A collision detection method for autonomous driving based on directional bounding box

Abstract

The present invention discloses an automatic driving collision detection method based on a directional bounding box, which relates to the technical field of automatic driving. Firstly, relatively reliable obstacle bounding box (OBB) data is obtained in real time from a vehicle perception system; secondly, a planned candidate trajectory is obtained from a vehicle trajectory planning module and discretized, and a vehicle OBB bounding box is created for each discrete point; thirdly, a distance between the center of the vehicle and the center of the obstacle is preliminarily calculated to clarify whether the two are absolutely safe (no collision will occur); finally, a separation axis test is performed on obstacles and vehicles with possible collision to determine whether a collision occurs, and the collision risk is calculated based on the shortest distance between the two, so as to make the next decision for the unmanned driving system. The present invention improves the adaptability, safety and accuracy of the automatic driving vehicle and the surrounding environment by analyzing the collision between dynamic obstacles and static obstacles.

Description

Technical Field

[0001] The present invention belongs to the technical field of driverless vehicles, and in particular relates to an automatic driving collision detection method based on a directional bounding box. Background Art

[0002] In the field of autonomous driving, one of the most core modules is trajectory planning, which includes collision detection between vehicles and obstacles. In traditional collision detection methods, vehicles are usually represented by particles or vehicle circumscribed circles, and obstacles are represented by AABB bounding boxes or OBB bounding boxes, and then the separating axis theorem is used directly to determine whether the vehicle collides with the obstacle. For obstacles in the environment, whether they are moving objects or static objects, whether they are distant obstacles or nearby obstacles, collision analysis is directly performed. Although this can determine whether the vehicle trajectory collides with obstacles in the surrounding environment, the calculation method is relatively rough and there are cases of false detection, which will cause the originally feasible trajectory on a macro scale to become inaccessible.

[0003] The bounding box collision detection method is a detection method based on the real environment space. In geometric space, it can be described as follows: if there is a straight line / plane between two bounding boxes that can separate the two, then the two bounding boxes do not intersect. In order to balance the speed and accuracy of collision detection, some researchers have combined the AABB bounding box with the OBB bounding box, but once the AABB bounding box is used, the accuracy of the object bounding box is inevitably reduced.

[0004] According to the movement conditions, obstacles are divided into static obstacles and dynamic obstacles. The collision detection of static obstacles is relatively simple. Since the position of dynamic obstacles in space is changing, if dynamic obstacles are directly processed in a static form, the accuracy of collision detection will be lower. For autonomous driving, simple environments with fewer obstacles and low-speed driving can meet the requirements, but for complex environments with more obstacles, it will cause the vehicle to be unable to pass and even pose a safety hazard. Summary of the invention

[0005] The purpose of the present invention is to provide an autonomous driving collision detection method based on a directional bounding box to solve the above-mentioned defects caused by the prior art.

[0006] An automatic driving collision detection method based on a directional bounding box comprises the following steps:

[0007] S1. Environmental perception and fusion and coordinate unification and conversion

[0008] Obtain relatively reliable obstacle OBB bounding box data from the vehicle perception system in real time, obtain the planned candidate trajectory from the vehicle trajectory planning module and discretize it, create a vehicle OBB bounding box for each discrete point, and unify the vehicle and the obstacle into the world coordinate system;

[0009] S2. Pre-analysis of collision between vehicle and obstacle

[0010] For the vehicle direction bounding box and the obstacle direction bounding box, when the comparison between the center distance of the two direction bounding boxes and their respective minimum circumscribed circle radius is greater than the safety threshold, it is considered that the two direction bounding boxes will not collide and are absolutely safe; otherwise, accurate collision analysis is required, calculated as follows:

[0011]

[0012] in:

[0013] S3. Accurate collision analysis

[0014] Project each side of the vehicle direction bounding box and the pedestrian direction bounding box on the normal vector of each axis. If the projection intervals of the two polygons do not overlap, it means that the two bounding boxes are separated on the normal vector of the axis, and the minimum distance is L. min，i ; If the projection intervals of the two polygons overlap, it means that the test of the two bounding boxes on the axis normal vector is intersection;

[0015] The same collision analysis is performed on each edge of the polygon. As long as there is an edge that is tested as separated, it means that the two bounding boxes have not collided and the minimum distance is {L min，i}The minimum value L of the set min ;

[0016] S4. Collision risk calculation

[0017] When the minimum distance between two bounding boxes is calculated, the collision risk of the two bounding boxes can be evaluated based on the distance: the closer the minimum distance, the greater the collision risk, and the farther the minimum distance, the smaller the collision risk. That is, the collision distance is inversely proportional to the collision risk. Specifically, it is expressed as follows:

[0018]

[0019] Preferably, the precise analysis of the collision between the vehicle and the obstacle includes collision detection analysis of static obstacles and collision detection analysis of dynamic obstacles.

[0020] Preferably, the static obstacle collision analysis method is: for the candidate trajectory, the set of vehicle bounding boxes discretized at equal time intervals T is {{OBB 车 ,t}, t=0,T,2T,3T,..., the set of static obstacle bounding boxes is {OBB 障碍物 , i}, i = 0, 1, 2, 3, ...; The collision analysis between the vehicle bounding box and the static obstacle bounding box is performed. As long as there is a static obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all static obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set {collision_risk i}, the maximum value among i=0,1,2,3,...

[0021] Preferably, the method for dynamic obstacle collision analysis is: for candidate trajectories, the set of vehicle bounding boxes discretized at equal time intervals T is {{OBB 车 , t},t=0,T,2T,3T,..., the set of predicted bounding boxes of dynamic obstacles is {OBB 障碍物 , i，t}, i = 0, 1, 2, 3, ..., t = 0, T, 2T, 3T, ...; The collision analysis between the vehicle bounding box and the dynamic obstacle bounding box is performed. As long as there is a dynamic obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all the dynamic obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set {collision_risk i}, the maximum value among i=0,1,2,3,...

[0022] Preferably, after calculating the maximum collision risk of dynamic obstacles and static obstacles, the maximum risk value of dynamic analysis and static analysis is taken as the maximum collision risk value of the candidate trajectory and obstacle set; the maximum collision risk value is used for the autonomous driving trajectory planning and decision-making module to evaluate obstacles in the environment and make reasonable actions.

[0023] Preferably, the reasonable actions include but are not limited to deceleration, obstacle avoidance and parking.

[0024] The advantages of the present invention are:

[0025] 1. By comparing the center distance between the vehicle direction bounding box OBB and the obstacle direction bounding box OBB with the minimum circumscribed circle radius of the two, the collision situation can be quickly predicted, and only obstacles with collision risks are accurately analyzed, which speeds up the collision detection.

[0026] 2. For complex obstacles, the obstacle contour is described in polygonal form, and collision detection is performed on the vehicle OBB and the obstacle OBB, and the projection method is used to detect collisions on each axis of the bounding box, which improves the accuracy of collision analysis. It can not only calculate whether two bounding boxes collide, but also calculate the minimum distance between two non-colliding bounding boxes.

[0027] 3. Static obstacles and dynamic obstacles are classified and detected, and trajectories of dynamic obstacles are predicted. Whether it is a static obstacle or a dynamic obstacle, the collision detection result will be more accurate, thereby improving the accuracy of trajectory planning and the safety of the autonomous driving system.

[0028] 4. For risky obstacles, the collision risk is calculated. The closer the obstacle is to the candidate trajectory, the greater the risk; the farther the obstacle is from the candidate trajectory, the smaller the risk. The autonomous driving system makes more reasonable decisions and plans based on the collision risk value, such as deceleration, obstacle avoidance, parking, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is the main process of autonomous driving collision detection.

[0030] Figure 2 Schematic diagram of pre-detection of vehicles and obstacles.

[0031] Figure 3 Schematic diagram of accurate collision detection between vehicle and obstacle.

[0032] Figure 4 Schematic diagram of collision detection between vehicle candidate trajectories and static obstacles.

[0033] Figure 5 Schematic diagram of collision detection between vehicle candidate trajectories and dynamic obstacles. DETAILED DESCRIPTION

[0034] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0035] Reference Figure 1 As shown, the present invention is a collision detection method for autonomous driving based on a directional bounding box, characterized by including: coordinate unification and coordinate conversion, vehicle-obstacle collision pre-analysis, vehicle-obstacle collision precise analysis, collision risk calculation, etc. Among them:

[0036] The coordinate unification and coordinate conversion method is as follows: in order to predict the trajectory of dynamic obstacles, discretize candidate trajectories and implement collision detection, it is necessary to unify the vehicle and the obstacle into the world coordinate system.

[0037] Furthermore, the candidate trajectories of the vehicle are discretized at equal time intervals, and the OBB of the vehicle at the discrete points is calculated to form an OBB set of discrete points of the candidate trajectories of the vehicle.

[0038] Furthermore, the trajectory prediction of dynamic obstacles can be performed from two dimensions (x, y). At any time, the movement of the obstacle is processed as uniform linear motion, that is, the center point position of the obstacle is:

[0039]

[0040] Finally, a set of dynamic obstacles OBB is formed according to different moments.

[0041] Furthermore, the vehicle and obstacle collision pre-analysis method is: referring to Figure 2 As shown in the figure, for the vehicle OBB and the obstacle OBB, when the comparison between the center distance of the two OBBs and the radius of their respective minimum circumscribed circles is greater than the safety threshold, it is considered that the two OBBs will not collide and are absolutely safe; otherwise, accurate collision analysis is required.

[0042] Furthermore, the calculation of the collision pre-analysis between the vehicle and the obstacle is as follows:

[0043]

[0044] in:

[0045] Furthermore, the precise analysis of the collision between the vehicle and the obstacle includes collision detection analysis of static obstacles and collision detection analysis of dynamic obstacles.

[0046] Further, accurate collision analysis is performed by referring to Figure 3 As shown, taking the collision detection of vehicles and pedestrians as an example, the vehicle direction bounding box OBB is a quadrilateral ABCD. The pedestrian direction bounding box OBB is a polygon EFGHI. The purpose is that pedestrians are irregular objects, and using polygons can more accurately describe the outline of obstacles.

[0047] Furthermore, when performing accurate collision analysis, it is necessary to perform projection calculations on the normal vectors of each axis for each edge of the two bounding boxes. The EI edge of the obstacle EFGHI is used as an example to provide a detailed description of collision detection. The normal vector perpendicular to the EI edge is The projection points of polygon ABCD on the normal vector are A', B', C', D', so the projection points of any point on the polygon on the normal vector will all be within the interval [A', C']; the projection points of polygon EFGHI on the normal vector are E', F', G', H', I', so the projection of any point on the polygon on the normal vector will all be within the interval [E', G']. If the projection intervals of the two polygons do not overlap, it means that the test of the two bounding boxes on the axis normal vector is separated (no collision occurs), and the minimum distance is L min，i ; If the projection intervals of the two polygons overlap, it means that the two bounding boxes are intersecting (colliding) when tested on the axis normal vector. The same collision analysis is performed on each edge of the polygon. As long as there is an edge that is separated when tested, it means that the two bounding boxes are not colliding, and the minimum distance is {L min，i}The minimum value L of the set min .

[0048] Furthermore, when the minimum distance between two bounding boxes is calculated, the collision risk of the two bounding boxes can be evaluated based on the distance. The closer the minimum distance, the greater the collision risk, and the farther the minimum distance, the smaller the collision risk, that is, the collision distance is inversely correlated with the collision risk. Specifically, it can be expressed as follows:

[0049]

[0050] Furthermore, precise collision analysis includes static obstacle collision analysis and dynamic obstacle collision analysis.

[0051] Static obstacle collision analysis, the method is: refer to Figure 4 As shown, for candidate trajectory 1, the set of vehicle bounding boxes discretized at equal time intervals T is {{OBB 车 , t}, t=0,T,2T,3T,..., the set of static obstacle bounding boxes is {OBB 障碍物 , i}, i = 0, 1, 2, 3, .... The collision analysis between the vehicle bounding box and the static obstacle bounding box is performed. As long as there is a static obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all static obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set {collision_risk i}, the maximum value among i=0,1,2,3,...

[0052] Dynamic obstacle collision analysis, the method is: refer to Figure 5As shown in the figure, for the candidate trajectory, the set of vehicle bounding boxes discretized at equal time intervals T is {{OBB 车 , t},t=0,T,2T,3T,..., the set of predicted bounding boxes of dynamic obstacles is {OBB 障碍物 , i，t}, i = 0, 1, 2, 3, ..., t = 0, T, 2T, 3T, .... The collision analysis between the vehicle bounding box and the dynamic obstacle bounding box is performed. As long as there is a dynamic obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all dynamic obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set {collision_risk i}, the maximum value among i=0,1,2,3,...

[0053] Furthermore, after calculating the maximum collision risk of dynamic obstacles and static obstacles, the maximum risk value of dynamic analysis and static analysis is taken as the maximum collision risk value of the candidate trajectory and obstacle set. This maximum collision risk value is used by the autonomous driving trajectory planning and decision module to evaluate obstacles in the environment and make reasonable actions, such as deceleration, obstacle avoidance, parking, etc.

[0054] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A collision detection method for autonomous driving based on a directional bounding box, characterized in that: The steps include: S1. Environmental perception and fusion and coordinate unification and conversion Obtain relatively reliable obstacle OBB bounding box data from the vehicle perception system in real time, obtain the planned candidate trajectory from the vehicle trajectory planning module and discretize it, create a vehicle OBB bounding box for each discrete point, and unify the vehicle and the obstacle into the world coordinate system; S2. Pre-analysis of collision between vehicle and obstacle For the vehicle direction bounding box and the obstacle direction bounding box, when the comparison between the center distance of the two direction bounding boxes and their respective minimum circumscribed circle radius is greater than the safety threshold, it is considered that the two direction bounding boxes will not collide and are absolutely safe; otherwise, accurate collision analysis is required, calculated as follows: in: ; S3. Accurate collision analysis Project the normal vector of each axis on each side of the vehicle direction bounding box and the pedestrian direction bounding box. If the projection intervals of the two polygons do not overlap, it means that the two bounding boxes are separated on the normal vector of the axis, and the minimum distance is ; If the projection intervals of the two polygons overlap, it means that the test of the two bounding boxes on the axis normal vector is intersection; The same collision analysis is performed on each edge of the polygon. As long as there is an edge that is tested as separated, it means that the two bounding boxes do not collide and the minimum distance is { }The minimum value of the set ; S4. Collision risk calculation When the minimum distance between the two bounding boxes is calculated, the collision risk of the two bounding boxes is evaluated based on the distance: the closer the minimum distance is, the greater the collision risk is, and the farther the minimum distance is, the smaller the collision risk is. That is, the collision distance is inversely correlated with the collision risk. Specifically, it is expressed as the following relationship: 。 2. The method for collision detection of autonomous driving based on directional bounding box according to claim 1, characterized in that: The precise analysis of the collision between the vehicle and the obstacle includes collision detection analysis of static obstacles and collision detection analysis of dynamic obstacles.

3. The automatic driving collision detection method based on directional bounding box according to claim 2 is characterized in that: The static obstacle collision analysis method is: for the candidate trajectory, the set of vehicle bounding boxes discretized at equal time intervals T is { , the set of static obstacle bounding boxes is ; The collision analysis between the vehicle bounding box and the static obstacle bounding box is performed. As long as there is a static obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all static obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set The maximum value in .

4. The method for collision detection of autonomous driving based on directional bounding box according to claim 2, characterized in that: The method for dynamic obstacle collision analysis is: for candidate trajectories, the set of vehicle bounding boxes discretized at equal time intervals T is { , the set of predicted bounding boxes of dynamic obstacles is ; The collision analysis between the vehicle bounding box and the dynamic obstacle bounding box is performed. As long as there is a dynamic obstacle that collides with the vehicle bounding box, it means that the candidate trajectory will collide; if all dynamic obstacle bounding boxes do not collide with the vehicle bounding box, it means that the candidate trajectory will not collide, and the collision risk of the trajectory is the collision risk set The maximum value in .

5. The method for collision detection of autonomous driving based on directional bounding box according to claim 2, characterized in that: After calculating the maximum collision risk of dynamic obstacles and static obstacles, the maximum risk value of dynamic analysis and static analysis is taken as the maximum collision risk value of the candidate trajectory and obstacle set; this maximum collision risk value is used by the autonomous driving trajectory planning and decision-making module to evaluate obstacles in the environment and make reasonable actions.

6. The method for collision detection of autonomous driving based on directional bounding box according to claim 5, characterized in that: The reasonable actions include but are not limited to deceleration, obstacle avoidance and parking.

Images