JP5630249B2 - Object recognition device - Google Patents

Description

  The present invention relates to an object recognition apparatus that recognizes a moving object.

  For example, Patent Document 1 discloses a pedestrian recognition device that can distinguish and recognize a stationary object and a moving object. In this pedestrian recognition device, the position of a reflecting object (that is, a reflection point) is detected by a laser radar, and the position on an absolute coordinate system divided into a plurality of meshes is determined. Subsequently, when the position data of the reflection point is continuously present in the same mesh when the number of detection periods by the laser radar reaches a predetermined number (for example, 4 times), a stationary object is located on the mesh. By deleting the position data on the assumption that it is to be performed, only the position data of the moving object is left in the absolute coordinate system. And a pedestrian is recognized by grouping the adjacent things among the positional data regarded as the thing of a moving object.

JP 2008-26997 A

  However, in the pedestrian recognition device disclosed in Patent Document 1, since the position data of the moving object can be specified only after the position data of the stationary object is specified, the stationary object is used to specify the moving object. More time than the time it takes to specify is required. In addition to this, in order to specify the position data of the stationary object, it is necessary to detect the position of the reflection point by the laser radar a plurality of times, for example, four times. Therefore, the position data of the moving object is specified to recognize the pedestrian. There was a problem that it took time. The reason for detecting the position of the reflection point by the laser radar a plurality of times, for example, four times, is considered to suppress the influence of erroneous detection and improve the accuracy of pedestrian recognition.

  The present invention has been made in view of the above-described conventional problems, and an object of the present invention is to provide an object recognition device that can recognize a moving object accurately and more quickly. .

In the object recognition device according to claim 1 or 2 , the coordinate position of the object in the absolute coordinate system determined based on the results of the transmission of the exploration wave and the reception of the reflected wave by the distance measuring sensor is used as the absolute coordinate system. The mapping means sequentially assigns the partitions divided into the size partition units. Then, the occupancy probability in each section at the predetermined time point (the actual object in each section is actually determined) is determined according to the information on whether or not the coordinate position is assigned by the mapping means at each of the predetermined time point and the past past The ratio of the certainty that has continued to exist) is calculated, and the object corresponding to the coordinate position assigned to the section having the occupation probability within a predetermined range is recognized as a moving object.

Occupancy probability is the percentage of certainty that an object in each section has actually existed, so the higher the occupancy probability, the more likely that the object will remain stationary in that section, and the lower the occupancy probability, the more Is not likely to exist in the compartment. When the occupation probability is intermediate between them, there is a high possibility that the object has moved to the section. According to the configurations of claims 1 and 2 , since an object corresponding to a coordinate position assigned to a section having an occupation probability within a predetermined range is recognized as a moving object, the object is assigned to a section having a high possibility that the object has moved. It becomes possible to recognize the object corresponding to the given coordinate position more accurately as a moving object.

  Since the occupation probability is determined according to information on whether or not the coordinate position is assigned by the mapping means between the predetermined time point and the latest past, the mapping position assignment process by the mapping means is performed at most twice. Thus, the occupation probability can be calculated, and the time required for recognition of the moving object can be further reduced. In addition, since recognition of a stationary object first is not an essential condition for recognizing a moving object, it may not require more time to identify a moving object than it takes to identify a stationary object. Can be done.

In the configuration of claim 3 , information indicating whether or not the coordinate position is assigned by the mapping means at a predetermined time (hereinafter, assignment information) and likelihood of whether or not the coordinate position is assigned by the mapping means are determined in advance. Based on the obtained likelihood information and the occupation probability for the most recent past of the predetermined time as a prior probability in Bayesian estimation, the occupation probability in each block at the predetermined time is calculated using Bayesian estimation. Will do.

In Bayesian estimation, the target probability can be accurately obtained with a small number of trials by multiplying the prior probability by a likelihood function. According to the configuration of claim 3 , since the occupation probability is calculated using Bayesian estimation based on the allocation information and likelihood information at a predetermined time point and the occupation probability for the most recent past at the predetermined time point, The occupation probability in each section at the time of can be calculated quickly and accurately. As a result, the moving object can be recognized more accurately and more quickly.

In the configuration of claim 1 , among the sections to which coordinate positions are assigned, those that are adjacent to each other are grouped by a grouping means, and the occupation probability (group occupation probability) for the group is calculated, and the group occupation probability is calculated. Is recognized as a moving object corresponding to a set of coordinate positions included in a group within a predetermined range. According to this, when the object from which the reflected wave is obtained is a part of a certain object, it is possible to handle the data of the constituent parts of the certain object in a group.

In the configuration of claim 1 , among the groups having the group occupation probability within a predetermined range, afterimage sections detected by the afterimage section detection means (sections to which coordinate positions have been assigned in the past) are adjacent by a predetermined number or more. For a certain group, an object corresponding to a set of coordinate positions included in the group is recognized as a moving object. If the section to which the coordinate position is assigned is adjacent to the afterimage section, there is a high possibility that the object corresponding to the coordinate position has moved from the afterimage section. An object corresponding to a set of coordinate positions included in an adjacent group is likely to be a moving object. Therefore, according to the above configuration, it is possible to further increase the accuracy of recognition of moving objects.

In the configuration according to claim 2 , among the sections having the occupation probability within a predetermined range, for a section where a predetermined number of afterimage sections detected by the afterimage section detecting means are adjacent, the coordinate position assigned to the section is supported. The object to be recognized is recognized as a moving object. If the section to which the coordinate position is assigned is adjacent to the afterimage section, there is a high possibility that the object corresponding to the coordinate position has moved from the afterimage section. An object corresponding to a coordinate position assigned to an adjacent section is highly likely to be a moving object. Therefore, also with the above configuration, it is possible to further increase the accuracy of recognition of moving objects, as in the configuration of claim 1 .

According to the configuration of claim 4 , since it becomes possible not to detect the section corresponding to the undetectable area that is the area where the exploration wave of the distance measuring sensor is unreachable, it corresponds to the undetectable area. It is possible to prevent erroneous recognition of a moving object that occurs when a section to be detected is detected as an afterimage section.

In the configuration of claim 5 , the object is a moving object or a stationary object depending on whether the coordinate position is assigned to a section where the occupation probability exceeds a predetermined range or the coordinate position is assigned to a section where the occupation probability exceeds a predetermined range. It will be recognized by the recognition means. Therefore, the distinction between the moving object and the stationary object is performed at the same timing, and the recognition of the moving object can be performed at the same time as that required for the recognition of the stationary object.

1 is a block diagram showing a schematic configuration of an object recognition system 100. FIG. It is a flowchart which shows the operation | movement flow regarding recognition of the moving body in ECU2. It is a schematic diagram for demonstrating calculation of the occupation probability of a group. It is a schematic diagram for demonstrating discrimination | determination with a movement candidate group and a stationary group. It is a schematic diagram for demonstrating the detection of an afterimage cell. It is a figure for demonstrating the comparison of the effect of a prior art and this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of an object recognition system 100 to which the present invention is applied. An object recognition system 100 shown in FIG. 1 is mounted on a vehicle and includes a distance measuring sensor 1 and an ECU (electronic control unit) 2. In the present embodiment, the object recognition system 100 detects a moving object located around the vehicle, particularly in front of the vehicle.

  The distance measuring sensor 1 is a laser radar, for example, and includes a light emitting unit 11, a light receiving unit 12, and a time measuring unit 13. The light emitting unit 11 and the light receiving unit 12 are provided at the front of the host vehicle so that an object existing in front of the host vehicle can be detected.

  The light emitting unit 11 includes a laser diode, and sweeps and scans (pulses) pulsed laser light in a range of predetermined angles in the vehicle width direction and the vehicle height direction according to a drive signal from the ECU 2. A detection area of the laser radar is defined by an angle range in which the laser beam is irradiated in each of the vehicle width direction and the vehicle height direction.

  The light receiving unit 12 receives reflected light, which is the laser light emitted from the light emitting unit 11 and reflected by an object, by a light receiving lens, and supplies the light receiving element (photodiode). The light receiving element outputs a voltage corresponding to the intensity of the reflected light. The output voltage of the light receiving element is amplified by an amplifier and then output to a comparator. The comparator compares the output voltage of the amplifier with the reference voltage, and outputs a predetermined light reception signal to the time measuring unit 13 when the output voltage becomes larger than the reference voltage.

  A drive signal output from the ECU 2 to the light emitting unit 11 is also input to the time measuring unit 13. The time measuring unit 13 measures the time from when the drive signal is output until the light receiving signal is generated, that is, the time difference between the time when the laser light is emitted and the time when the reflected light is received, and information on this time difference (hereinafter referred to as the time difference information). , Measurement time information) is input to the ECU 2.

  When the light emitting unit 11 emits laser light to detect an object that actually exists in front of the host vehicle, the laser light scans sequentially in a predetermined pattern within the irradiation area (that is, the detection area). As described above, a drive signal is output from the ECU 2 to the light emitting unit 11. Thus, since the pattern scanned by the laser light is determined, when the reflected light is received, the irradiation angle of the laser light from which the reflected light is obtained is uniquely determined.

  The ECU 2 is composed mainly of a microcomputer composed of a CPU, ROM, RAM, backup RAM, etc., and executes various processes by executing various control programs stored in the ROM based on input information. To do. The ECU 2 corresponds to the object recognition device in the claims.

  When the measurement time information (that is, the time difference between the laser beam emission time and the reflected light reception time) is input from the time measurement unit 13 of the distance measurement sensor 1, the ECU 2 determines the distance to the object based on the time difference. calculate. The ECU 2 creates position data of the object (specifically, the reflection point of the laser light) based on the calculated distance to the object and the irradiation angle of the laser light from which the reflected light is obtained. Specifically, the center of the light emitting unit 11 and the light receiving unit 12 is the origin (0, 0, 0), the vehicle width direction is the X axis, the vehicle longitudinal direction is the Y axis, and the height direction is the Z axis. The coordinates, Y coordinate, and Z coordinate are obtained. That is, the relative position of the object with respect to the center of the light emitting unit 11 and the light receiving unit 12 (that is, corresponding to the position of the host vehicle) is obtained.

  The ECU 2 further receives sensor signals from a vehicle speed sensor that detects the traveling speed of the host vehicle, a yaw rate sensor that detects the magnitude of the yaw rate acting on the host vehicle, and a steering angle sensor that detects the steering angle of the steering wheel. The The ECU 2 uses the sensor signals from these sensors to calculate the amount of movement of the host vehicle.

  Here, the operation flow related to the recognition of the moving object in the ECU 2 will be described with reference to FIG. FIG. 2 is a flowchart showing an operation flow related to recognition of a moving object in the ECU 2. This flow is started when, for example, an IG switch or an ACC switch of the host vehicle is turned on and power is supplied to the object recognition system 100. Also, this flow ends when the supply of power to the object recognition system 100 is stopped.

  This flow is started when the speed of the host vehicle becomes higher than a predetermined value such as 5 km / h based on the sensor signal of the vehicle speed sensor (speed of detection limit of the vehicle speed sensor). It is good also as a structure which is complete | finished when this speed becomes below a predetermined value.

  First, in step S1, an observation process is performed and the process proceeds to step S2. In the observation process, a driving signal is output to the distance measuring sensor 1, the laser light is swept toward the detection area, and input of measurement time information output from the distance measuring sensor 1 is accepted. The measurement cycle by the distance measuring sensor 1 is set to 100 msec, for example.

  In step S2, a mapping process is performed and the process proceeds to step S3. In the mapping process, the distance to the reflection point is calculated based on the measurement time information input from the distance measuring sensor 1 (that is, the time difference between the emission time of the laser beam and the reception time of the reflected light), and as described above, Create position data of reflection points in the coordinate system.

  In the mapping process, the movement amount of the own vehicle is calculated using the vehicle speed, yaw rate, steering angle, etc. of the own vehicle, and the own vehicle position in the absolute coordinate system is calculated based on the calculated movement amount of the own vehicle. The point that is the origin of the absolute coordinate system can be arbitrarily set. For example, the center of the light emitting unit 11 and the light receiving unit 12 at the time when measurement by the distance measuring sensor 1 is started (t = 0) is set as the origin. In the absolute coordinate system, the vehicle width direction is the X axis, the vehicle longitudinal direction is the Y axis, and the height direction is the Z axis.

The calculation of the position of the host vehicle in the absolute coordinate system may be performed based on the movement amount of the host vehicle calculated using the front and rear two-wheel model, for example, as disclosed in Patent Document 1. . In addition to calculating the movement amount of the own vehicle by applying it to a vehicle model such as a front and rear two-wheel model, the self-position is estimated by scan matching using position data of a fixed object obtained by measurement by the distance measuring sensor 1. The vehicle may be configured to calculate using the wheel speed difference of the four wheels of the own vehicle (for example, in the case of a four-wheel vehicle), information on the position change of the own vehicle detected by GPS,
Then, after determining the position of the vehicle in the absolute coordinate system, the position of each reflection point in the absolute coordinate system is obtained using the position data (coordinate position) of the sensor coordinate system. That is, the reflection point position data in the sensor coordinate system is converted into position data in the absolute coordinate system. Specifically, in order to convert the coordinate position of the reflection point in the sensor coordinate system to the coordinate position in the absolute coordinate system, the vehicle front-rear direction (Y-axis) in the sensor coordinate system is the traveling direction of the host vehicle in the absolute coordinate system. While rotating the sensor coordinate system so as to match, the vehicle width direction (X-axis), the vehicle longitudinal direction (Y-axis), and the height direction (Z-axis) based on the coordinate position in the absolute coordinate system of the host vehicle What is necessary is just to apply a three-dimensional coordinate to an absolute coordinate system. Therefore, step S2 corresponds to the absolute position determining means in the claims.

The absolute coordinate system is divided into grid cells (for example, 50 cm ³ ) having a certain size, and the coordinate position converted into the coordinate position in the absolute coordinate system is assigned to the grid cell corresponding to the coordinate position. become. Therefore, step S2 corresponds to the mapping means of the claims, and the grid cell corresponds to the section of the claims. Hereinafter, for the sake of convenience, the description will be made using grid cells arranged in two planes in the vehicle width direction (X axis) and the vehicle longitudinal direction (Y axis). The same applies to grid cells arranged in two planes in the vertical direction (Z-axis), the vehicle longitudinal direction (Y-axis), and the height direction (Z-axis).

  In step S3, a grouping process is performed and the process proceeds to step S4. In the grouping process, among the grid cells to which coordinate positions are assigned in step S2, grid cells adjacent to each other are grouped together. Therefore, step S3 corresponds to grouping means in claims.

  In step S4, the occupation probability of each grid cell in the absolute coordinate system is calculated. Therefore, step S4 corresponds to the occupation probability calculation means in the claims. The occupation probability of the grid cell is calculated using Bayesian estimation depending on whether or not a coordinate position is assigned to the grid cell. Hereinafter, a grid cell to which a coordinate position is assigned is referred to as an occupied cell, and a grid cell to which no coordinate position is assigned is referred to as an unoccupied cell.

For the occupied cell, the occupation probability is calculated from the following equation (1). In the equation (1), p (x _t | z _t ) is a reflection point that is actually applied to a grid cell when a coordinate position is assigned to the grid cell (that is, a reflection point in the grid cell is observed). Indicates the probability of existence. p (x _t ) is a prior probability in Bayesian estimation, and indicates the probability that a reflection point actually exists in the grid cell. p (x _−t ) indicates the probability that a reflection point does not actually exist in the grid cell, and is expressed as p (x _−t ) = 1−p (x _t ). It is assumed that the following symbol 1 originally applies to the part of x− _{t, and} the same applies to sentences other than the following mathematical expressions.

また、ｐ（ｘ_ｔ）は、過去の時点についての占有確率が算出済みの場合には、ｐ（ｘ_ｔ）＝ｐ（ｘ_ｔ−１｜ｚ_ｔ−１）となるが、測距センサ１による計測を開始した時点（ｔ＝０）のように、過去の時点についての占有確率が算出済みでない場合には、確率は１／２であるのでｐ（ｘ₀）＝０．５となる。さらに、ｐ（ｚ_ｔ｜ｘ_ｔ）およびｐ（ｚ_ｔ｜ｘ_―ｔ）は予め定められた定数であって、ｐ（ｚ_ｔ｜ｘ_ｔ）はベイズ推定における尤度にあたる。ｐ（ｚ_ｔ｜ｘ_ｔ）は、物体が存在するときに物体（反射点）が観測される確率に相当し、ｐ（ｚ_ｔ｜ｘ_―ｔ）は物体が存在しないときに物体（反射点）が観測される確率に相当する。ｐ（ｚ_ｔ｜ｘ_ｔ）およびｐ（ｚ_ｔ｜ｘ_―ｔ）は、測距センサ１の測定精度に応じて設定すればよく、本実施形態では、ｐ（ｚ_ｔ｜ｘ_ｔ）＝０．８、ｐ（ｚ_ｔ｜ｘ_―ｔ）＝０．１とする。なお、０．０００１≦ｐ（ｘ_ｔ｜ｚ_ｔ）≦０．９９９９であるものとする。

Further, p (x _t ) is p (x _t ) = p (x _t−1 | z _t−1 ) when the occupation probability for the past time point has been calculated. When the occupancy probability for the past time point has not been calculated, such as the time point when the measurement by (1) is started (t = 0), the probability is ½, so p (x ₀ ) = 0.5. Further, p (z _t | x _t ) and p (z _t | x _−t ) are predetermined constants, and p (z _t | x _t ) is a likelihood in Bayesian estimation. p (z _t | x _t ) corresponds to the probability that an object (reflection point) is observed when the object is present, and p (z _t | x _−t ) is the object (reflection point when the object is not present). ) Corresponds to the probability of being observed. p (z _t | x _t ) and p (z _t | x _−t ) may be set according to the measurement accuracy of the distance measuring sensor 1. In this embodiment, p (z _t | x _t ) = 0. .8, p (z _t | x _−t ) = 0.1. Note that 0.0001 ≦ p (x _t | z _t ) ≦ 0.9999.

For unoccupied cells, the occupancy probability is calculated from the following equation (2). In equation (2), p (x _t | z _−t ) is actually assigned to a grid cell when no coordinate position is assigned to the grid cell (ie, a reflection point in the grid cell is observed). The probability that a reflection point exists is shown. p (x _t ) and p (x _−t ) are the same as described in the equation (1). It is assumed that the following symbol 2 originally applies to the part of pz- _{t, and} the same applies to sentences other than the following mathematical expressions.

また、ｐ（ｚ_―ｔ｜ｘ_ｔ）およびｐ（ｚ_―ｔ｜ｘ_―ｔ）は予め定められた定数であって、ｐ（ｚ_―ｔ｜ｘ_ｔ）はベイズ推定における尤度にあたる。ｐ（ｚ_―ｔ｜ｘ_ｔ）は、物体が存在するときに物体（反射点）が観測されない確率に相当し、ｐ（ｚ_―ｔ｜ｘ_―ｔ）は物体が存在しないときに物体（反射点）が観測されない確率に相当する。ｐ（ｚ_―ｔ｜ｘ_ｔ）はｐ（ｚ_―ｔ｜ｘ_ｔ）＝１−ｐ（ｚ_ｔ｜ｘ_ｔ）と表され、ｐ（ｚ_―ｔ｜ｘ_―ｔ）は、ｐ（ｚ_―ｔ｜ｘ_―ｔ）＝１−ｐ（ｚ_ｔ｜ｘ_―ｔ）と表される。本実施形態では、ｐ（ｚ_―ｔ｜ｘ_ｔ）＝０．２、ｐ（ｚ_―ｔ｜ｘ_―ｔ）＝０．９となる。なお、０．０００１≦ｐ（ｘ_ｔ｜ｚ_―ｔ）≦０．９９９９であるものとする。

_{Further, p (z -t | x t} ) and _p (z _{-t | x} -t) is a predetermined _{constant, p (z -t | x t} ) is equivalent to the likelihood in Bayesian estimation. p (z _−t | x _t ) corresponds to the probability that an object (reflection point) will not be observed when an object exists, and p (z _−t | x _−t ) represents an object (reflection) when no object exists. This corresponds to the probability that a point) is not observed. p _(z -t | _{x t)} is _p expressed as _{| | (x t z t)} , p (z -t x t) = 1-p (z -t | x -t) is, p (z _{- t} | x _−t ) = 1−p (z _t | x _−t ) In this embodiment, p (z _−t | x _t ) = 0.2 and p (z _−t | x _−t ) = 0.9. Note that 0.0001 ≦ p (x _t | z _−t ) ≦ 0.9999.

  When a grid cell corresponds to an occlusion area, the occupation probability is not calculated for the grid cell. The occlusion region is a region where the laser beam cannot reach, and indicates a region outside the detection area of the distance measuring sensor 1 or a region corresponding to the shadow of the object. The region outside the detection area may be configured to be specified by the ECU 2 based on the scan range. The area corresponding to the shadow of the object is identified by estimating the range where the laser beam is blocked by the object reflecting the laser beam from the irradiation angle of the laser beam when the reflected wave is obtained. That's fine. Therefore, the ECU 2 corresponds to the non-detectable area determination means in the claims. When the grid cell corresponds to the occlusion area, the occupation probability may be set to 0.5.

  Whether or not a grid cell corresponds to an occlusion area may be determined to be an occlusion area when a part of the grid cell includes an occlusion area, or may be a predetermined value such as a half or more of the grid cell. A configuration may be adopted in which an occlusion area exceeding the range is included and determined to be an occlusion area.

  In the present embodiment, a configuration is shown in which the occupation probability of each grid cell in the absolute coordinate system is calculated after performing the grouping process, but the present invention is not limited to this. For example, the grouping process may be performed after calculating the occupation probability of each grid cell in the absolute coordinate system.

  In step S5, the group occupation probability (hereinafter, group occupation probability) is calculated, and the process proceeds to step S6. Therefore, step S5 corresponds to the group occupation probability calculating means in the claims. The group occupation probability is calculated according to the specific gravity of the coordinate position based on the occupation probability in each grid cell calculated in step S4 and the number of coordinate positions assigned to each grid cell in step S2.

  Here, a specific example of calculating the occupation probability of the group will be shown using FIG. FIG. 3 is a schematic diagram for explaining calculation of the group occupation probability. Each cell in the figure indicates a grid cell, and the number in each cell indicates the occupation probability of each grid cell. A grid cell group surrounded by a broken line indicates a group (hereinafter referred to as a group BCDE), and a set of points indicated by a frame A indicates a coordinate position group included in the group BCDE.

  In the example of FIG. 3, the grid cell indicated by B includes five coordinate positions and the occupation probability is 0.1, and the grid cell indicated by C includes two coordinate positions and the occupation probability is 0. 3 and D, the grid cell indicated by D has one coordinate position and the occupation probability is 0.2, and the grid cell indicated by E has one coordinate position and the occupation probability is 0.5. In this case, the group occupation probability is (0.1 × 5 + 0.3 × 2 + 0.2 × 1 + 0.5 × 1) /9=0.2.

  In step S6, based on the group occupation probability calculated in step S5, it is determined whether or not it is a movement candidate group (that is, determination of a movement candidate group and a stationary group). Specifically, if the group occupancy probability is equal to or greater than a predetermined lower limit threshold and equal to or lower than a predetermined upper limit threshold (that is, a predetermined range), it is determined as a movement candidate group (YES in step S6), and step S7 is performed. Move on. On the other hand, if the group occupancy probability is greater than the predetermined upper limit threshold (that is, exceeds the predetermined range), it is determined as a stationary group (NO in step S6), and the process proceeds to step S10. The predetermined upper limit threshold and the predetermined lower limit threshold are values that can be arbitrarily set.

  Here, a specific example of discrimination between a movement candidate group and a stationary group will be described with reference to FIG. FIG. 4 is a schematic diagram for explaining discrimination between a movement candidate group and a stationary group. In the example of FIG. 4, it is assumed that the predetermined upper limit threshold is 0.92, and the predetermined lower limit threshold is 0.002. 4 is the group BCDE shown in FIG. 3, and the group occupation probability is 0.2.

  In the example of FIG. 4, when the group occupation probability is in the range of 0.002 or more and 0.92 or less, it is determined as a movement candidate group, and when the group occupation probability is greater than 0.92, it is determined as a stationary group. Is done. Since the group occupation probability of the group BCDE is 0.2, the group BCDE is determined to be a movement candidate group in the example of FIG.

  In step S7, an afterimage cell search process is performed and the process proceeds to step S8. In the afterimage cell search process, among the grid cells in the absolute coordinate system, the grid cells that do not correspond to the above-described occlusion area and to which no coordinate position is assigned in step S2 are equal to or greater than a predetermined lower limit threshold value. In addition, a grid cell that is not more than a predetermined upper limit threshold (that is, a predetermined range) is detected as an afterimage cell. Therefore, step S7 corresponds to the afterimage section detection means in the claims.

  Here, a specific example of afterimage cell detection will be described with reference to FIG. FIG. 5 is a schematic diagram for explaining detection of an afterimage cell. A group composed of grid cells indicated by B to E shown in FIG. 5 is a group BCDE shown in FIG. In the example of FIG. 5, it is assumed that the occupation probability of grid cells other than B to H is 0.001, except that the grid cells indicated by F to H have an occupation probability of 0.4. In addition, the predetermined upper limit threshold and the predetermined lower limit threshold are the same values as those used for discrimination between the movement candidate group and the stationary group. In the example of FIG. 5, it is assumed that there is no grid cell corresponding to the occlusion area.

  In the example of FIG. 5, among the grid cells to which no coordinate position is assigned, those having an occupation probability in the range of 0.002 or more and 0.92 or less are grid cells indicated by F to H. The grid cells indicated by F to H are detected as afterimage cells.

  The predetermined upper limit threshold and the predetermined lower limit threshold used for detecting the afterimage cell are different from the predetermined upper limit threshold and the predetermined lower limit threshold used for distinguishing the movement candidate group from the stationary group. It is good also as composition to do. In addition, since the occupancy probability of the grid cell to which the coordinate position is not assigned in step S2 does not increase, the afterimage cell search process does not correspond to the aforementioned occlusion area from among the grid cells of the absolute coordinate system, and A grid cell that is not assigned a coordinate position in step S2 and that is equal to or greater than a predetermined lower limit threshold may be detected as an afterimage cell.

  In step S8, it is determined whether or not there are two or more (that is, a plurality of) afterimage cells in the vicinity of the movement candidate group. If there are a plurality (YES in step S8), the process proceeds to step S9. If there is not a plurality (NO in step S8), the process proceeds to step S10. The vicinity mentioned here is, for example, within a predetermined number of grid cells from the movement candidate group. For example, it may be within the range of several grid cells from the movement candidate group, or may be within the range of one grid cell from the movement candidate group (that is, adjacent to the movement candidate group).

  In step S9, a moving object determination process is performed, the process returns to step S1, and the flow is repeated. In the moving object determination process, a set of coordinate positions included in a movement candidate group in which a plurality of afterimage cells are adjacent to each other is determined as a moving object. The moving object determination process may be configured to determine the type of moving object such as a pedestrian according to the size of the range occupied by the set of coordinate positions.

  For example, in the example shown in FIG. 5, since there are three afterimage cells F to H adjacent to the group BCDE that is the movement candidate group, a set of coordinate positions included in the group BCDE is determined to be a moving object. become. In the present embodiment, a configuration is described in which when a plurality of adjacent afterimage cells exist, a set of coordinate positions included in the movement candidate group is determined as a moving object, but the present invention is not necessarily limited thereto. For example, when there is even one afterimage cell that is adjacent, a set of coordinate positions included in the movement candidate group may be determined as a moving object.

  In step S10, a stationary object determination process is performed, the process returns to step S1, and the flow is repeated. In the stationary object determination process, a set of coordinate positions included in the group is determined as a stationary object. Therefore, Steps S8 to S10 correspond to recognition means in the claims.

When this flow is repeated and the occupation probability of each grid cell at the most recent past time has already been calculated at a predetermined time (for example, the current time), the occupation at the most recent past time is obtained in step S4. Using the probability p (x _t-1 | z _t-1 ) as the current prior probability p (x _t ), the occupation probability in each grid cell at the current time is calculated.

Using the occupation probability p (x _t-1 | z _t-1 ) of the most recent past time as the prior probability p (x _t ) at the present time, the occupation probability in each grid cell at the present time is calculated. As is clear from the equation (1), the grid cell in which the object continues to exist has an occupancy probability close to 1. In addition, the grid cell in which the state in which the object does not exist continues to have a value close to 0, and the grid cell in which the state in which the object exists has changed, the intermediate value between them (the object continues to exist) Smaller value than the grid cell).

  Thus, the occupation probability is a ratio of the certainty that the object in each grid cell has actually existed, and according to the configuration of the present embodiment, the occupation probability is assigned to a grid cell in a predetermined range. Since the object corresponding to the coordinate position is recognized as a moving object, the object corresponding to the coordinate position assigned to the grid cell to which the object has a high possibility of movement can be more accurately recognized as the moving object. .

  In Bayesian estimation, the target probability can be obtained with a small number of trials by multiplying the prior probability by the likelihood function. However, according to the above configuration, the allocation information and likelihood at the present time and the latest Since the occupation probability is calculated using Bayesian estimation based on the occupation probability of the past, the occupation probability in each grid cell at the present time can be calculated quickly and accurately.

  Further, according to the above configuration, the grid cell corresponding to the region where the exploration wave of the distance measuring sensor 1 cannot reach (hereinafter referred to as the undetectable region) is not detected as the afterimage cell, and thus corresponds to the undetectable region. It is possible to prevent erroneous recognition of a moving object that occurs when a grid cell is detected as an afterimage cell.

  Here, the effect in this invention is demonstrated concretely using FIG. When comparing the number of misrecognitions when the pedestrian recognition is performed by the pedestrian recognition device disclosed in Patent Document 1 and the number of misrecognitions when the pedestrian is recognized by the object recognition system 100, Compared to the pedestrian recognition device disclosed in Patent Document 1, the object recognition system 100 had fewer erroneous recognitions.

  Further, in the pedestrian recognition apparatus disclosed in Patent Document 1, in order to recognize a stationary object, laser light is swept (scanned), and the coordinate position of the object is converted to a mesh (that is, a grid cell) on the absolute coordinate system. The assignment work is repeated, but it is necessary to repeat this work for seven scans in order to suppress erroneous recognition. Since a moving object is recognized after recognizing a stationary object, it takes time corresponding to 8 scans of the above work.

  On the other hand, in the present embodiment, the occupation probability is calculated using Bayesian estimation based on the allocation information and likelihood at the present time and the occupation probability for the most recent past at the present time. Recognition. In this embodiment, since the measurement accuracy of the distance measuring sensor 1 is incorporated into the likelihood in Bayesian estimation, it is possible to recognize whether the object is a moving object or a stationary object even with a small number of trials while suppressing erroneous recognition. Specifically, moving object recognition is stationary in a time corresponding to two scans of the task of assigning the coordinate position of the object to the mesh (that is, grid cell) on the absolute coordinate system by sweeping irradiation (scanning) of the laser beam. It can also recognize objects. That is, compared with the pedestrian recognition apparatus disclosed in Patent Document 1, it is possible to recognize a moving object with high accuracy and more quickly.

  In the present embodiment, the coordinate position in the absolute coordinate system of the host vehicle is rotated while rotating the sensor coordinate system so that the vehicle longitudinal direction (Y axis) of the sensor coordinate system matches the traveling direction of the host vehicle in the absolute coordinate system. By applying the three-dimensional coordinates in the vehicle width direction (X axis), the vehicle longitudinal direction (Y axis), and the height direction (Z axis) to the absolute coordinate system, the coordinate position of the reflection point in the sensor coordinate system is Although the configuration for converting to the coordinate position in the absolute coordinate system is shown, it is not necessarily limited to this. For example, while rotating the sensor coordinate system so that the vehicle front-rear direction (Y-axis) of the sensor coordinate system matches the traveling direction of the host vehicle in the absolute coordinate system, the vehicle is based on the coordinate position in the absolute coordinate system of the host vehicle. The two-dimensional coordinates in the width direction (X axis) and the vehicle front-rear direction (Y axis) may be applied to the absolute coordinate system.

  Further, in the present embodiment, it is determined whether or not there are a plurality of afterimage cells adjacent to the movement candidate group, and when there are a plurality of afterimage cells, a set of coordinate positions included in the movement candidate group is determined as a moving object. However, the present invention is not limited to this. For example, it is determined whether there are a plurality of afterimage cells adjacent to a grid cell (hereinafter referred to as a movement candidate cell) whose occupation probability is 0.002 or more and 0.92 or less. The cells may be grouped together, and a set of coordinate positions included in the group may be determined as a moving object.

  In the present embodiment, a configuration in which a laser radar is used as the distance measuring sensor 1 has been described. However, the configuration is not necessarily limited thereto. Any device such as an ultrasonic sensor or an infrared sensor that can measure the distance and position of an object based on the results of transmission of the exploration wave and reception of the reflected wave of the exploration wave can be used as the distance measurement sensor 1.

  In the present embodiment, the configuration in which the occupancy probability of the grid cell is calculated using Bayesian estimation is shown, but the present invention is not necessarily limited thereto. For example, information on whether or not a coordinate position has been assigned to a grid cell in a mapping process at a predetermined time (for example, the current time) and information on whether or not a coordinate position has been assigned to a grid cell in a mapping process in the past in the past Any other method may be used to calculate the occupancy probability of the grid cell as long as the occupancy probability determined according to the original can be estimated.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the technical means disclosed in different embodiments can be appropriately combined. Such embodiments are also included in the technical scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Distance sensor, 2 ECU (object recognition apparatus, undetectable area determination means), 11 Light emission part, 12 Light reception part, 13 Time measurement part, 100 Object recognition system, S2 Absolute position determination means, Mapping hand, S3 Grouping means, S4 occupation probability calculation means, S5 group occupation probability calculation means, S7 afterimage section detection means, S8 to S10 recognition means

Claims (5)

Mounted on the vehicle,
Based on the results of the transmission of the exploration wave and the reception of the reflected wave by the distance measuring sensor that transmits the exploration wave at a predetermined period and receives the reflected wave reflected by the object, the predetermined position is set as the origin. Absolute position determining means for sequentially determining the coordinate position of the object in the absolute coordinate system;
Each time the absolute position determination means determines the coordinate position of the object in the absolute coordinate system, a section corresponding to the coordinate position of the sections obtained by dividing the absolute coordinate system into partition units of a predetermined size, Mapping means for assigning the coordinate position, and an object recognition device comprising:
It is determined according to information on whether or not the coordinate position has been assigned by the mapping means at a predetermined time point and information on whether or not the coordinate position has been assigned by the mapping means in the immediate past of the predetermined time point, An occupancy probability calculating means for calculating an occupancy probability which is a ratio of certainty that the object in each of the sections at the predetermined time has actually existed;
Based on the occupancy probability calculated by the occupancy probability calculation means, a recognition means for recognizing the object corresponding to the coordinate position assigned to the section having the occupancy probability in a predetermined range as a moving object ;
Among the sections to which the coordinate positions are assigned by the mapping means, grouping means for grouping adjacent ones into groups,
A group occupancy probability that is the occupancy probability for the group grouped by the grouping means based on the occupancy probability in each section calculated by the occupancy probability calculating means and the number of coordinate positions assigned to each section. Group occupation probability calculating means for calculating
Based on the fact that the coordinate position is not assigned by the mapping means and the occupation probability calculated by the occupation probability calculation means is greater than a predetermined value, An afterimage section detecting means for detecting a certain afterimage section from among the sections ,
The recognizing unit is a group in which the group occupancy probability calculated by the group occupancy probability calculating unit is in a predetermined range, and the group having a predetermined number or more of afterimage sections detected by the afterimage section detecting unit, object recognition apparatus characterized that you recognizing an object corresponding to a set of coordinate positions in that group as a moving object. Mounted on the vehicle,
Based on the results of the transmission of the exploration wave and the reception of the reflected wave by the distance measuring sensor that transmits the exploration wave at a predetermined period and receives the reflected wave reflected by the object, the predetermined position is set as the origin. Absolute position determining means for sequentially determining the coordinate position of the object in the absolute coordinate system;
Each time the absolute position determination means determines the coordinate position of the object in the absolute coordinate system, a section corresponding to the coordinate position of the sections obtained by dividing the absolute coordinate system into partition units of a predetermined size, Mapping means for assigning the coordinate position, and an object recognition device comprising:
It is determined according to information on whether or not the coordinate position has been assigned by the mapping means at a predetermined time point and information on whether or not the coordinate position has been assigned by the mapping means in the immediate past of the predetermined time point, An occupancy probability calculating means for calculating an occupancy probability which is a ratio of certainty that the object in each of the sections at the predetermined time has actually existed;
Based on the occupancy probability calculated by the occupancy probability calculation means, a recognition means for recognizing the object corresponding to the coordinate position assigned to the section having the occupancy probability in a predetermined range as a moving object ;
Based on the fact that the coordinate position is not assigned by the mapping means and the occupation probability calculated by the occupation probability calculation means is greater than a predetermined value, An afterimage section detecting means for detecting a certain afterimage section from among the sections ,
The recognizing unit is configured to determine a segment having a predetermined number or more of the afterimage segments detected by the afterimage segment detecting unit among the segments having the occupation probability calculated by the occupation probability calculating unit within a predetermined range. object recognition apparatus characterized that you recognize objects as a moving object corresponding to the assigned coordinate positions on. In claim 1 or 2 ,
The occupancy probability calculating means is information on whether or not the coordinate position has been assigned by the mapping means at the predetermined time, and a likelihood that is determined in advance as to whether or not the coordinate position is assigned by the mapping means. Based on the degree information and the occupancy probability of the past in the past as the prior probability in Bayesian estimation, the occupancy probability in each section at the predetermined time point is calculated using Bayesian estimation. And
If the occupancy probability for the past has already been calculated, the calculated occupancy probability is used to calculate the occupancy probability in each section at the predetermined time point, while the occupancy probability for the past is If not already calculated, a predetermined value is used as the occupancy probability to calculate the occupancy probability in each of the sections at the predetermined time point. In any one of Claims 1-3 ,
An undetectable area determination means for determining an undetectable area that is an area where the exploration wave of the ranging sensor is unreachable
The afterimage section detecting means is determined by the undetectable area determining means in addition to the fact that the coordinate position is not assigned by the mapping means and the occupation probability calculated by the occupation probability calculating means is greater than a predetermined value. An object recognition apparatus, wherein the afterimage section is detected from among the sections based on the fact that it does not correspond to the undetectable area. In any one of Claims 1-4 ,
The recognizing unit recognizes the object as a moving object in the case of an object corresponding to the coordinate position assigned to the section having the occupation probability within a predetermined range, while being assigned to the section with the occupation probability exceeding the predetermined range. An object recognition apparatus characterized by recognizing an object corresponding to a coordinate position as a stationary object.

Images