Collision detection method and device and vehicle

Abstract

The invention provides a collision detection method and device and a vehicle. The method comprises the following steps: determining an associated object of the vehicle from an adjacent frame image of a driving area; predicting a collision moment of the vehicle and the associated object based on the pixel width of the associated object in each frame of image and the distance between the associated object and the vehicle in each frame of image; and predicting a collision result of the vehicle and the associated object based on the motion trail of the vehicle in the target duration and the motion trail of the associated object in the target duration. According to the invention, the pixel width of the associated object in each frame of image and the distance between the associated object in each frame of image and the vehicle can be combined to more accurately determine the movement track change condition of the associated object, so that the collision moment of the vehicle and the associated object can be accurately predicted according to the movement track change condition of the associated object. Therefore, the collision result of the vehicle and the associated object can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a collision detection method, device and vehicle. Background Art

[0002] In the automotive field, collision detection can be understood as detecting whether a collision will occur between two vehicles or between a vehicle and an obstacle. Currently, vehicle collision detection mainly relies on the driver's intuitive judgment, however, this intuition often deviates from the actual collision situation. In other words, the driver's prediction accuracy of collision detection based on his own intuition is relatively low. Summary of the invention

[0003] The present invention provides a collision detection method, device and vehicle to solve the defects in the prior art.

[0004] The present invention provides a collision detection method, comprising the following steps: Determine the associated objects of the vehicle from adjacent frame images of the driving area; Predicting a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; Based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, the collision result of the vehicle and the associated object is predicted, the starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision time.

[0005] According to a collision detection method provided by the present invention, the prediction of the collision moment between the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image includes: Determining a first collision time based on a pixel width of the associated object in each frame of image and a capture time of each frame of image; Determining a second collision time based on the distance between the associated object and the vehicle in each frame image and the acquisition time of each frame image; A collision moment between the vehicle and the associated object is determined based on the first collision moment and the second collision moment.

[0006] According to a collision detection method provided by the present invention, determining the collision moment between the vehicle and the associated object based on the first collision moment and the second collision moment includes: Based on a preset weight, the first collision moment and the second collision moment are weighted and added to determine a collision moment between the vehicle and the associated object.

[0007] According to a collision detection method provided by the present invention, the prediction of the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within a target duration and the motion trajectory of the associated object within the target duration includes: If the motion trajectory of the vehicle within the target duration overlaps with the motion trajectory of the associated object within the target duration, it is determined that there is a collision risk between the vehicle and the associated object.

[0008] According to a collision detection method provided by the present invention, the predicting of the collision result between the vehicle and the associated object further includes: Obtaining an actual collision result between the vehicle and the associated object; If the actual collision result is different from the collision result, a vehicle braking system parameter is adjusted based on the difference between the actual collision result and the collision result.

[0009] According to a collision detection method provided by the present invention, the obtaining of the actual collision result between the vehicle and the associated object comprises: Based on the collision detection model, applying the frame image of the driving area, determining the actual collision result between the vehicle and the associated object; The collision detection model is trained based on sample frame images of a sample vehicle in a sample driving area, sample associated objects of the sample vehicle, and collision result labels between the sample vehicle and the sample associated objects.

[0010] The present invention also provides a collision detection device, comprising the following modules: A determination unit, used to determine the associated object of the vehicle from adjacent frame images of the driving area; A prediction unit, configured to predict a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; A detection unit is used to predict the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, wherein the starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision time.

[0011] The present invention also provides a vehicle, comprising: a collision detection device as described above and a vehicle braking system, wherein the collision detection device is used to adjust the vehicle braking system parameters based on the difference between the actual collision result and the collision result when an actual collision result between the vehicle and the associated object is different from the collision result.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned collision detection methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the collision detection method described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the collision detection method described above is implemented.

[0015] The collision detection method, device and vehicle provided by the present invention capture the change of the motion trajectory of the associated object at a distance based on the pixel width of the associated object in each frame image, and capture the change of the motion trajectory of the associated object on a two-dimensional plane based on the distance between the associated object and the vehicle in each frame image, so that the change of the motion trajectory of the associated object can be more accurately determined in combination with the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, and then the collision moment of the vehicle and the associated object can be accurately predicted according to the change of the motion trajectory of the associated object. In addition, based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, the present invention can accurately determine whether the motion trajectory of the vehicle and the associated object overlap within the target time length, and then accurately predict the collision result of the vehicle and the associated object. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of the collision detection method provided by the present invention.

[0018] Figure 2 It is a flowchart diagram of an implementation of step 120 in the collision detection method provided by the present invention.

[0019] Figure 3 It is a schematic diagram of the OBB collision model provided by the present invention.

[0020] Figure 4 It is a schematic diagram of collision detection of each frame image provided by the present invention.

[0021] Figure 5 It is a schematic diagram for determining the collision moment provided by the present invention.

[0022] Figure 6 It is a structural schematic diagram of the collision detection device provided by the present invention.

[0023] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] At present, vehicle collision detection mainly relies on the intuitive judgment of the driver, that is, the driver intuitively predicts whether two vehicles or a vehicle and an obstacle will collide based on the visual image of the driving area. However, this intuition often deviates from the actual collision moment. In other words, the accuracy of collision detection based on the driver's own intuition is relatively low.

[0026] In this regard, the present invention provides a collision detection method, which is based on adjacent frame images of a driving area input externally, objectively reflects the pixel width of an associated object and the distance between a vehicle and the associated object from the adjacent frame images, and predicts the collision moment between the vehicle and the managed object based on the objectively reflected pixel width and distance, and compares the motion trajectories of the vehicle and the associated object within a target time period to determine whether the vehicle and the associated object will collide.

[0027] Figure 1 It is a flow chart of the collision detection method provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 and step 130 .

[0028] Step 110: Determine objects associated with the vehicle from adjacent frame images of the driving area.

[0029] Specifically, the driving area can be understood as the range of the vehicle driving, and the adjacent frame images refer to two or more images within a continuous time interval, and the adjacent frame images are used to represent the road information in the driving area, the size, shape, speed, motion trajectory and other information of the associated objects in the driving area. Optionally, a camera can be set in front of the vehicle to collect adjacent frame images of the driving area while the vehicle is driving in the driving area; or the personnel in the vehicle can collect adjacent frame images during the driving of the vehicle through a camera device, which is not specifically limited in the embodiment of the present invention.

[0030] It is understandable that after acquiring adjacent frame images, the adjacent frame images can be preprocessed, such as graying, denoising, image enhancement, geometric transformation, etc., to obtain high-quality adjacent frame images, so that collision detection can be accurately performed based on the high-quality adjacent frame images.

[0031] In addition, associated objects refer to objects that pose a potential risk of collision with the vehicle during the driving process. These objects may be other vehicles, pedestrians, non-motor vehicles, road obstacles, etc. After acquiring adjacent frame images of the driving area, image recognition can be performed on the adjacent frame images to identify multiple candidate objects, and the trajectory of each candidate object can be tracked based on the position change of each candidate object in the adjacent frame images to determine the motion trajectory of each candidate object. Based on the motion trajectory of each candidate object, the relative position relationship between each candidate object and the vehicle is determined. If any candidate object is in the same lane as the vehicle and is close to the vehicle, it is considered that the candidate object may have a risk of collision with the vehicle, and the candidate object can be used as an associated object.

[0032] In addition, the associated physics can also be determined by multi-frame tracking, that is, target detection is performed on adjacent frame images to obtain the target object in each frame image, and the similarity between the target objects in each frame image is calculated, so that objects with higher similarity in each frame image are associated, and the objects with higher similarity are the associated objects.

[0033] Step 120 : predicting the collision moment between the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image.

[0034] Specifically, the pixel width of the associated object refers to the number of horizontal pixels occupied by the associated object in each frame image, and the pixel width can be used to represent the actual width of the associated object. For example, the actual width of the associated object can be determined based on the following formula: ; in, Indicates the actual width of the associated object. Indicates the pixel width of the associated object, Indicates the actual distance between the associated object and the vehicle. Indicates the focal length of the camera.

[0035] Although the number of pixels of the associated object in the distance is small, any slight change in pixel width may represent a large movement change of the associated object in reality. Based on the pixel width of the associated object, the embodiment of the present invention can subtly perceive any slight change of the associated object in the distance, and thus can capture the potential danger of collision between the associated object and the vehicle as early as possible.

[0036] However, the change in pixel width of the associated object cannot accurately represent the complex movement of the object in the two-dimensional plane. For example, when the associated object rotates in the two-dimensional plane, the change will not be reflected in the pixel width, and thus the change in the motion trajectory of the associated object in the two-dimensional plane cannot be known through the pixel width.

[0037] Unlike the pixel width, the distance between the associated object and the vehicle in each frame image takes into account the position information of the associated object and the vehicle on the two-dimensional plane. Based on the positions of the two, the relative position relationship between the associated object and the vehicle on the two-dimensional plane can be determined. The change in the relative position relationship can determine the motion trajectory of the associated object, that is, the change in the motion trajectory of the associated object on the two-dimensional plane.

[0038] However, the distance between the associated object and the vehicle in each frame image focuses on the motion trajectory on the two-dimensional plane, and there may be measurement errors for the motion changes of the associated object in the distance. For example, when the motion of the associated object in the distance changes, the distance change between the associated object and the vehicle in each frame image may be very small, and thus the motion change of the associated object in the distance cannot be reflected, which leads to the inability to capture the potential collision risk of the associated object in the distance in time.

[0039] In summary, the embodiments of the present invention capture the changes in the motion trajectory of the associated object at a distance based on the pixel width of the associated object in each frame image, and capture the changes in the motion trajectory of the associated object in a two-dimensional plane based on the distance between the associated object and the vehicle in each frame image. Therefore, the changes in the motion trajectory of the associated object can be determined more accurately in combination with the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, and further, the collision moment of the vehicle and the associated object can be accurately predicted according to the changes in the motion trajectory of the associated object.

[0040] Exemplarily, the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image can be combined to determine the actual distance between the vehicle and the associated object and the relative speed between the vehicle and the associated object, and the collision moment of the vehicle and the associated object can be determined based on the actual distance and relative speed.

[0041] Step 130: predict the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within the target duration and the motion trajectory of the associated object within the target duration. The start time of the target duration is the acquisition time of the start frame image in the adjacent frame images, and the end time of the target duration is the collision time.

[0042] Specifically, the predicted collision moment refers to the moment when the vehicle and the associated object may collide when the vehicle and the associated object continue to move in the current motion state. The reason why the moment of possible collision is determined based on the relative speed and relative distance, that is, the collision indicates the moment when the vehicle and the associated object may collide at the current relative speed and relative distance, but the collision requires that the movement trajectories of the vehicle and the associated object need to overlap. If the vehicle only catches up with the associated object in parallel, but the movement trajectories of the two do not overlap, then the two have not collided.

[0043] In this regard, after determining the collision time, the embodiment of the present invention uses the acquisition time of the start frame image as the start time of the target duration, and the collision time as the end time of the target duration, to determine the motion trajectory of the vehicle within the target duration, and the motion trajectory of the associated object within the target duration. If the motion trajectories of the two overlap, it indicates that there is a collision risk between the vehicle and the associated object within the target duration. If the motion trajectories of the two do not overlap, it indicates that there is no collision risk between the vehicle and the associated object within the target duration.

[0044] The collision detection method provided by the embodiment of the present invention captures the change of the motion trajectory of the associated object at a distance based on the pixel width of the associated object in each frame image, and captures the change of the motion trajectory of the associated object on a two-dimensional plane based on the distance between the associated object and the vehicle in each frame image, so that the change of the motion trajectory of the associated object can be more accurately determined in combination with the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, and then the collision moment of the vehicle and the associated object can be accurately predicted based on the change of the motion trajectory of the associated object. In addition, based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, the embodiment of the present invention can accurately determine whether the motion trajectories of the vehicle and the associated object overlap within the target time length, and then accurately predict the collision result between the vehicle and the associated object.

[0045] Based on the above embodiments, Figure 2 is a flow chart of an implementation of step 120 in the collision detection method provided by the present invention, such as Figure 2 As shown, step 120 predicts the collision moment between the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, including: Step 121 : Determine the first collision time based on the pixel width of the associated object in each frame image and the acquisition time of each frame image.

[0046] Specifically, the acquisition time of each frame image can be understood as the timestamp corresponding to each frame image. Exemplarily, the first collision time can be determined here based on the pixel width of the associated object in two adjacent frames of images and the acquisition time of the two adjacent frames of images. It is also possible to select multiple groups of two adjacent frames of images, and calculate the first collision time corresponding to each group of two adjacent frames of images, and use the average value of the first collision time of each group as the final first collision time.

[0047] Taking the calculation of the first collision moment of two adjacent frames of images as an example, the first collision moment can be determined based on the following formula: ; in, represents the first collision moment, Indicates the acquisition time of the start frame, Indicates the acquisition time of the next frame corresponding to the start frame. Indicates the pixel width of the associated object in the starting frame, Indicates the pixel width of the associated object in the next frame corresponding to the start frame.

[0048] Optionally, before determining the first collision moment, each frame image may be projected as a perspective view, so that the change in the motion trajectory of the distant associated object may be subtly observed.

[0049] Step 122: Determine the second collision time based on the distance between the associated object and the vehicle in each frame image and the acquisition time of each frame image.

[0050] Specifically, the acquisition time of each frame image can be understood as the timestamp corresponding to each frame image. Exemplarily, the second collision time can be determined based on the distance between the associated object and the vehicle in two adjacent frames of images and the acquisition time of the two adjacent frames of images. It is also possible to select multiple groups of two adjacent frames of images, calculate the second collision time corresponding to each group of two adjacent frames of images, and use the average value of the second collision time of each group as the final second collision time.

[0051] Taking the calculation of the second collision moment of two adjacent frames of images as an example, the second collision moment can be determined based on the following formula: ; in, represents the second collision moment, Indicates the acquisition time of the start frame, Indicates the acquisition time of the next frame corresponding to the start frame. represents the distance between the associated object and the vehicle in the starting frame, Indicates the distance between the associated object and the vehicle in the next frame corresponding to the start frame.

[0052] Optionally, in order to improve the ability to perceive, understand and predict the surrounding environment, before determining the second collision moment, a Bird's Eye View (BEV) transformation can be performed on adjacent frame images, that is, the adjacent frame images are converted into a perspective that is centered on the vehicle and looks down at the vehicle's surroundings from a high altitude.

[0053] Step 123: Determine the collision moment between the vehicle and the associated object based on the first collision moment and the second collision moment.

[0054] Specifically, the first collision time is determined based on the pixel width of the associated object in each frame image, and thus the first collision time can accurately represent the collision risk of the associated object at a distance. The second collision time is determined based on the distance between the associated object and the vehicle in each frame image, and thus the second collision time can accurately represent the collision risk of the associated object on a two-dimensional plane.

[0055] In this regard, the embodiment of the present invention determines the collision moment between the vehicle and the associated object in combination with the first collision moment and the second collision moment, taking into account both the collision risk of the associated object at a distance and the collision risk of the associated object on a two-dimensional plane.

[0056] Optionally, the nearest collision moment can be selected from the first collision moment and the second collision moment as the collision moment between the vehicle and the associated object, or the average of the first collision moment and the second collision moment can be used as the collision moment between the vehicle and the associated object. This embodiment of the present invention does not specifically limit this.

[0057] Based on any of the above embodiments, determining the collision moment between the vehicle and the associated object based on the first collision moment and the second collision moment includes: Based on the preset weights, the first collision moment and the second collision moment are weighted and added to determine the collision moment between the vehicle and the associated object.

[0058] Specifically, although the first collision moment can characterize the collision risk of the associated object at a distance, the first collision moment cannot characterize the collision risk of the associated object on a two-dimensional plane. Although the second collision moment can characterize the collision risk of the associated object on a two-dimensional plane, the second collision moment cannot reflect the collision risk of the associated object at a distance. In this regard, the embodiment of the present invention adds weights to the first collision moment and the second collision moment based on preset weights, so that the determined collision moment can reflect both the collision risk of the associated object at a distance and the collision risk of the associated object on a two-dimensional plane.

[0059] Optionally, the collision moment of the vehicle and the associated object may be determined based on the following formula: ; in, Indicates the collision moment between the vehicle and the associated object, represents the first collision moment, represents the second collision moment, represents the weight corresponding to the first collision moment, represents the weight corresponding to the second collision moment, where .

[0060] Based on any of the above embodiments, based on the motion trajectory of the vehicle within the target duration and the motion trajectory of the associated object within the target duration, predicting the collision result between the vehicle and the associated object includes: If the motion trajectory of the vehicle within the target duration overlaps with the motion trajectory of the associated object within the target duration, it is determined that there is a risk of collision between the vehicle and the associated object.

[0061] Specifically, the overlap here can be understood as the position of the vehicle being the same as the position of the associated object at any time within the target duration. If the position of the vehicle is the same as the position of the associated object at any time within the target duration, it indicates that there is a risk of collision between the vehicle and the associated object. Conversely, if there is no time within the target duration when the position of the vehicle is the same as the position of the associated object, it indicates that there is no risk of collision between the vehicle and the associated object.

[0062] Figure 3 is a schematic diagram of the OBB collision model provided by the present invention, such as Figure 3 As shown in the figure, OBB (Oriented Bounding Box) is a bounding box established according to the shape and orientation of the object. The OBB corresponding to the vehicle and the target object within the target duration is determined, and the two OBBs are projected onto the two axes of the coordinate axis to determine whether the OBBs projected on the two axes overlap. If so, it indicates that there is a risk of collision between the vehicle and the associated object. If not, it indicates that there is no risk of collision between the vehicle and the associated object. Figure 3 It can be seen that the OBBs projected on the two axes overlap, which determines that there is a risk of collision between the vehicle and the associated object.

[0063] Figure 4 Schematic diagram of collision detection of each frame image provided by the present invention, such as Figure 4 As shown, the corresponding frame images are collected every 0.1s within the target time, and each frame image is traversed to determine whether the OBBs corresponding to the vehicle and the associated object in each frame image overlap. If so, it is determined that there is a collision, otherwise, it is determined that there is no collision.

[0064] In addition, with the development of autonomous driving technology, the automatic emergency braking system (AEB) is an important safety function, and its performance accuracy and reliability are crucial to ensuring driving safety. However, AEB may be triggered by mistake in some cases, resulting in unnecessary emergency braking, affecting the driving experience and even causing accidents. Among them, the reason for AEB false triggering is the wrong estimation of the collision result, that is, the collision result is inconsistent with the actual collision result.

[0065] Therefore, it is necessary to evaluate the AEB system to confirm whether there is a false triggering situation. At present, the evaluation of the AEB system mainly relies on real vehicle testing, but this method is costly, inefficient, and difficult to cover all possible false triggering scenarios. In addition, real vehicle testing often cannot fully simulate the complex environment of false triggering, resulting in limitations in the evaluation results.

[0066] In this regard, the embodiment of the present invention uses the above embodiment method to determine the collision result, and compares the collision result with the actual collision result to evaluate whether the AEB system may be falsely triggered. This method does not require actual vehicle testing, which greatly saves testing costs. Specifically, the evaluation method is as follows: Predict the collision results between the vehicle and the associated objects, and then include: Obtain the actual collision results between the vehicle and the associated object; If the actual collision result is different from the collision result, the vehicle braking system parameters are adjusted based on the actual collision result and the difference between the collision result.

[0067] Specifically, the actual collision result refers to the actual result of whether the vehicle collides with the associated object. The actual collision result can be obtained through simulation testing or through monitoring by sensors (such as radar) configured on the vehicle.

[0068] If the actual collision result is different from the collision result, it indicates that the vehicle braking system (such as the AEB system) may be triggered by mistake. In this case, the difference between the actual collision result and the collision result can be analyzed, the parameters that need to be adjusted can be identified, and the parameters of the vehicle braking system can be adjusted. The adjusted parameters may include braking force distribution, braking response time, braking distance, etc.

[0069] It can be seen that the embodiment of the present invention avoids false triggering of the vehicle braking system by comparing the actual collision result with the collision result and adjusting the vehicle braking system parameters when the two are inconsistent.

[0070] Based on any of the above embodiments, obtaining an actual collision result between the vehicle and the associated object includes: Based on the collision detection model, the frame image of the driving area is applied to determine the actual collision result between the vehicle and the associated object; The collision detection model is trained based on sample frame images of sample vehicles in sample driving areas, sample associated objects of sample vehicles, and labels of collision results between sample vehicles and sample associated objects.

[0071] Specifically, the sample frame images of the sample vehicle in the sample driving area can be obtained from an open source dataset, such as CNCAP2024, and the scenes involved may include complex environments such as urban roads, highways, rainy and snowy weather, etc. Among them, in selecting the sample frame images, the frame images corresponding to the manual emergency brake can be eliminated to avoid interference with subsequent detection results.

[0072] After collecting sample frame images, the sample frame images are annotated, including obstacle and lane line annotations. Obstacle annotations include vehicles, pedestrians, bicycles and other objects that may interact with the vehicle. Lane line annotations are used to determine the vehicle's driving path and lane boundaries.

[0073] In addition to the above-mentioned annotations, the collision result between the sample vehicle and the sample-associated object, that is, the collision result label, is also required. The label can be the collision moment between the sample vehicle and the sample-associated object. The collision moment can be determined based on the relative distance and relative speed between the sample vehicle and the sample-associated object.

[0074] In addition, before training based on sample frame images of sample vehicles in sample driving areas, sample associated objects of sample vehicles, and collision result labels between sample vehicles and sample associated objects, a field of interest (FOI) can be annotated in the sample frame images. The field of interest usually includes an area within a certain distance and angle range in front of the sample vehicle, so that the sample associated object can be determined by focusing on the area.

[0075] After the collision detection model is trained, the actual collision result between the vehicle and the associated object can be determined based on the collision detection model and the frame image of the driving area.

[0076] Based on any of the above embodiments, the present invention further provides a collision detection method, comprising: Figure 5 Schematic diagram of determining the collision moment provided by the present invention, such as Figure 5As shown, the associated objects of the vehicle are determined from the adjacent frame images of the driving area by means of multi-frame tracking. In the perspective view, the first collision moment is determined based on the pixel width (widthT0 and widthT1) of the associated objects in each frame image, and the acquisition time (T0 and T1) of each frame image. In the BEV perspective, the second collision moment is determined based on the distance (DepthT0 and DepthT1) between the associated object and the vehicle in each frame image, and the acquisition time of each frame image. Then, based on the preset weights, the weights of the first collision moment and the second collision moment are added to determine the collision moment between the vehicle and the associated object.

[0077] During the target duration, the corresponding frame images are collected every 0.1s, and each frame image is traversed to determine whether the OBBs corresponding to the vehicle and the associated object on each frame image overlap. If so, it is determined that there is a collision, if not, it is determined that there is no collision, and finally the collision result between the vehicle and the associated object is obtained. Among them, the starting time of the target duration is the collection time of the starting frame image in the adjacent frame images, and the end time of the target duration is the collision time.

[0078] Next, based on the collision detection model, the frame image of the driving area is used to determine the actual collision result between the vehicle and the associated object; wherein the collision detection model is trained based on the sample frame image of the sample vehicle in the sample driving area, the sample associated object of the sample vehicle, and the collision result label between the sample vehicle and the sample associated object.

[0079] If the actual collision result is different from the collision result, the vehicle braking system parameters are adjusted based on the actual collision result and the difference between the collision result.

[0080] The collision detection device provided by the present invention is described below. The collision detection device described below and the collision detection method described above can be referenced to each other.

[0081] Based on any of the above embodiments, Figure 6 is a schematic diagram of the structure of the collision detection device provided by the present invention, such as Figure 6 As shown, the device comprises: A determination unit 610, configured to determine an associated object of the vehicle from adjacent frame images of the driving area; A prediction unit 620, configured to predict a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; The detection unit 630 is used to predict the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length. The starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision time.

[0082] The collision detection device provided by the embodiment of the present invention captures the change of the motion trajectory of the associated object at a distance based on the pixel width of the associated object in each frame image, and captures the change of the motion trajectory of the associated object on a two-dimensional plane based on the distance between the associated object and the vehicle in each frame image, so that the change of the motion trajectory of the associated object can be more accurately determined in combination with the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, and then the collision moment of the vehicle and the associated object can be accurately predicted according to the change of the motion trajectory of the associated object. In addition, based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, the embodiment of the present invention can accurately determine whether the motion trajectories of the vehicle and the associated object overlap within the target time length, and then accurately predict the collision result between the vehicle and the associated object.

[0083] Based on any of the above embodiments, based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image, predicting the collision moment between the vehicle and the associated object includes: Determine the first collision time based on the pixel width of the associated object in each frame image and the acquisition time of each frame image; Determine the second collision time based on the distance between the associated object and the vehicle in each frame image and the acquisition time of each frame image; Based on the first collision moment and the second collision moment, a collision moment between the vehicle and the associated object is determined.

[0084] Based on any of the above embodiments, determining the collision moment between the vehicle and the associated object based on the first collision moment and the second collision moment includes: Based on the preset weights, the first collision moment and the second collision moment are weighted and added to determine the collision moment between the vehicle and the associated object.

[0085] Based on any of the above embodiments, based on the motion trajectory of the vehicle within the target duration and the motion trajectory of the associated object within the target duration, predicting the collision result between the vehicle and the associated object includes: If the motion trajectory of the vehicle within the target duration overlaps with the motion trajectory of the associated object within the target duration, it is determined that there is a risk of collision between the vehicle and the associated object.

[0086] Based on any of the above embodiments, predicting the collision result between the vehicle and the associated object, and then further comprising: Obtain the actual collision results between the vehicle and the associated object; If the actual collision result is different from the collision result, the vehicle braking system parameters are adjusted based on the actual collision result and the difference between the collision result.

[0087] Based on any of the above embodiments, obtaining an actual collision result between the vehicle and the associated object includes: Based on the collision detection model, the frame image of the driving area is applied to determine the actual collision result between the vehicle and the associated object; The collision detection model is trained based on sample frame images of sample vehicles in sample driving areas, sample associated objects of sample vehicles, and labels of collision results between sample vehicles and sample associated objects.

[0088] Based on any of the above embodiments, the present invention also provides a vehicle, comprising: a collision detection device and a vehicle braking system as described in any of the above embodiments, the collision detection device being used to adjust vehicle braking system parameters based on the difference between the actual collision result and the collision result when an actual collision result between the vehicle and an associated object is different from the collision result.

[0089] Specifically, if the actual collision result is different from the collision result, it indicates that the vehicle braking system (such as the AEB system) may be triggered by mistake. In this case, the difference between the actual collision result and the collision result can be analyzed, the parameters that need to be adjusted can be identified, and the parameters of the vehicle braking system can be adjusted. The adjusted parameters may include braking force distribution, braking response time, braking distance, etc.

[0090] It can be seen that the embodiment of the present invention avoids false triggering of the vehicle braking system by comparing the actual collision result with the collision result and adjusting the vehicle braking system parameters when the two are inconsistent.

[0091] Figure 7 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the collision detection method, which includes: determining the associated object of the vehicle from the adjacent frame images of the driving area; predicting the collision moment of the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image; predicting the collision result of the vehicle and the associated object based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, wherein the starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision moment.

[0092] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0093] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the collision detection method provided by the above-mentioned methods, which includes: determining an associated object of the vehicle from adjacent frame images of a driving area; predicting a collision moment between the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image; predicting a collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within a target time length and the motion trajectory of the associated object within the target time length, the starting time of the target time length being the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length being the collision moment.

[0094] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the collision detection method provided by the above-mentioned methods, the method comprising: determining an associated object of a vehicle from adjacent frame images of a driving area; predicting a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; predicting a collision result between the vehicle and the associated object based on a motion trajectory of the vehicle within a target time length and a motion trajectory of the associated object within the target time length, the starting time of the target time length being the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length being the collision moment.

[0095] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collision detection method, characterized in that: include: Determine the associated objects of the vehicle from adjacent frame images of the driving area; Predicting a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; Based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, the collision result of the vehicle and the associated object is predicted, the starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision time.

2. The collision detection method according to claim 1, characterized in that: The predicting of the collision moment between the vehicle and the associated object based on the pixel width of the associated object in each frame image and the distance between the associated object and the vehicle in each frame image includes: Determining a first collision time based on a pixel width of the associated object in each frame of image and a capture time of each frame of image; Determining a second collision time based on the distance between the associated object and the vehicle in each frame image and the acquisition time of each frame image; A collision moment between the vehicle and the associated object is determined based on the first collision moment and the second collision moment.

3. The collision detection method according to claim 2, characterized in that: The determining the collision moment between the vehicle and the associated object based on the first collision moment and the second collision moment includes: Based on a preset weight, the first collision moment and the second collision moment are weighted and added to determine a collision moment between the vehicle and the associated object.

4. The collision detection method according to any one of claims 1 to 3, characterized in that: The predicting the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within the target duration and the motion trajectory of the associated object within the target duration includes: If the motion trajectory of the vehicle within the target duration overlaps with the motion trajectory of the associated object within the target duration, it is determined that there is a collision risk between the vehicle and the associated object.

5. The collision detection method according to any one of claims 1 to 3, characterized in that: The predicting of the collision result between the vehicle and the associated object further includes: Obtaining an actual collision result between the vehicle and the associated object; If the actual collision result is different from the collision result, a vehicle braking system parameter is adjusted based on the difference between the actual collision result and the collision result.

6. The collision detection method according to claim 5, characterized in that: The obtaining of the actual collision result between the vehicle and the associated object comprises: Based on the collision detection model, applying the frame image of the driving area, determining the actual collision result between the vehicle and the associated object; The collision detection model is trained based on sample frame images of a sample vehicle in a sample driving area, sample associated objects of the sample vehicle, and collision result labels between the sample vehicle and the sample associated objects.

7. A collision detection device, characterized in that: include: A determination unit, used to determine the associated object of the vehicle from adjacent frame images of the driving area; A prediction unit, configured to predict a collision moment between the vehicle and the associated object based on a pixel width of the associated object in each frame image and a distance between the associated object and the vehicle in each frame image; A detection unit is used to predict the collision result between the vehicle and the associated object based on the motion trajectory of the vehicle within the target time length and the motion trajectory of the associated object within the target time length, wherein the starting time of the target time length is the acquisition time of the starting frame image in the adjacent frame images, and the ending time of the target time length is the collision time.

8. A vehicle, characterized in that: include: The collision detection device and vehicle braking system as described in claim 7, wherein the collision detection device is used to adjust the vehicle braking system parameters based on the difference between the actual collision result and the collision result when the actual collision result between the vehicle and the associated object is different from the collision result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the collision detection method according to any one of claims 1 to 6 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the collision detection method according to any one of claims 1 to 6 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the collision detection method according to any one of claims 1 to 6 is implemented.

Images