larger size of the size of the object in the size and prediction data so that the position of the most recent contact and the ratio of each length at each partial side in each quadrant are the same before and after the replacement.

Since the size of the object in the recognition data is replaced in this way, the recognition unit 27 obtains the position of the object (center position of the cluster frame) by the size of the replaced object, and the position of the obtained object is the object in the recognition data. The position of the object in the recognition data is replaced so as to be the position of.

This gives the size, position and velocity of the object in the final recognition data.

The recognition data utilization unit 29 carries out a predetermined process using the recognition data obtained by the recognition unit 27. For example, the recognition data utilization unit 29 is a collision determination unit that determines whether or not the vehicle VC collides with an object detected by the first to fifth detection units 1-1 to 1-5. The collision determination unit determines the presence or absence of the collision based on, for example, the position and speed of the vehicle VC and the size, position, and speed in the recognition data of the object. When the collision determination unit determines that there is a collision, the collision determination unit may notify an alarm or the like. Alternatively, for example, the recognition data utilization unit 29 is an automatic driving processing unit that performs an automatic driving process for automatically changing lanes of the vehicle VC. For example, when changing lanes, the automatic driving processing unit detects a white line from the image of the camera, recognizes the traveling lane of the vehicle VC and the lane to be changed, and determines the position and speed of the vehicle VC and the size in the recognition data of the object. , Position and speed to change lanes while avoiding collisions between the vehicle VC and the object.

The control processing unit 2 and the storage unit 3 in such an object recognition device D can be configured by a computer called an so-called ECU (Electronic Control Unit).

In addition, the first size correction unit 24 of the object in the observation data so that the size of the object in the observation data and the size of the object in the prediction data match before the correlation processing is performed in the correlation unit. It corresponds to an example of a size correction unit that corrects at least one of the size and the size of the object in the prediction data.

Next, the operation of this embodiment will be described. FIG. 5 is a flowchart showing the operation of the object recognition device regarding the generation and use of recognition data. FIG. 6 is a flowchart showing the operation of the object recognition device regarding the first size correction process, the correlation process, and the size reset shown in FIG. FIG. 7 is a flowchart showing the operation of the object recognition device regarding the second size correction process and the generation of recognition data shown in FIG.

When the vehicle VC starts operation, such an object recognition device D executes necessary initialization of each part and starts the operation. By executing the control processing program, the control processing unit 2 includes a control unit 21, a clustering unit 22, a replacement unit 23, a first size correction unit 24, a correlation unit 25, a second size correction unit 26, a recognition unit 27, and a prediction unit. The unit 28 and the recognition data utilization unit 29 are functionally configured.

The object recognition device D repeatedly executes each of the following operations shown in FIG. 5 in a predetermined sampling cycle T in recognizing an object around the vehicle VC (moving body).

In FIG. 5, the object recognition device D detects the detection points detected by the first to fifth detection units 1-1 to 1-5 at the time _tk of the sampling timing of the first to fifth detection units 1-. Obtained from 1 to 1-5 (S1).

Next, the object recognition device D processes an object around the vehicle VC (moving body) in the first to fifth detection ranges RF, RRF, RRB, RLF, and RLB by the clustering unit 22 of the control processing unit 2. The observation data of the object is obtained by clustering one or a plurality of detection points acquired from the first to fifth detection units 1-1 to 1-5 in (S2).

Next, the object recognition device D selects one size from a plurality of predefined sizes based on the size of the object in the observation data obtained by the clustering unit 22 by the replacement unit 23 of the control processing unit 2. , The size and position of the object in the observation data are replaced so that the selected size is the size of the object in the observation data and the position obtained from the selected size is the position of the object in the observation data (. S3).

Next, the object recognition device D obtains the prediction data based on the past observation data by the prediction unit 28 of the control processing unit 2 (S4). In the present embodiment, the size _spk , position x _pk , and velocity v _pk of the object in the prediction data are _spk = s _sk-1 , x _pk = x _sk-1 , + T × v _sk-1 , v _pk =, respectively. Obtained by v _sk-1 .

The object recognition device D executes the first size correction process, the correlation process, and the size reset by the control processing unit 2 (S5).

More specifically, in this process S5, each operation shown in FIG. 6 is executed.

In FIG. 6, the object recognition device D determines whether or not the size of the object in the observation data and the size of the object in the prediction data are different by the first size correction unit 24 of the control processing unit 2 (S51). In this determination, since the size of the object is one of the first to fourth size categories in the present embodiment by the replacement unit 23, the same or not is determined by the size category. As a result of this determination, when each size classification of the two data is different (Yes), the object recognition device D executes the process S53 after the next process S52. On the other hand, as a result of the determination, when each size classification of both data is the same (No), the object recognition device D next executes the process S53.

In this process S52, in the present embodiment, the object recognition device D uses the first size correction unit 24 to match the size of the object in the observation data with the size of the object in the prediction data. Correct the size.

In the processing S53, the object recognition device D uses the correlation unit 25 of the control processing unit 2 to determine whether or not the object represented by the observation data and the object represented by the prediction data are the same as the observation data. Judgment is made by correlation processing with the prediction data. As a result of this determination, when each object of both data is the same (Yes), the object recognition device D executes the process S55 after the next process S54. On the other hand, as a result of the determination, when the objects of both data are not the same (No), the object recognition device D then executes the process S55.

In this process S54, the object recognition device D associates the object represented by the observation data with the object represented by the prediction data by the correlation unit 25. As a result, object tracking is performed at each sampling timing.

In the process S55, the object recognition device D returns the size of the object corrected by the first size correction unit 24 before the correction in order to reset the size by the correlation unit 25, and ends the process S5.

Returning to FIG. 5, next, the object recognition device D executes the second size correction process, the correlation process, and the generation of the recognition data by the control processing unit 2 (S6).

More specifically, in this process S6, each operation shown in FIG. 7 is executed.

In FIG. 7, the object recognition device D determines whether or not the size of the object in the observation data and the size of the object in the prediction data are different by the second size correction unit 26 of the control processing unit 2 in the present embodiment. (S61). As a result of this determination, when each size classification of the two data is different (Yes), the object recognition device D executes the process S63 after the next process S62. On the other hand, as a result of the determination, when each size classification of both data is the same (No), the object recognition device D next executes the process S63.

In this process S62, in the present embodiment, the object recognition device D has the object in the observation data so that the size of the object in the observation data and the size of the object in the prediction data match by the second size correction unit 26. Correct the size.

In the processing S63, the object recognition device D obtains the position x _sk of the object in the recognition data by the recognition unit 27 of the control processing unit 2 using the above equation 1 in the present embodiment.

Next, the object recognition device D obtains the velocity _vsk of the object in the recognition data by the recognition unit 27 using the above equation 2 in the present embodiment (S64).

Next, the object recognition device D selects the larger size of the size of the object in the observation data and the size of the object in the prediction data by the recognition unit 27, and uses this selected size as the size of the object in the recognition data. The size of the object in the recognition data is replaced with the above (S65).

Then, the object recognition device D obtains the position of the object (center position of the cluster frame) by the size of the object replaced by the process S65 by the recognition unit 27, and the position of the obtained object is the position of the object in the recognition data. The position of the object in the recognition data is replaced with the above-mentioned recognition data, and the process S6 is terminated.

Returning to FIG. 5, next, the object recognition device D performs a predetermined process using the recognition data obtained in the process S6 by the recognition data utilization unit 29 of the control processing unit 2 (S7).

Each of the processes S2 to S7 is executed for each object represented by the observation data, and the process at the current sampling timing is completed. As described above, when it is determined that the object represented by the observation data is not the same as all the objects represented by the prediction data, the correlation unit 25 determines the object represented by this uncorrelated observation data. Treat it as a newly recognized object. On the other hand, the correlation unit 25 continuously deviates from the surroundings of the vehicle VC an object represented by the predicted data which is not determined to be the same as any of the objects represented by the observation data for a predetermined number of times. Treat as an object.

As described above, the object recognition device D and the object recognition method implemented therein in the embodiment are described in the object size in the observation data and the prediction data before the correlation processing is performed by the first size correction unit 24. Since at least one of the size of the object in the observation data and the size of the object in the prediction data is corrected so as to match the size of the object, the decrease in the determination accuracy of the correlation due to the difference in size can be prevented. It can be suppressed.

The object recognition device D and the object recognition method are described in the observation data so that the size of the object in the observation data and the size of the object in the prediction data match before the recognition data is obtained by the second size correction unit 26. Since at least one of the size of the object and the size of the object in the prediction data is corrected, it is possible to suppress a decrease in at least the detection accuracy of the size of the object and the detection accuracy of the position due to the difference in size. In the present embodiment, it is possible to suppress a decrease in speed detection accuracy.

FIG. 8 is a diagram for explaining the effect of the object recognition device. 8A shows the prediction data based on the previous observation data, FIG. 8B shows the current detection point, and FIG. 8C shows the observation data obtained by clustering based on the detection point shown in FIG. 8B. The observation data shown in FIG. 8C is the observation data before being replaced by the size classification.

In general, a detector (sensor, corresponding to the first to fifth detection units 1-1 to 1-5 in this embodiment) for detecting (sensing) an object, such as a radar device, is a predetermined predetermined device. An object is detected within the detection range. In the example shown in FIG. 8, the detector arranged on the right front side of the vehicle VC detects the object in the detection range RRF which is a fan shape of about 120 degrees from the right front side to the middle of the right side surface, and detects the object on the right side of the vehicle VC. The detector arranged at the rear detects the object in the detection range RRB which is a fan shape of about 120 degrees from the middle of the right side surface to the right rear. Due to the performance limit of the detector, even if an object exists within the detection range, for example, as shown in FIG. 8B, a detection point is generated in a part of the object, and a detection point in the rest of the object is generated. It may leak. In particular, the detection accuracy tends to decrease near the boundary between the detection ranges RRF and RRB, and such leakage may occur even near the overlap between the detection range RRF and the detection range RRB. The cluster frame generated by clustering based on the detection points on a part of the object shown in FIG. 8C is different from the true cluster frame because it is the detection point on a part of the object. Therefore, the predicted data based on the previous observation data shown in FIG. 8A and the observation data obtained by clustering based on the detection point shown in FIG. 8B shown in FIG. 8C are the objects detected last time based on the size. When determining the correlation with the object detected this time, the accuracy of determining the correlation is lowered. For example, when the determination is made by the correlation processing as in the present embodiment, the positional relationship of the representative points depending on the frame size is different from the true positional relationship, so that the determination accuracy of the correlation is lowered.

However, in the object recognition device D and the object recognition method in the embodiment, as described above, the size is corrected by the first size correction unit 24 before the correlation processing is performed, so that the correlation due to the difference in size is caused. It is possible to suppress a decrease in determination accuracy.

On the other hand, when the position of a predetermined point (representative point) of the size of the object in the area of the object is the position of the object, if the detection accuracy of the size of the object decreases as described above, the area of the object based on the detection result. Is deviated from the region of the true object, so that the accuracy of the position of the object is also reduced, and as a result, the accuracy of the speed of the object is also reduced.

However, as described above, the object recognition device D and the object recognition method in the embodiment correct the size by the second size correction unit 26 before obtaining the recognition data, so that the size of the object is caused by the difference in size. It is possible to suppress the deterioration of the detection accuracy of the, the detection accuracy of the position, and the detection accuracy of the speed.

The object recognition device D and the object recognition method correct the positions of the recent contacts having high detection accuracy so that they are the same before and after the correction by the first size correction unit 24. Therefore, the position accuracy is corrected after the correction. Can be maintained or its decline can be suppressed. Since the object recognition device D and the object recognition method correct the positions of the recent contacts so that they are the same before and after the correction by the second size correction unit 26, the position accuracy is maintained or deteriorated after the correction. Can be suppressed.

In the object recognition device D and the object recognition method, the ratio of each length in each partial side in each quadrant in the nearest edge, which tends to have high detection accuracy, is corrected before and after the correction by the first size correction unit 24. Since the correction is made so that they are the same, the position accuracy can be maintained or the deterioration thereof can be suppressed after the correction. The object recognition device D and the object recognition method correct the ratio of each length in each partial side of each quadrant in the nearest quadrant so as to be the same before and after the correction by the second size correction unit 26. Therefore, the position accuracy can be maintained or its deterioration can be suppressed after the correction.

If the rectangular frame of the prediction data exists beyond the detection range, the size of the object in the observation data may be missing in the portion outside the detection range. In the object recognition device D and the object recognition method, the ratio of each length in each partial side of the detection range in the nearest edge of the rectangular frame of the prediction data is corrected by the first size correction unit 24 with the prediction data. Since the size of the object in the observation data is corrected so that it becomes the same as the observation data later, the defect can be compensated, and the position accuracy can be maintained or the deterioration thereof can be suppressed after the correction. In the object recognition device D and the object recognition method, the ratio of each length in each partial side of the detection range in the nearest edge of the rectangular frame of the prediction data is corrected by the second size correction unit 26 with the prediction data. Since the size of the object in the observation data is corrected with the observation data later, the defect can be compensated, and the position accuracy can be maintained or the decrease thereof can be suppressed after the correction.

Since the object recognition device D and the object recognition method replace the size selected from a plurality of predefined sizes with the size of the object in the observation data, it is possible to suppress the fluctuation of the size due to the detection omission of the detection point.

The second size correction unit 26 may correct according to the size of the smaller object. Above The object recognition device D and the object recognition method select the larger size of the object size in the observation data and the object size in the prediction data before the correction by the second size correction unit 26. Since the size of the object in the recognition data is replaced so that the selected size is the size of the object in the recognition data, the above case can be avoided and the recognition data having an appropriate object size can be obtained.

Since the object recognition device D and the object recognition method obtain the speed of the object in the recognition data based on the position of the object in the observation data and the position and speed of the object in the prediction data, the position of the object in which the deterioration of the detection accuracy is suppressed can be obtained. Based on this, it is possible to suppress a decrease in speed detection accuracy.

In the above-described embodiment, the explanation is given in two dimensions, but the same can be explained in three dimensions. In this case, the cluster frame is a rectangular parallelepiped, and the line segment included in the rectangle as the cluster frame is a plane.

Further, in the above-described embodiment, the object state information of the observation data, the prediction data, and the recognition data includes the size, the position, and the velocity, but may further include the acceleration. In this case, at time _tk , the acceleration of the object in the observation data is a _ok , the acceleration of the object in the prediction data is a _pk , the acceleration of the object in the recognition data is a _sk , and a predetermined third parameter. When is γ, the position x _sk and the velocity v _sk of the object in the recognition data are obtained by the above equations 1 and 2, and the acceleration a _sk of the object in the recognition data is a _sk = a _pk + γ /. It is obtained by T ² × (x _ok − x _pk ). In this case, the position x _pk , velocity v _pk and acceleration a _pk of the object in the prediction data are x _pk = x _sk-1 + T x v _sk-1 + (T ^2/2 ) x a _sk-1 , v _pk . = V _sk-1 + T × a _sk-1 , a _pk = a _sk-1 . The third parameter γ is a gain with respect to the prediction dependence of acceleration in the so-called α-β-γ filter, and is appropriately set.

In order to express the present invention, the present invention has been appropriately and sufficiently described through the embodiments with reference to the drawings described above, but those skilled in the art can easily change and / or improve the above-described embodiments. It should be recognized that it is possible. Therefore, unless the modified or improved form implemented by a person skilled in the art is at a level that deviates from the scope of rights of the claims stated in the claims, the modified form or the improved form is the scope of rights of the claims. It is interpreted to be included in.

D Object recognition device FR (FRb1 to FRb8, FRa1 to FRa8) Cluster frame CP (CP _b1 to CP _b8 , CP _a1 to CP _a8 ) Central position (example of representative point)
CS (CP _b1 to CP _b3 , CP _b5 , CP _b7 , CP _a1 to CP _a3 , CP _a5 , CP _a7 ) Recent contacts R (RF, RRF, RRB, RLF, RLB) Detection range 1-1 to 1-5 1st to 5th detection unit 2 Control processing unit 3 Storage unit 21 Control unit 22 Clustering unit 23 Replacement unit 24 1st size correction unit 25 Correlation unit 26 2nd size correction unit 27 Recognition unit 28 Prediction unit 29 Recognition data utilization unit (ex) .Collision judgment unit)

Claims (6)

A clustering unit that obtains observation data that is object state information including the size and position of the object by clustering one or a plurality of detection points obtained by detecting an object around the own machine within a predetermined detection range. When,
A prediction unit that obtains prediction data that predicts the object state information based on past observation data,
A correlation unit that determines whether or not the object represented by the observation data and the object represented by the prediction data are the same by the correlation processing between the observation data and the prediction data.
Before performing the correlation processing in the correlation unit, the size of the object in the observation data and the size of the object in the prediction data are matched so that the size of the object in the observation data and the size of the object in the prediction data match. It is equipped with a size correction unit that corrects at least one of them.
Object recognition device. The size of the object is the size of the frame surrounding the area where the object exists.
The size correction unit corrects the position of the nearest contact point closest to the own machine in the frame so that it is the same before and after the correction.
The object recognition device according to claim 1. The size of the object is the size of a rectangular frame surrounding the area where the object exists.
When the size correction unit sets a predetermined first coordinate axis, a second coordinate axis orthogonal to the first coordinate axis, and a coordinate system having the center position of the own machine as the coordinate origin, the sides of the rectangular frame are set. When intersecting with the first coordinate axis or the second coordinate axis, the first length of the first partial side extending to one quadrant and the second portion extending to the other quadrant at the closest tangent to the own machine at the intersecting side. Correct so that the ratio to the second length of the side is the same before and after the correction.
The object recognition device according to claim 1 or 2. The size of the object is the size of a rectangular frame surrounding the area where the object exists.
When the rectangular frame of the prediction data exists beyond the detection range, the size correction unit exceeds the detection range at the second nearest contact side closest to the own machine at the side intersecting the detection range. The observation so that the ratio of the third length of the extending third partial side to the fourth length of the fourth partial side extending within the detection range is the same in the predicted data and the corrected observation data. Correct the size of the object in the data,
The object recognition device according to any one of claims 1 to 3. Before correction by the size correction unit, one size is selected from a plurality of predefined sizes based on the size of the object in the observation data obtained by the clustering unit, and the selected size is used as the observation data. Further includes a replacement part that replaces the size and position of the object in the observation data so that the position obtained from the selected size is the position of the object in the observation data.
The object recognition device according to any one of claims 1 to 4. A clustering step of obtaining observation data which is object state information including the size and position of the object by clustering one or a plurality of detection points obtained by detecting an object around the own machine within a predetermined detection range. When,
A prediction process for obtaining prediction data that predicts the object state information based on past observation data, and
A correlation processing step of determining whether or not the object represented by the observation data and the object represented by the prediction data are the same by the correlation processing between the observation data and the prediction data.
Before performing the correlation processing in the correlation processing step, the size of the object in the observation data and the size of the object in the prediction data so that the size of the object in the observation data and the size of the object in the prediction data match. A size correction step for correcting at least one of them is provided.
Object recognition method.

Images