Object detection method and object detection device

Abstract

A controller (20) estimates an object model (92) obtained by approximating an object (91) as a rectangle from a plurality of reflection points (Q) output from a peripheral detection sensor (11), calculates a point group density from the number of reflection points (Q) included in the object model (92), performs feature point base estimation processing for estimating the movement speed of the object (91) from the movement amounts of the positions of feature points using at least one of the plurality of reflection points (Q) as a feature point when the point group density is equal to or greater than a first threshold value, and calculates a center position (C) of the object model (92) and performs center position base estimation processing for estimating the movement speed of the object (91) from the movement amount of the center position (C) when the point group density is less than the first threshold value.

Description

Object detection method and object detection device

[0001] The present invention relates to an object detection method and an object detection device.

[0002] Conventionally, there is known a device for detecting another vehicle (hereinafter referred to as a cut-in vehicle) crossing in front of a vehicle (see, for example, Patent Document 1). Patent Document 1 discloses a method for detecting reflection points on an object by scanning a radar wave within a scanning range and receiving the radar wave reflected by the object. From the positions of each reflection point, a line segment capturing the outline of the vehicle is calculated, and the center position of the vehicle is calculated based on the calculated line segment. When the entire cut-in vehicle is located within the scanning range, status recognition is performed using a method using the center position. On the other hand, when the rear end position of the vehicle is located in an edge region of the scanning range, the leading end position is calculated from the center position, and status recognition is performed using the leading end position. It also discloses that when the leading end position and the center position are close to each other, status recognition may also be performed using a method using the center position.

[0003] JP 2011-196943 A

[0004] However, when a technology such as that disclosed in Patent Document 1 is applied to estimating the speed of a vehicle traveling alongside or ahead, it may not be possible to accurately switch between state recognition based on the center position and state recognition based on the leading edge position, and the accuracy of speed estimation may not be ensured. For example, in Patent Document 1, state recognition using the center position is performed for other vehicles in the edge region of the scanning range, but if only a portion of the other vehicle is detected, there is a problem in that state recognition using the center position contains errors and highly accurate object detection cannot be performed.

[0005] An object of the present invention is to provide an object detection method and an object detection device that are capable of highly accurate object detection.

[0006] In the object detection method of the present invention, a detection sensor transmits electromagnetic waves and detects reflection points at which the electromagnetic waves are reflected by an object, and a controller detects the object and estimates its movement speed based on the multiple reflection points. The controller approximates the object with a rectangle using the multiple reflection points and calculates a point cloud density from the number of reflection points included in the approximated rectangle. If the point cloud density is equal to or greater than a first threshold, the controller directly uses at least one of the multiple reflection points as a feature point and performs a feature-point-based estimation process to estimate the object's movement speed from the amount of movement of the feature point. On the other hand, if the point cloud density is less than the first threshold, the controller calculates the center position of the rectangle and performs a center-position-based estimation process to estimate the object's movement speed from the amount of movement of the center position.

[0007] In the present invention, by switching the velocity estimation method depending on the point cloud density, velocity estimation can be performed using an appropriate method according to the reliability of the reflection points, thereby enabling highly accurate object detection.

[0008] 5 is a block diagram showing a schematic configuration of an object detection device of the present embodiment. A diagram showing an example of point cloud data detected by a perimeter detection sensor and an example of an object approximated by a rectangle based on the point cloud data in the present embodiment. A diagram showing an example of dividing an approximated rectangular object model into a plurality of sections in the present embodiment. A diagram showing another example of point cloud data detected by a perimeter detection sensor. A flowchart showing an object detection method of the present embodiment. A flowchart showing an overview of the speed estimation process of step S12 in FIG.

[0009] A first embodiment of the present disclosure will now be described. Fig. 1 is a block diagram showing a schematic configuration of an object detection device 10 of this embodiment. The object detection device 10 of this embodiment is mounted on a vehicle and detects objects around the vehicle, particularly surrounding vehicles, and estimates their movement speeds. As shown in Fig. 1, the object detection device 10 includes a surrounding detection sensor 11 and a controller 20 that estimates the movement speeds of surrounding vehicles based on observation information measured by the surrounding detection sensor 11.

[0010] The surroundings detection sensor 11 is provided on the periphery of the vehicle to detect objects around the vehicle. The number and installation location of the surroundings detection sensor 11 are not particularly limited. For example, multiple surroundings detection sensors 11 may be provided to cover the front, rear, and sides of the vehicle so as to detect objects within a 360° range around the vehicle. The surroundings detection sensor 11 transmits electromagnetic waves and receives the electromagnetic waves reflected by objects, thereby detecting the reflection points of the electromagnetic waves on the objects. Examples of such surroundings detection sensors 11 include light detection and ranging (LiDAR), millimeter-wave radar, and radar, and a combination of these may also be used. The surroundings detection sensor 11 may also be configured to include a depth estimation sensor. When depth estimation is provided, a host vehicle movement amount detection device that estimates the host vehicle movement amount may be further added to cancel out the host vehicle movement amount. Examples of host vehicle movement amount detection devices include a device that measures the host vehicle's position, such as a global navigation satellite system (GNSS), an inertial measurement unit (IMU) that measures acceleration and angular velocity, and a wheel encoder. Furthermore, when a depth estimation sensor is used, the amount of movement may be estimated using SLAM (Simultaneous Localization and Mapping).

[0011] The surroundings detection sensor 11 measures the position of the reflection point reflected by the object based on the time from the timing of transmitting the electromagnetic wave to the timing of receiving the electromagnetic wave reflected by the object and the transmitting direction of the electromagnetic wave. The surroundings detection sensor 11 also transmits the electromagnetic wave within a predetermined scanning range to measure multiple reflection points within the scanning range and obtains one frame of observation results including point cloud data of these multiple reflection points. The surroundings detection sensor 11 repeatedly performs measurements at a predetermined cycle and outputs the observation results of multiple consecutive frames to the controller 20.

[0012] The controller 20 is configured by a computer and detects objects and estimates the movement speed of the objects based on the observation results (frames including point cloud data) input from the surrounding detection sensor 11. Specifically, the controller 20 includes a storage device that stores various data and programs, and a processor configured with an arithmetic circuit such as a CPU (Central Processing Unit). As shown in FIG. 1 , the controller 20 functions as an object extraction unit 21, a rectangle estimation unit 22, a point cloud density calculation unit 23, and a speed estimation unit 24 by the processor reading and executing the program from the storage device. The controller 20 also includes a frame buffer 25 that records acquired frames, the movement speed estimation results calculated based on the frames, and the reliability of the estimation results.

[0013] The object extraction unit 21 divides the point cloud data obtained from the surroundings detection sensor 11 into point cloud subsets for each object 91 (moving object). A clustering method such as DBSCAN (Density-based spatial clustering of applications with noise) may be used as an object extraction method. When extracting objects, objects that are not to be detected (e.g., road surfaces) may be removed in advance as necessary. The object extraction unit 21 obtains a point cloud set for each object, and detects this point cloud set as a single object.

[0014] FIG. 2 illustrates an example of point cloud data detected by a perimeter detection sensor and an example of an object approximated by a rectangle based on the point cloud data. In FIG. 2 , a vehicle (host vehicle 90) equipped with an object detection device 10 detects an object 91 (e.g., another vehicle) present within a predetermined scanning range Ar using a perimeter detection sensor 11. In FIG. 2 , the object 91 moves from position P1 to position P2. The object extraction unit 21 acquires frames containing a point cloud of reflection points included in the scanning range at a predetermined interval. For example, in FIG. 2 , the object 91 is located at position P1 at a first timing and at position P2 at a second timing. In this case, the frame acquired at the first timing includes point cloud data of a reflection point Q on the object 91 at position P1, and the frame acquired at the second timing includes point cloud data of a reflection point Q on the object 91 at position P2. Note that while Figure 2 shows the two-dimensional position of each reflection point Q when the object 91 is viewed from the normal direction of the ground, in reality, these reflection points Q have three-dimensional (x, y, z) coordinates.

[0015] The rectangle estimation unit 22 approximates the object with a rectangular object model 92 based on each reflection point Q of the acquired point cloud data. Specifically, the point cloud set for each extracted object 91 is approximated with a rectangular object model. Here, the rectangle refers to a shape that is rectangular when viewed from the normal direction of the ground, as shown in FIG. 2 , and may include, for example, a rectangular parallelepiped or a cylinder. As a method for approximating the extracted object with the object model 92, parameters that best fit the point cloud to the model are estimated. Specifically, parameters that minimize the distance between each surface of the object model and each reflection point Q of the point cloud are estimated by optimization. As a method for this optimization, an algorithm such as L-shape fitting can be used, for example.

[0016] As a result, as shown in Fig. 2, the rectangle estimation unit 22 can approximate the object 91 to an object model 92 that is rectangular when viewed from the normal direction of the ground. For example, in the example of Fig. 2, the periphery detection sensor 11 acquires a reflection point Q on a first surface 91A (right surface) of the object 91 and a reflection point Q on a second surface 91B (rear surface) that is orthogonal to the first surface 91A. Therefore, the rectangle estimation unit 22 can estimate the first surface 91A and the second surface 91B of the object model from these reflection points Q, and can approximate the object to a rectangle.

[0017] Furthermore, if the area of ​​the approximated rectangle is equal to or greater than a predetermined second threshold, the rectangle estimation unit 22 further divides the rectangle approximating the object 91 into multiple sections. FIG. 3 is a diagram showing an example of dividing an approximated rectangular object model into multiple sections. For example, if the object 91 is a large vehicle such as a truck or bus, point cloud data of reflection points Q uniformly distributed on the circumferential surface of the object 91 cannot be obtained, and the reflection points Q may be concentrated in a specific area, as shown in FIG. 3. In such a case, the reliability of the reflection points Q decreases in areas where the intervals between the reflection points Q are long. Therefore, if the area of ​​the rectangular approximated object model 92 is equal to or greater than a predetermined second threshold, the rectangle estimation unit 22 divides the object model 92 into multiple sections 92A. The sections 92A are divided, for example, along the direction in which the reflection points Q in the rectangle vary. In the example of FIG. 3, the reflection points Q vary along the first surface 91A, which is the longitudinal direction, and the variation in the reflection points Q on the second surface 91B is small. 3, the object model 92 is divided into a plurality of sections 92A in the direction along the first surface 91A. If the reflection points Q also vary in the direction along the second surface 91B, the object model 92 may be divided into a plurality of sections 92A in the direction along the first surface 91A and the direction along the second surface 91B.

[0018] The point cloud density calculation unit 23 calculates the density (point cloud density) of each reflection point Q in the object model 92 approximated to a rectangle. Specifically, the point cloud density calculation unit 23 calculates (point cloud density) = (number of reflection points Q) / (rectangle volume) from the number of reflection points Q of the point cloud included within the rectangle of the object model 92. If the object model 92 is divided into multiple sections 92A, the point cloud density of the reflection points Q is calculated for each of the sections 92A.

[0019] FIG. 4 illustrates another example of point cloud data detected by a perimeter detection sensor. As shown in FIG. 4, the point cloud data may include a case in which only one surface (e.g., a first surface 91A) of the object 91 is observed, and a reflection point Q on a second surface 91B perpendicular to the first surface 91A is not observed. In this case, the rectangular size in the depth direction cannot be accurately determined. In response to this, the point cloud density calculation unit 23 may calculate the point cloud density for each of the horizontal and vertical directions. When dividing the object model 92 into the horizontal and vertical directions, the vertical point cloud density is calculated by dividing the object model 92 into multiple horizontal regions 93A in the vertical direction along a dividing line 93 parallel to the horizontal direction, and determining whether each region 93A contains a reflection point Q. In this case, the vertical point cloud density is calculated as follows: (vertical point cloud density) = (number of horizontal regions containing reflection point Q) / (total number of regions). Although not shown in the figure, the calculation of the point cloud density in the horizontal direction is similar, where the object model 92 is divided into a plurality of vertical regions in the horizontal direction by dividing lines parallel to the vertical direction, and whether each region contains a reflection point Q is determined by (horizontal point cloud density) = (number of vertical regions containing reflection point Q) / (total number of regions). In this way, by calculating the point cloud density separately for the horizontal and vertical directions, more information can be obtained, and accordingly, it becomes possible to perform velocity estimation with high accuracy.

[0020] The speed estimation unit 24 calculates the moving speed of the object 91 based on each reflection point Q in a plurality of consecutive frames. Specifically, as shown in FIG. 1 , the speed estimation unit 24 functions as an estimation method determination unit 241, a first speed estimation unit 242, a second speed estimation unit 243, and a speed integration unit 244.

[0021] The estimation method determination unit 241 determines an estimation method for performing velocity estimation based on the point cloud density. Specifically, when the point cloud density is equal to or greater than a predetermined first threshold, the estimation method determination unit 241 determines that the reliability of the reflection points Q included in the point cloud is high, and determines feature-point-based velocity estimation, which directly uses the reflection points Q, as the estimation method. More preferably, when the point cloud density is equal to or greater than the first threshold in both of two consecutive frames input from the surroundings detection sensor 11, the estimation method determination unit 241 determines feature-point-based velocity estimation as the estimation method. On the other hand, when the point cloud density is less than the predetermined first threshold, the estimation method determination unit 241 determines that the reliability of the reflection points Q is low, and determines center-position-based velocity estimation, which uses the center position C (see FIG. 2 ) of the object model 92 approximated to a rectangle, as the estimation method. When the object model 92 is divided into multiple sections 92A, the estimation method determination unit 241 determines the estimation method based on the section 92A with the highest point cloud density among the multiple sections 92A. For example, the estimation method determination unit 241 may select the section 92A with the highest point cloud density as the determination section, or may select a predetermined number of sections 92A in descending order of point cloud density as the determination sections.

[0022] Furthermore, when the point cloud density calculation unit 23 calculates the point cloud density separately for the horizontal direction and the vertical direction, the estimation method determination unit 241 determines the estimation method of the moving speed for each direction. When the speed estimation method for the horizontal direction and the speed estimation method for the vertical direction are different, the first speed estimation unit 242 and the second speed estimation unit 243 described below calculate the moving speed for each direction individually, and then the speed integration unit 244 integrates these speeds.

[0023] The first speed estimator 242 calculates the moving speed of the object 91 from the point cloud data by a feature point-based speed estimation method. In this feature point-based speed estimation, the first speed estimator 242 estimates the moving speed of the object 91 from the reflection point Q by estimating the moving speed of the object 91 from the feature point Q that has a high degree of distinctiveness in at least two consecutive frames (P1 and P2 in FIG. 2). Ｒ (See Figure 2) are identified. Then, these feature points Q Ｒ The moving speed is estimated based on the moving distance between the two frames and the time difference between the two frames.

[0024] Feature point Q ＲThe reflection points Q are preferably reflection points Q that can be identified as the same point between two frames, and examples thereof include a corner where the first surface 91A and the second surface 91B intersect in the object model 92, a high reflection point (such as the edge of a license plate) where the reflection intensity (brightness value) is higher than other surrounding reflection points Q, or points surrounding the high reflection point. The positions of the reflection points Q on the side of the moving object 91 vary depending on the timing of irradiation of the electromagnetic wave from the surrounding detection sensor 11, but the positions of the corners where two side surfaces (for example, the first surface 91A and the second surface 91B) intersect can be stably acquired even if the object 91 moves. In this case, a plurality of reflection points Q along the edge where two side surfaces of the object model 92 intersect are called feature points Q. Ｒ In addition, all of the reflection points Q within a predetermined distance range from the side may be set as the feature points Q. Ｒ Such a corner has no uncertainty in the horizontal and vertical directions, and the feature point Q Ｒ The correspondence between them can be stably established.

[0025] Furthermore, the high reflection points are filtered to exclude points with a brightness value less than a predetermined value, or the difference value between the high reflection points Q and the surrounding reflection points Q is calculated to identify the feature points Q with a high degree of characteristic. Ｒ A high reflection point is a point with a specifically high brightness value, and such high reflection points are a minority of the total reflection points Q, so that the characteristic points Q between two frames can be stably detected. Ｒ When LiDAR is used as the surrounding detection sensor 11, a highly reflective point with a high brightness value is, in principle, estimated to have a depth shorter than the actual depth. Therefore, the point obtained by averaging the highly reflective point with a high brightness value and its surrounding reflection points Q is called the feature point Q. Ｒ It may also be possible to use the following.

[0026] Also, feature point Q Ｒ However, if there is sufficient resource, all reflection points Q can be treated as feature points Q. Ｒ All downsampled reflection points Q may be used as feature points Q Ｒ In this case, the feature point Q Ｒ Since the approximation by is not relied upon, the correspondence between the reflection points Q between two frames can be stably established.

[0027] Feature points Q between two frames before and after Ｒ The correspondence between the feature points Q and Q is achieved by, for example, aligning the center positions C of the object models 92 to the same position. Ｒ The distance from the center position C to the feature point Q Ｒ The distance and angle to each feature point Q Ｒ The corresponding feature point Q is calculated based on the feature level (for example, the reflection intensity or the reflection intensity of the surrounding reflection point Q). Ｒ Then, the combination with the smallest total cost is found, and the corresponding feature point Q Ｒ The combination determination algorithm may be, for example, a greedy method or a Hungarian method. Ｒ Examples of the distance include the Mahalanobis distance between the two, the norm of the characteristic degree, or a weighted sum thereof.

[0028] The first speed estimation unit 242 estimates the moving speed of the object 91 by, for example, (moving speed)=(feature point Q Ｒ The time difference between two consecutive frames can be calculated by dividing the average of the movement amounts of the pair of frames by the time difference between two consecutive frames.

[0029] As described above, the point cloud density calculation unit 23 may calculate the point cloud densities in both the horizontal and vertical directions. In this case, a moving speed estimation method is determined for each of the horizontal and vertical directions. Here, if it is determined that feature point-based estimation processing is to be performed for at least one of the horizontal and vertical directions, the first speed estimation unit 242 estimates the moving speed using the feature point-based estimation processing in a direction perpendicular to the direction of high density. For example, as shown in FIG. 4 , a reflection point Q may be detected only on one side (e.g., the first surface 91A) of the object 91. In such a case, the first speed estimation unit 242 can accurately calculate the moving speed in the direction perpendicular to the side.

[0030] The first speed estimation unit 242 estimates the characteristic point Q ＲIn addition to calculating the moving speed based on the feature point Q, the reliability of the calculated moving speed is also calculated. The reliability of the speed estimation result of the feature point based speed estimation can be calculated based on the point cloud density of the reflection points Q. For example, Ｒ The reliability is calculated so that the larger the number of feature points Q, the higher the reliability is set. Ｒ The residual error between the individual moving speeds of the feature points Q Ｒ The value obtained by dividing the number of times the number of times the reliability is obtained may be used as the reliability.

[0031] The second speed estimator 243 estimates the moving speed of the object 91 using a center-position-based speed estimation method. Specifically, the second speed estimator 243 calculates the center position C of the object model 92 to be estimated for each of at least two consecutive frames (the previous and next frames). The second speed estimator 243 then calculates the moving speed by using the formula (movement speed) = (movement distance of the center position C in two consecutive frames) / (time difference between two consecutive frames). Similarly to the first speed estimator 242, the second speed estimator 243 calculates the moving speeds in both the horizontal and vertical directions. Note that the moving speed is estimated by the second speed estimator 243 when the point cloud density is less than the first threshold, and the reliability of the moving speed estimation result is lower than that of feature-point-based speed estimation. For example, the second speed estimator 243 may set a certain fixed value as the reliability.

[0032] The velocity integration unit 244 estimates the moving velocity of the object 91 by integrating the moving velocity estimated by the first velocity estimation unit 242 and the second velocity estimation unit 243, and the velocity information of the previous frame. In this embodiment, the estimation result of the moving velocity of the object 91 estimated by the velocity estimation unit 24 and the velocity information including the reliability thereof are temporarily stored in the frame buffer 25. When integrating the velocity information of the previous frame, it is preferable that the velocity integration unit 244 uses a filter such as a Kalman filter, an extended Kalman filter, or a particle filter. In this case, the moving velocity up to the previous frame and its variance, and the moving velocity estimated based on the current frame and its variance are used. An example of a velocity model used in the filter is a constant acceleration motion model. When a constant acceleration motion model is used, the updated velocity is calculated as V ｔ＋１ , the speed before update is V ｔ , acceleration a ｔ The updated velocity after the time dt has elapsed is V ｔ＋１ =V ｔ +a ｔ ×dt. The velocity integration unit 244 estimates an updated moving velocity (current estimated velocity) from the velocity model based on the velocity information of the previous frame, and corrects the moving velocity using the observation results of the current frame, thereby integrating the velocity information based on the previous frame with the moving velocity based on the current frame. At this time, the velocity integration unit 244 also updates the variance of the moving velocity based on the integrated moving velocity and the moving velocity calculated up to the previous frame. Furthermore, if the difference between the moving velocity based on the current frame and the estimated velocity estimated by the velocity model is large compared to the variance of the moving velocity, the filter does not need to be updated. For example, if the residual of the estimated velocity is three times or more the variance, there is a high possibility that it is an outlier, so no update is performed. Furthermore, if it is determined that the estimated velocity is greater than the possible speed of the object 91, no update is performed.

[0033] Furthermore, when integrating the movement speed based on the previous frame and the movement speed based on the new frame, the speed integration unit 244 refers to the reliability of the estimation result of the movement speed based on each frame and performs weighting according to the reliability. For example, if the reliability of the previous frame is lower than the reliability of the current frame, the speed integration unit 244 increases the weight value of the movement speed of the current frame and decreases the weight value of the movement speed based on the previous frame.

[0034] [Object Detection Method] Next, an object detection method of the object detection device 10 of this embodiment will be described. Fig. 5 is a flowchart showing the object detection method of this embodiment. In the object detection device 10 of this embodiment, the controller 20 first extracts an object by using the object extraction unit 21 to acquire a frame including point cloud data of a plurality of reflection points Q from the surroundings detection sensor 11 (step S1).

[0035] Next, the rectangle estimation unit 22 estimates an object model 92 that approximates the object 91 with a rectangle based on the point cloud data of each frame obtained in step S1 (step S2). For example, as shown in Fig. 2, when a first surface 91A and a second surface 91B that sandwich a corner of the object 91 are detected, a rectangle with the first surface 91A and the second surface 91B as its sides can be estimated.

[0036] The rectangle estimation unit 22 further determines whether the area of ​​the object model 92 estimated as a rectangle is equal to or greater than a predetermined second threshold (step S3). If the determination in step S3 is YES, the rectangle estimation unit 22 divides the estimated rectangle into a plurality of sections 92A (step S4). As described above, it is preferable to divide the sections 92A along the direction of variation of the reflection points Q. This allows the object model 92 to have different point cloud densities in each section 92A.

[0037] After step S4, the point cloud density calculation unit 23 calculates the point cloud density of each section 92A (step S5), and adopts the section 92A with the highest point cloud density as the determination section for determining the estimation method for the moving speed of the object 91 (step S6). Note that in this embodiment, the section 92A with the highest point cloud density is used as the determination section, but multiple sections 92A with high point cloud densities may be used as the determination sections.

[0038] On the other hand, if the determination in step S3 is NO, the point cloud density calculation unit 23 calculates the point cloud density of the rectangular estimated object model 92 (step S7). Note that in step S6 or step S7, the point cloud density calculation unit 23 may calculate the point cloud density in each direction, separately for the vertical direction and the horizontal direction, as described above.

[0039] Next, the estimation method determination unit 241 of the speed estimation unit 24 determines a method for estimating the moving speed of the object 91 based on the point cloud density of the section 92A that was determined as the determination section in step S6 or the point cloud density calculated in step S7. Specifically, the estimation method determination unit 241 determines whether the point cloud density is equal to or greater than a first threshold value (step S8).

[0040] If step S8 returns "YES," the estimation method determination unit 241 further determines whether the point cloud density of the object model 92 for the same object 91 in the most recently acquired frame is equal to or greater than a first threshold (step S9). If the point cloud densities are calculated separately for the vertical and horizontal directions in steps S6 and S7, the estimation method determination unit 241 determines whether the point cloud density is equal to or greater than the first threshold for each direction. If step S9 returns "YES," the estimation method determination unit 241 determines feature-point-based velocity estimation as the method for estimating the moving velocity of the object 91 (step S10). On the other hand, if step S8 returns "NO" and if step S9 returns "NO," the estimation method determination unit 241 determines center-position-based velocity estimation as the method for estimating the moving velocity of the object 91 (step S11).

[0041] Next, the speed estimation unit 24 performs speed estimation processing for the current frame based on the estimation method determined in step S10 or step S11 (step S12). FIG. 6 is a flowchart showing an outline of the speed estimation processing in step S12. First, the speed estimation unit 24 determines whether the estimation method is feature-point-based speed estimation (step S21). If the determination in step S21 is YES, the first speed estimation unit 242 performs feature-point-based speed estimation based on point cloud data of the reflection point Q. The first speed estimation unit 242 reads out the frame acquired one frame previously (the immediately preceding frame) from the frame buffer 25 (step S22). Next, the first speed estimation unit 242 estimates the feature point Q in the newly acquired current frame. Ｒ (step S23), and the corresponding feature point Q Ｒ and associate them (step S24). Ｒ As described above, examples of the reflection points Q include points on the corners where the first surface 91A and the second surface 91B intersect, high reflection points whose reflection intensity is higher than the surrounding reflection points Q, and reflection points (low reflection points) around the high reflection points. If there are sufficient resources, all reflection points Q can be treated as characteristic points Q. Ｒ It may also be possible to use the following.

[0042] Then, the first velocity estimation unit 242 estimates the feature point Q Ｒ From the feature point Q of the current frame, Ｒ The first speed estimator 242 estimates the moving speed of the object 91 based on the distance to the target object 91 and the time difference between the acquisition time of the previous frame and the acquisition time of the current frame (step S25). Furthermore, the first speed estimator 242 calculates the reliability of the estimated moving speed (step S26).

[0043] On the other hand, if the determination in step S21 is NO, the second velocity estimator 243 performs center-position-based velocity estimation based on the center position C of the object model 92. Specifically, the second velocity estimator 243 reads the previous frame from the frame buffer (step S27). Next, the second velocity estimator 243 calculates the center position C of the object model 92 for the previous frame and the center position C of the object model 92 for the current frame (step S28). Then, the second velocity estimator 243 estimates the moving velocity of the object 91 based on the movement distance from the center position C of the object model 92 in the previous frame to the center position C of the object model 92 in the current frame and the time difference between the acquisition time of the previous frame and the acquisition time of the current frame (step S29).

[0044] Furthermore, the second speed estimator 243 calculates the reliability of the estimated moving speed (step S30). The center position based speed estimation may use a fixed value as described above.

[0045] Returning to FIG. 5 , after the velocity estimation process in step S12, the velocity integration unit 244 reads velocity information including the moving velocity of the object 91, the reliability of the moving velocity, and the variance of the moving velocity, calculated based on the previous frame (previous frame) stored in the frame buffer 25 (step S13). The velocity integration unit 244 then integrates the moving velocity calculated based on the current frame in step S12 with the estimated velocity at the time of acquiring the current frame, which is estimated based on the velocity information of the previous frame, to calculate the final moving velocity of the object 91 (step S14). Note that if point cloud densities are calculated separately for the vertical and horizontal directions in step S6 or step S7, the moving velocity estimation process in step S12 is performed for each direction. In this case, the velocity integration unit 244 integrates the moving velocities in the vertical and horizontal directions to calculate the moving velocity based on the current frame. Furthermore, in step S14, as described above, weighting is performed based on the reliability of the movement speed calculated based on the previous frame and the reliability of the movement speed calculated based on the current frame, and the movement speeds are integrated with emphasis on the movement speed with high reliability.

[0046] [Effects of the Present Embodiment] The object detection device 10 of the present embodiment includes a perimeter detection sensor 11 and a controller 20. The controller 20 functions as an object extraction unit 21, a rectangle estimation unit 22, a point cloud density calculation unit 23, and a speed estimation unit 24 by having a processor read and execute a program stored in the device. The perimeter detection sensor 11 emits electromagnetic waves and detects the electromagnetic waves reflected by an object 91, thereby outputting a frame including reflection points Q of the electromagnetic waves on the object 91 to the controller 20. The object extraction unit 21 acquires a frame including multiple reflection points Q from the perimeter detection sensor 11 and extracts the object 91. The rectangle estimation unit 22 estimates an object model 92 that approximates the object 91 with a rectangle from the multiple reflection points Q (steps S2 to S4). The point cloud density calculation unit 23 calculates the point cloud density from the number of reflection points Q included in the object model 92 (steps S5 to S7). The speed estimation unit 24 functions as an estimation method determination unit 241, a first speed estimation unit 242, a second speed estimation unit 243, etc., and estimates the moving speed of the object 91 (steps S8 to S14). When the point cloud density is equal to or greater than a first threshold, the estimation method determination unit 241 of the speed estimation unit 24 determines that at least one of the multiple reflection points Q is a feature point Q Ｒ and the feature point Q Ｒ The first speed estimation unit 242 performs the feature point-based estimation process. When the point cloud density is less than the first threshold, the estimation method determination unit 241 calculates the center position C of the object model 92 and determines the movement speed of the object 91 from the movement amount of the center position C as the estimation method. The second speed estimation unit 243 performs the center position-based estimation process.

[0047] This makes it possible to perform highly accurate object detection by switching the object detection method depending on the point cloud density. In other words, when the point cloud density of the reflection points Q in the object model 92 is high, it indicates that the reliability of each reflection point Q is high. In such a case, at least some or all of the reflection points Q are treated as feature points Q. Ｒ and the feature point Q ＲThe feature-point-based estimation process estimates the moving speed based on the distance traveled by the object 91. In particular, a scene in which a large number of reflection points Q are obtained is likely to involve a situation in which another vehicle, the object 91, is located relatively close to the host vehicle. In such a case, the center-position-based estimation process may not detect the entire object 91 by the periphery detection sensor 11. Therefore, estimating the moving speed based on the center position C of the rectangle estimated from a portion of the object 91 may result in an error. In contrast, the feature-point-based estimation process directly uses the reflection points Q to calculate the moving speed of the object 91, thereby enabling a more accurate estimation of the moving speed than the center-position-based speed estimation process. On the other hand, when the object 91 is located far away, the number of reflection points Q reflected by the object 91 decreases, resulting in a lower point cloud density in the object model 92. In this case, the reliability of the reflection points Q decreases, resulting in a decrease in the accuracy (reliability) of the moving speed estimated by the feature-point-based speed estimation. However, when the object 91 is located far away, the entire object 91 is likely to be included in the scanning range, and the error in the center position C of the rectangular object model 92 decreases. Therefore, by switching the speed estimation method depending on the point cloud density as in this embodiment, it is possible to select an appropriate speed estimation method according to the reliability of the reflection point Q, and it is possible to detect the object 91 with high accuracy and estimate the moving speed of the detected object. Therefore, it is possible to estimate with higher accuracy the speed vector required for determining whether another vehicle is cutting in ahead of the host vehicle.

[0048] In this embodiment, in step S3, the rectangle estimation unit 22 determines whether the area of ​​the rectangularly approximated object model 92 is equal to or greater than a second threshold. If the area of ​​the object model 92 is equal to or greater than the second threshold, the object model 92 is divided into multiple sections 92A. When the object model 92 is divided into multiple sections 92A, the point cloud density calculation unit 23 calculates the point cloud density of each section 92A in steps S5 and S6, and uses the section 92A with the highest point cloud density as the judgment section for determining the travel speed estimation method. When performing feature point-based estimation processing, the first speed estimation unit 242 of the speed estimation unit 24 estimates the travel speed of the object 91 based on the judgment section with the highest point cloud density. As a result, even if a large-sized object 91, such as a large vehicle, is approximated with a rectangular object model 92, resulting in a bias in the point cloud density, speed estimation using the reflection points Q in the low point cloud density area is not performed, and a highly accurate travel speed estimation process can be performed using the reflection points Q with the high point cloud density.

[0049] In this embodiment, the point cloud density calculation unit 23 calculates the point cloud density for each of the vertical direction and the horizontal direction when calculating the point cloud density of the object model 92. This makes it possible to estimate the moving speed based on information on the reflection points Q for the vertical direction and the reflection points Q for the horizontal direction, and enables more accurate estimation of the moving speed compared to when the vertical direction and the horizontal direction are determined uniformly.

[0050] In this embodiment, when the point cloud densities in the vertical and horizontal directions are calculated for one surface of the object 91, a speed estimation method is determined for each direction based on these point cloud densities. If the point cloud density in at least one of the vertical and horizontal directions is equal to or greater than a first threshold, the first speed estimation unit 242 estimates the movement speed in a direction perpendicular to the one surface of the object 91 using a feature-point-based estimation process. When only reflection points Q from one surface of the object 91 are acquired, fewer reflection points are obtained upward in the depth direction of the object model 92. In contrast, as described above, when the point cloud densities in the vertical and horizontal directions are calculated and a movement speed estimation method is determined based on each point cloud density, information in each direction is obtained, enabling a movement speed to be estimated based on more information. Therefore, even when reflection points are only obtained from one surface of the object 91, a highly accurate movement speed can be estimated even in the depth direction perpendicular to the one surface.

[0051] In this embodiment, the estimation method determination unit 241 of the speed estimation unit 24 adopts the feature point-based estimation process as the estimation method when the point cloud density is equal to or greater than a first threshold value in two consecutive frames. As a result, the first speed estimation unit 242 determines the distance between corresponding reflection points Q (feature points Q Ｒ ) can be obtained with high accuracy, and the reliability of the estimation accuracy of the moving speed of the object 91 can also be improved.

[0052] In this embodiment, the first velocity estimation unit 242 of the velocity estimation unit 24 estimates a feature point Q of the object 91 from a plurality of reflection points Q in each of two consecutive frames. Ｒ and identify the feature points Q in two consecutive frames. Ｒ The moving speed of the object 91 is estimated based on the moving distance of the feature point Q between the two frames and the time difference between the two consecutive frames. Ｒ The moving speed of the object 91 can be estimated from the amount of movement and the acquisition interval between two consecutive frames.

[0053] In this embodiment, the first speed estimator 242 estimates the speed at the characteristic point Q ＲIt is preferable to specify the corners where the sides of the rectangle of the object model 92 intersect as the feature points Q. The corners of the rectangle are, for example, the positions where two surfaces (first surface 91A and second surface 91B) of the object 91 intersect, and can be detected with high accuracy even if the position of the object 91 as seen from the vehicle itself or the orientation of the object 91 changes depending on the timing of transmitting the electromagnetic waves from the surroundings detection sensor 11. For this reason, the corners of the rectangle are specified as the feature points Q. Ｒ This allows stable speed estimation and improves the estimation accuracy.

[0054] The first speed estimation unit 242 estimates the characteristic point Q Ｒ As the reflection points, high reflection points having a higher reflection intensity than the surrounding reflection points Q, or points around the high reflection points, may be used. The high reflection points and the points around the high reflection points are points detected by the unevenness or edge portions of the object 91, and are points with a high degree of distinctiveness on the object 91. For this reason, such high reflection points and the surrounding reflection points Q are called characteristic points Q. Ｒ Even by doing so, it is possible to perform stable speed estimation and improve the estimation accuracy.

[0055] In this embodiment, the speed estimation unit 24 records the result of estimating the moving speed based on a previously acquired frame in the frame buffer 25. Then, the speed integration unit 244 integrates the moving speed of the object 91 estimated based on the reflection point Q of the newly acquired current frame with the result of estimating the moving speed based on the previous frame, to estimate (correct) the moving speed of the object 91. This makes it possible to estimate the moving speed with high accuracy while suppressing sudden changes in the moving speed of the object 91. In other words, the speed integration unit 244 integrates the moving speed of the object 91 estimated based on the reflection point Q of the newly acquired current frame with the result of estimating the moving speed based on the previous frame between two consecutive frames. Ｒ In the estimation of the moving speed of the object 91 based on the differential value of the moving distance of the feature point Q, the moving speed may suddenly increase. In general, it is rare for the moving speed of the object 91 to suddenly change. Ｒ If the moving speed is estimated based only on the differential value of the moving distance, there is a possibility of erroneously determining a cut-in vehicle crossing in front of the vehicle 90. In contrast, by integrating the estimated speed from the previous frame and the moving speed estimated based on the newly acquired current frame, it is possible to suppress erroneous determination of a sudden change in moving speed, and thereby reduce erroneous determination of a cut-in vehicle.

[0056] In this embodiment, the first speed estimating unit 242 and the second speed estimating unit 243 further calculate the reliability of the estimation result of the moving speed of the object 91. Then, the speed integrating unit 244 compares the reliability of the moving speed based on the previous frame, which was previously acquired, with the reliability of the moving speed based on the current frame, and performs speed estimation that prioritizes the frame with the higher reliability. For example, the estimated moving speed is weighted according to the reliability, and then the moving speed based on the previous frame and the moving speed based on the current frame are integrated. This makes it possible to respond to a truly sudden change in the moving speed of the object 91 and to properly determine a sudden cut-in by another vehicle.

[0057] [Modifications] The present invention is not limited to the above-described embodiment, and includes the following modifications within the scope that allows the object of the present invention to be achieved. For example, in the above-described embodiment, an example was shown in which the speed integration unit 244 integrates the speed estimation result based on the previous frame previously acquired and the speed estimation result based on the newly acquired frame. However, without performing speed integration, the moving speed estimated by the first speed estimating unit 242 or the second speed estimating unit 243 may be determined as the moving speed of the object 91, and for example, a cut-in of another vehicle may be determined.

[0058] Furthermore, the speed integration unit 244 performs reliability-based movement speed integration by weighting the speed estimation result based on the previous frame and the speed estimation result based on the newly acquired frame based on the reliability, but is not limited to this. For example, the speed integration unit 244 may perform movement speed integration when the reliability of the speed estimation result based on the previous frame is equal to or greater than a predetermined value, and may not perform integration when the reliability is less than the predetermined value.

[0059] 10...Object detection device, 11...Surroundings detection sensor, 20...Controller, 21...Object extraction unit, 22...Rectangle estimation unit, 23...Point cloud density calculation unit, 24...Speed ​​estimation unit, 25...Frame buffer, 90...Own vehicle, 91...Object, 92...Object model, 92A...Division, 93...Division line, 93A...Area, 241...Estimation method determination unit, 242...First speed estimation unit, 243...Second speed estimation unit, 244...Speed ​​integration unit, Ar...Scanning range, C...Center position, Q...Reflection point, Q Ｒ ...Feature points.

Claims

1. An object detection method using a surroundings detection sensor that irradiates electromagnetic waves and detects the electromagnetic waves reflected by an object, thereby outputting a frame including reflection points of the electromagnetic waves on the object, and a controller that detects the object based on a plurality of the reflection points included in the frame, wherein the controller: approximates the object with a rectangle from the plurality of reflection points; calculates a point cloud density from the number of reflection points included in the rectangle; and estimates a movement speed of the object based on the reflection points; and when estimating the movement speed, if the point cloud density is equal to or greater than a predetermined first threshold, the controller performs a feature point-based estimation process that sets at least one of the plurality of reflection points as a feature point and estimates the movement speed of the object from the amount of movement of the position of the feature point; and if the point cloud density is less than the first threshold, performs a center position-based estimation process that calculates the center position of the rectangle and estimates the movement speed of the object from the amount of movement of the center position.

2. The object detection method of claim 1, wherein the controller, when approximating the object with a rectangle, divides the rectangle into multiple sections if the area of ​​the approximated rectangle is equal to or greater than a predetermined second threshold; when calculating the point cloud density, calculates the point cloud density for each of the multiple sections; and when estimating the movement speed, when performing the feature point-based estimation process, estimates the movement speed of the object based on the section with the highest point cloud density.

3. The object detection method according to claim 1 or 2, wherein, when calculating the point cloud density, the controller calculates the point cloud density in each of a vertical direction and a horizontal direction perpendicular to the vertical direction.

4. The object detection method described in claim 3, wherein, when estimating the moving speed, if the point cloud density in at least one of the vertical direction and the horizontal direction for one surface of the object is equal to or greater than the first threshold, the controller estimates the moving speed in a direction perpendicular to the one surface using the feature point-based estimation process.

5. An object detection method according to any one of claims 1 to 4, wherein the controller performs the feature point-based estimation process when estimating the moving speed if the point cloud density is equal to or greater than the first threshold value in two consecutive frames.

6. An object detection method according to any one of claims 1 to 5, wherein, when estimating the moving speed, the controller identifies the feature points of the object from a plurality of the reflection points in each of two consecutive frames, and estimates the moving speed of the object based on the moving distance of the feature points in the two consecutive frames and the time difference between the two consecutive frames.

7. The object detection method according to claim 6, wherein the feature points are corners where the sides of the rectangle intersect.

8. The object detection method according to claim 6, wherein the feature point is a high-reflection point having a higher reflection intensity than surrounding reflection points, or a point surrounding the high-reflection point.

9. An object detection method according to any one of claims 1 to 8, wherein, when estimating the moving speed, the controller records an estimation result of the moving speed based on the previously acquired frame, and estimates the moving speed of the object by integrating the moving speed of the object estimated based on the reflection points in the newly acquired frame and the estimation result of the moving speed based on the previously acquired frame.

10. The object detection method according to claim 9, wherein, when estimating the moving speed, the controller further calculates a reliability of the speed estimation result, stores the speed estimation result based on the previously acquired frame and the reliability, compares the reliability of the speed estimation result based on the newly acquired frame with the reliability of the speed estimation result based on the previously acquired frame, and performs the speed estimation by prioritizing the frame with the higher reliability.

11. An object detection device comprising: a detection sensor that irradiates electromagnetic waves and detects reflection points of the electromagnetic waves on an object by detecting the electromagnetic waves reflected by the object; and a controller that detects the object based on the detected multiple reflection points, wherein the controller comprises: a rectangle estimation unit that approximates the object with a rectangle from the multiple reflection points; a point cloud density calculation unit that calculates a point cloud density from the number of reflection points included in the rectangle; and a speed estimation unit that estimates a moving speed of the object based on the reflection points, wherein the speed estimation unit, when the point cloud density is equal to or greater than a predetermined first threshold, performs a feature point-based estimation process that sets at least one of the multiple reflection points as a feature point and estimates the moving speed of the object from the amount of movement of the feature point, and when the point cloud density is less than the first threshold, performs a center position-based estimation process that calculates the center position of the rectangle and estimates the moving speed of the object from the amount of movement of the center position.

Images