Target detection method, device, vehicle and storage medium

Abstract

The present disclosure relates to a target detection method, device, vehicle and storage medium, and belongs to the field of autonomous driving and target detection technology. The method includes: obtaining a first image information set, wherein the first image information set includes M frames of first image information; determining the first image information of each of N target detection networks from the first image information set based on preset constraints, wherein the constraint condition is that the number of overlaps of the first image information of any two target detection networks is minimized; distributing the first image information of each of the N target detection networks. Thus, the first image information set can be distributed in consideration of the constraints, so that the distributed first image information satisfies the constraint that the number of overlaps of the first image information of any two target detection networks is minimized, that is, the first image information processed by any two target detection networks should not overlap as much as possible, and the staggered distribution of the first image information set can be achieved, which helps to improve the efficiency of target detection.

Description

Technical Field

[0001] The present disclosure relates to the fields of autonomous driving and target detection technology, and in particular to a target detection method, device, vehicle and storage medium. Background Art

[0002] At present, target detection technology has been widely used in the fields of autonomous driving, security monitoring, etc. For example, in the field of autonomous driving, vehicles, pedestrians, traffic signals, etc. on the road can be detected based on target detection technology, and in the field of security monitoring, abnormal behaviors can be detected based on target detection technology. However, the target detection methods in related technologies have the problem of low target detection efficiency. Summary of the invention

[0003] The present disclosure provides a target detection method, device, vehicle, and computer-readable storage medium to at least solve the problem of low target detection efficiency in the related art. The technical solution of the present disclosure is as follows:

[0004] According to a first aspect of an embodiment of the present disclosure, a target detection method is provided, comprising: acquiring a first image information set, wherein the first image information set comprises M frames of first image information, where M is a positive integer; determining, from the first image information set, first image information of each of N target detection networks based on a preset constraint condition, wherein the constraint condition is that the number of overlaps between the first image information of each of any two target detection networks is minimum, and N is a positive integer; and distributing the first image information of each of the N target detection networks.

[0005] According to a second aspect of an embodiment of the present disclosure, a target detection device is provided, including: an acquisition module, configured to acquire a first image information set, wherein the first image information set includes M frames of first image information, M is a positive integer; a determination module, configured to determine the first image information of each of N target detection networks from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps between the first image information of any two target detection networks is minimum, and N is a positive integer; a distribution module, configured to distribute the respective first image information to the N target detection networks.

[0006] According to a third aspect of an embodiment of the present disclosure, a vehicle is provided, comprising a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the steps of the method described in the first aspect of the embodiment of the present disclosure.

[0007] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the method described in the first aspect of the embodiment of the present disclosure are implemented.

[0008] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: the first image information set can be distributed taking into account the constraint conditions, so that the distributed first image information satisfies the constraint that the number of overlaps of the first image information of any two target detection networks is minimized, that is, the first image information processed by any two target detection networks try not to overlap, and the staggered distribution of the first image information set can be achieved, which helps to improve the target detection efficiency. In addition, the first image information set can be processed by multiple target detection networks, which helps to improve the resource utilization of the target detection network, and can achieve multi-classification target detection, which is suitable for traffic signal detection scenarios. In addition, each target detection network only needs to process the distributed first image information, and there is no need to process the entire first image information set, which helps to reduce the amount of data that a single target detection network needs to process, helps to improve the target detection efficiency, and a single target detection network is more lightweight, which is suitable for application scenarios where multiple lightweight target detection networks are deployed at the same time.

[0009] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0011] Figure 1 The figure is a flow chart of a target detection method according to an exemplary embodiment.

[0012] Figure 2 is a flowchart of a target detection method according to another exemplary embodiment.

[0013] Figure 3 is a flowchart of a target detection method according to another exemplary embodiment.

[0014] Figure 4 is a flowchart of a target detection method according to another exemplary embodiment.

[0015] Figure 5 is a schematic diagram of a target detection model according to an exemplary embodiment.

[0016] Figure 6 is a flowchart of a target detection method according to another exemplary embodiment.

[0017] Figure 7 The figure is a block diagram of a target detection device according to an exemplary embodiment.

[0018] Figure 8 is a block diagram of a vehicle according to an exemplary embodiment. DETAILED DESCRIPTION

[0019] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0021] Figure 1 is a flow chart of a target detection method according to an exemplary embodiment. Figure 1 As shown, the target detection method of the embodiment of the present disclosure includes the following steps.

[0022] S101, obtaining a first image information set, wherein the first image information set includes M frames of first image information, where M is a positive integer.

[0023] It should be noted that the executor of the target detection method of the embodiment of the present disclosure is an electronic device, such as a vehicle-mounted terminal, a vehicle-mounted controller, a server, an ISP (Image Signal Processor), etc. The target detection method of the embodiment of the present disclosure can be executed by the target detection device of the embodiment of the present disclosure, and the target detection device of the embodiment of the present disclosure can be configured in any electronic device to execute the target detection method of the embodiment of the present disclosure.

[0024] It should be noted that there are no excessive restrictions on the first image information, for example, it may include the original image, the result after the original image is processed, etc. The result after the original image is processed may include the image after the original image is processed, the image features of the original image (such as feature maps), initial detection results, etc. The initial detection results may include whether the original image contains the set object, the category of the pixels in the original image, the category of the image area of ​​the original image, and the detection frame of the set object in the original image, etc. There are no excessive restrictions on the processing method of the original image, for example, it may include feature extraction, noise reduction, brightness adjustment, normalization, color space conversion, cropping, rotation, flipping, etc. There are no excessive restrictions on the original image and the image after the original image is processed, for example, it may include RGB, HSV, HSL, YCbCr, Lab, YUV, etc.

[0025] S102, determining first image information of each of N target detection networks from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps of the first image information of any two target detection networks is minimal, and N is a positive integer.

[0026] It should be noted that there are no excessive restrictions on the target detection network.

[0027] Exemplarily, the target detection network may include a single-stage detector, a two-stage detector, an anchor-free detector, a Transformer-based detector, etc. It should be noted that the Transformer model is a neural network model based on a self-attention mechanism.

[0028] Exemplarily, the target detection network may include an inference network, a post-processing network, etc. It should be noted that the inference network is used to perform feature extraction, image classification, detection frame regression and other processing on the first image information to obtain an initial detection result, and the post-processing network is used to post-process the initial detection result to obtain a final detection result. There is no excessive limitation on the specific post-processing method, for example, it may include tracking smoothing (such as Kalman filtering, particle filtering, trajectory interpolation, etc.), NMS (Non-Maximum Suppression), threshold filtering (such as filtering out detection frames with scores below a certain threshold), detection frame cropping, category filtering (such as filtering out detection results of certain categories), etc. The initial detection result and the final detection result may include whether the image contains a set object, the category of pixels in the image, the category of the image area, and the detection frame of the set object.

[0029] Exemplarily, the target detection network may include a traffic signal detection network, an obstacle detection network, etc. It should be noted that the traffic signal detection network is used to detect signal lights, traffic signs, lane lines, etc., and the obstacle detection network is used to detect static obstacles (such as roadblocks, buildings, etc.), dynamic obstacles (such as vehicles, pedestrians, animals, etc.), etc.

[0030] Exemplarily, the target detection processes of multiple subnetworks in the target detection network are serial, and the frame rates of the multiple subnetworks in the target detection network are consistent. It is understandable that an target detection network can be composed of multiple subnetworks with consistent frame rates. The frame rate refers to the number of frames processed per second, which can be expressed by FPS (Frames Per Second). For example, a frame rate of 30 FPS is used to represent that the number of frames processed per second is 30, that is, 30 frames are processed per second.

[0031] For example, target detection network 1 is a traffic signal detection network, which may include a signal light detection subnetwork, a traffic sign detection subnetwork and a lane line detection subnetwork; target detection network 2 is an obstacle detection network, which may include a static obstacle detection subnetwork, a vehicle detection subnetwork, a pedestrian detection subnetwork and an animal detection subnetwork.

[0032] It is understandable that the first image information of the target detection network refers to the image information to be processed by the target detection network, that is, the image information distributed to the target detection network. Different target detection networks may have different first image information, and there may be overlap between the first image information of different target detection networks, and the number of first image information of different target detection networks may be different.

[0033] It is understandable that the first image information sets at different moments may be different, and the first image information of the same target detection network at different moments may be different. Determining the first image information of each of the N target detection networks from the first image information set based on a preset constraint condition may include determining the first image information of each of the N target detection networks at the tth moment from the first image information set at the tth moment based on the constraint condition.

[0034] In one embodiment, the first image information of each of the N target detection networks is determined based on preset constraints from the first image information set, including constructing an objective function between the first image information of each of the N target detection networks and the number of overlaps between the first image information of each of any two target detection networks, optimizing the objective function, and obtaining the first image information of each of the N target detection networks when the objective function is minimized as the final first image information of each of the N target detection networks. Thus, the objective function can be constructed, and the first image information of each of the N target detection networks when the objective function is minimized can be solved as the final first image information of each of the N target detection networks.

[0035] It should be noted that the optimization of the objective function can be achieved by using any function optimization method in the relevant technology, and no further limitation is given here. The function optimization method may include gradient descent, Newton method, quasi-Newton method, conjugate gradient method, trust region method, etc.

[0036] In one embodiment, first image information of each of N target detection networks is determined from the first image information set based on preset constraints, including constructing a search space based on the first image information set, performing parameter search in the search space based on the constraints, and obtaining the first image information of each of the N target detection networks.

[0037] S103: Distribute respective first image information to N object detection networks.

[0038] For example, taking N=3, M=30, 3 target detection networks are all inference networks, and the first image information is an image feature, the first image information set includes 30 frames of image features. From the first image information set, based on the preset constraints, the first image information of the inference network 1 can be determined as the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames of image features, and the first image information of the inference network 2 can be determined as the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames of image features, and the first image information of the inference network 3 can be determined as the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frames of image features. It can be understood that at this time, the number of overlaps of the first image information of any two inference networks is 0.

[0039] The image features of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames can be distributed to inference network 1, the image features of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames can be distributed to inference network 2, and the image features of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frames can be distributed to inference network 3.

[0040] For example, taking N=3, M=30, 3 target detection networks are all inference networks, and the first image information is an image feature, the first image information set includes 30 frames of image features. From the first image information set, based on the preset constraints, the first image information of the inference network 1 can be determined as the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames of image features, and the first image information of the inference network 2 can be determined as the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames of image features, and the first image information of the inference network 3 can be determined as the 6th, 9th, 12th, 15th, 18th, 21st, 24th, and 27th frames of image features. It can be understood that at this time, the number of overlaps of the first image information of any two inference networks is 0.

[0041] The image features of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frames can be distributed to inference network 1, the image features of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames can be distributed to inference network 2, and the image features of the 6th, 9th, 12th, 15th, 18th, 21st, 24th, and 27th frames can be distributed to inference network 3.

[0042] In one implementation, the target detection processes of the N target detection networks are performed in parallel, which helps to improve target detection efficiency.

[0043] The target detection method provided by the embodiments of the present disclosure obtains a first image information set, wherein the first image information set includes M frames of first image information, M is a positive integer, and the first image information of each of N target detection networks is determined from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps between the first image information of any two target detection networks is minimal, N is a positive integer, and the first image information of each of the N target detection networks is distributed to each of the N target detection networks. Therefore, the first image information set can be distributed taking into account the constraints, so that the distributed first image information satisfies the constraint that the number of overlaps of the first image information of any two target detection networks is minimized, that is, the first image information processed by any two target detection networks try not to overlap, and the staggered distribution of the first image information set can be achieved, which helps to improve the target detection efficiency. In addition, multiple target detection networks can be used to process the first image information set, which helps to improve the resource utilization of the target detection network, and can achieve multi-classification target detection, which is suitable for traffic signal detection scenarios. In addition, each target detection network only needs to process the distributed first image information, and there is no need to process the entire first image information set, which helps to reduce the amount of data that a single target detection network needs to process, helps to improve the target detection efficiency, and a single target detection network is more lightweight, which is suitable for application scenarios where multiple lightweight target detection networks are deployed at the same time.

[0044] Figure 2is a flow chart of a target detection method according to another exemplary embodiment. Figure 2 As shown, the target detection method of the embodiment of the present disclosure includes the following steps.

[0045] S201, obtaining a first image information set, wherein the first image information set includes M frames of first image information, where M is a positive integer.

[0046] The relevant contents of step S201 can be found in the above embodiment and will not be described again here.

[0047] S202, obtaining the frame rates of the N target detection networks.

[0048] S203: Determine the number of frames K of the first image information of the i-th target detection network based on the frame rate of the i-th target detection network, where i is a positive integer not greater than N, and K is a positive integer.

[0049] It is understandable that different target detection networks may have different frame rates, and the number of frames K of the first image information of different target detection networks may be different.

[0050] In one embodiment, determining the number of frames K of the first image information of the i-th target detection network based on the frame rate of the i-th target detection network includes determining the number of frames K of the first image information of the i-th target detection network to be X in response to the frame rate of the i-th target detection network being X FPS.

[0051] For example, in response to the frame rate of the target detection network 1 being 10 FPS, the frame number K of the first image information of the target detection network 1 is determined to be 10.

[0052] For example, in response to the frame rate of the target detection network 2 being 30 FPS, the frame number K of the first image information of the target detection network 2 is determined to be 30.

[0053] S204: Determine, from the first image information set, K frames of first image information of the i-th object detection network based on the constraint condition.

[0054] For example, taking N=3, M=30, and the three target detection networks are all inference networks, and the first image information is an image feature, the first image information set includes 30 frames of image features, and the frame rates of inference networks 1 to 3 are all 10FPS, then the number of frames K of the first image information of inference networks 1 to 3 is all 10.

[0055] From the first image information set, based on the preset constraints, it can be determined that the 10 frames of first image information of the inference network 1 are the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frame image features, and the 10 frames of first image information of the inference network 2 are the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frame image features, and the 10 frames of first image information of the inference network 3 are the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frame image features. It can be understood that at this time, the number of overlaps of the first image information of any two inference networks is 0.

[0056] For example, taking N=3, M=30, and the three target detection networks are all inference networks, and the first image information is an image feature, the first image information set includes 30 frames of image features, and the frame rates of inference networks 1 to 3 are 10FPS, 10FPS, and 8FPS, respectively. Then the frame numbers K of the first image information of inference networks 1 to 3 are 10, 10, and 8, respectively.

[0057] From the first image information set, based on the preset constraints, it can be determined that the 10 frames of first image information of the inference network 1 are the image features of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames, and the 10 frames of first image information of the inference network 2 are the image features of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames, and the 8 frames of first image information of the inference network 3 are the image features of the 6th, 9th, 12th, 15th, 18th, 21st, 24th, and 27th frames. It can be understood that at this time, the number of overlaps of the first image information of any two inference networks is 0.

[0058] In one embodiment, K frames of first image information of the i-th target detection network are determined from the first image information set based on constraints, including consecutively numbering the M frames of first image information in the first image information set according to numbers 1 to M, and in response to M being an integer multiple of K, performing a modulo operation on the number of the first image information and M / K, and using the K frames of first image information with a remainder of zero as the K frames of first image information of the i-th target detection network.

[0059] For example, taking M=30, if the frame rate of the target detection network 1 is 10FPS, the frame number K of the first image information of the target detection network 1 is 10, and the 30 frames of first image information in the first image information set can be numbered consecutively from 1 to 30, and the first image information numbered 3, 6, 9, 12, 15, 18, 21, 24, 27, and 30 can be used as 10 frames of first image information of the target detection network 1.

[0060] For example, taking M = 30 as an example, if the frame rate of the target detection network 2 is 15 FPS, then the number of frames K of the first image information of the target detection network 1 is 15. The 30 frames of the first image information in the first image information set can be continuously numbered according to the numbers from 1 to 30. The first image information with the numbers 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30 can be used as the 15 frames of the first image information of the target detection network 2.

[0061] S205. Distribute the respective first image information to N target detection networks.

[0062] For the relevant content of step S205, reference can be made to the above-mentioned embodiments, which will not be elaborated here.

[0063] The target detection method provided by the embodiments of the present disclosure obtains the respective frame rates of N target detection networks, and determines the number of frames K of the first image information of the i-th target detection network based on the frame rate of the i-th target detection network, where i is a positive integer not greater than N, and K is a positive integer. From the first image information set, K frames of the first image information of the i-th target detection network are determined based on the constraint conditions. Thus, the frame rate of the target detection network can also be considered to distribute the first image information set, so that the distributed first image information meets the frame rate constraints of each target detection network.

[0064] Figure 3 is a flowchart of a target detection method shown according to another exemplary embodiment. As Figure 3 shown, the target detection method of the embodiments of the present disclosure includes the following steps.

[0065] S301. Obtain a first image information set, where the first image information set includes M frames of first image information, and M is a positive integer.

[0066] S302. Obtain the respective frame rates of N target detection networks.

[0067] S303. Determine the number of frames K of the first image information of the i-th target detection network based on the frame rate of the i-th target detection network, where i is a positive integer not greater than N, and K is a positive integer.

[0068] For the relevant content of steps S301 - S303, reference can be made to the above-mentioned embodiments, which will not be elaborated here.

[0069] S304. In response to K < M, extract K frames of the first image information from the first image information set at intervals of the target number of frames based on the constraint conditions as the K frames of the first image information of the i-th target detection network.

[0070] It should be noted that the target number of frames is not overly limited. For example, it can be 1, 2, 3, etc.

[0071] In one embodiment, the method further includes, in response to M being an integer multiple of K, using (M / K-1) as the target frame number, and in response to M not being an integer multiple of K, using a maximum positive integer not greater than (M / K-1) as the target frame number.

[0072] For example, taking M=30 and K=10 as an example, 30 / 10-1=2, 2 can be used as the target frame number, and 10 frames of first image information can be extracted from the first image information set at intervals of 2 frames based on the constraint condition as the 10 frames of first image information of the target detection network. For example, the image features of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames can be used as the 10 frames of first image information of the target detection network.

[0073] For example, taking M=30 and K=8 as an example, the largest positive integer not greater than (30 / 8-1) is 2, and 2 can be used as the target frame number. From the first image information set, 8 frames of first image information are extracted at intervals of 2 frames based on the constraint condition as the 8 frames of first image information of the target detection network. For example, the image features of the 6th, 9th, 12th, 15th, 18th, 21st, 24th, and 27th frames can be used as the 8 frames of first image information of the target detection network.

[0074] In one embodiment, the method also includes extracting Q consecutive frames of first image information from the first image information set as the first frame first image information of each of the Q target detection networks in response to the consistent frame rates of each of the Q target detection networks, wherein Q is a positive integer greater than 1, so that the first frame first image information of each of the multiple target detection networks with the consistent frame rate is different, so that the consecutive multiple frames of first image information in the first image information set are staggered and distributed to the multiple target detection networks with the consistent frame rate, so that the first image information processed by any two target detection networks with the consistent frame rate do not overlap.

[0075] It can be understood that Q is a positive integer greater than 1 and not greater than N.

[0076] Exemplarily, extracting consecutive Q frames of first image information from the first image information set as the first frame of first image information of each of the Q target detection networks includes extracting the first Q frames of first image information from the first image information set as the first frame of first image information of each of the Q target detection networks.

[0077] For example, taking N = Q = 3, M = 30, all 3 target detection networks being inference networks, and the first image information being image features as an example, the first image information set includes 30 frames of image features, the frame rates of inference networks 1 to 3 are all 10 FPS, then the number of frames K of the first image information of inference networks 1 to 3 is all 10. Three consecutive frames of the first image information can be extracted from the first image information set as the first frame of the first image information for each of inference networks 1 to 3. For example, the 1st to 3rd frames of the first image information can be used as the first frame of the first image information for each of inference networks 1 to 3.

[0078] 30 / 10 - 1 = 2. 2 can be used as the target number of frames. From the first image information set, 10 frames of the first image information are extracted at an interval of 2 frames based on the constraint condition as the 10 frames of the first image information of inference network i, where i is a positive integer not greater than 3.

[0079] For example, the first frame of the first image information of inference network 1 is the 1st frame of the first image information, the first frame of the first image information of inference network 2 is the 2nd frame of the first image information, and the first frame of the first image information of inference network 3 is the 3rd frame of the first image information. The 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames of image features are used as the 10 frames of the first image information of inference network 1, the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames of image features are used as the 10 frames of the first image information of inference network 2, and the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frames of image features are used as the 10 frames of the first image information of inference network 3. It can be understood that at this time, the number of overlapping first image information of any two inference networks is 0.

[0080] S305. Distribute the respective first image information to N target detection networks.

[0081] For the relevant content of step S305, reference can be made to the above embodiments and will not be elaborated here.

[0082] The target detection method provided by the embodiments of the present disclosure, in response to K < M, extracts K frames of the first image information from the first image information set at an interval of the target number of frames based on the constraint condition as the K frames of the first image information of the i-th target detection network. Thus, considering extracting K frames of the first image information at an interval of the target number of frames based on the constraint condition as the K frames of the first image information of a certain target detection network can make the first image information processed by a target detection network evenly distributed in the first image information set.

[0083] Figure 4 is a flowchart of a target detection method shown according to another exemplary embodiment. As Figure 4 shown, the target detection method of the embodiments of the present disclosure includes the following steps.

[0084] S401, obtaining a first image information set, wherein the first image information set includes M frames of first image information, where M is a positive integer.

[0085] S402, determining first image information of each of N target detection networks from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps of the first image information of any two target detection networks is minimal, and N is a positive integer.

[0086] The relevant contents of steps S401-S402 can be found in the above embodiment and will not be repeated here.

[0087] S403, in response to the frame rates of the Q target detection networks being consistent, and the number of overlaps of the first image information of the Q target detection networks determined based on the constraint condition reaching a set condition, where Q is a positive integer greater than 1.

[0088] It should be noted that there are no excessive restrictions on the setting conditions.

[0089] In one embodiment, the method further includes determining that the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaches the set condition in response to the first image information of each of the Q target detection networks determined based on the constraint condition being consistent, and / or the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaching a set threshold. Thus, when the first image information of each of the target detection networks with the same frame rate is consistent, and / or the number of overlaps of the first image information of each of the target detection networks with the same frame rate is large, it can be determined that the set condition is met.

[0090] It should be noted that there are no excessive restrictions on the set threshold, for example, it can be 5.

[0091] S404, merging parameters of the Q target detection networks to obtain a merged network.

[0092] It should be noted that merging the parameters of the Q target detection networks to obtain a merged network means merging the parameters of the Q target detection networks into the parameters of a merged network, that is, converting the Q target detection networks into a merged network. Parameter merging of the Q target detection networks can be achieved by methods such as reparameterization and model combination, which are not limited here. For example, the frame rate of the merged network is consistent with the frame rate of the Q target detection networks.

[0093] S405: Replace the Q object detection networks with a merged network, and determine the first image information of the merged network from the first image information set based on the constraint condition.

[0094] S406: Distribute the first image information of the merged network to the merged network.

[0095] It should be noted that, for determining the relevant content of the first image information of the merged network based on the constraint condition from the first image information set, reference may be made to the relevant content of step S102 in the above embodiment, which will not be repeated here.

[0096] For example, taking N=5 and M=30 as an example, the frame rates of target detection networks 1 to 3 are each 10FPS, and the frame rates of target detection networks 4 and 5 are each 8FPS. The first image information of target detection networks 1 to 5 can be determined from the first image information set based on the constraint conditions.

[0097] In response to the consistent frame rates of the target detection networks 1 to 3, and the number of overlaps of the first image information of the target detection networks 1 to 3 determined based on the constraint condition reaching the set condition, the target detection networks 1 to 3 are parameter-merged to obtain the merged network 6, and the target detection networks 1 to 3 are replaced by the merged network 6. For example, the frame rate of the merged network 6 is 10 FPS.

[0098] In response to the frame rates of the target detection networks 4 and 5 being consistent and the number of overlaps of the first image information of the target detection networks 4 and 5 determined based on the constraint condition not reaching the set condition, the target detection networks 4 and 5 are maintained unchanged.

[0099] The first image information of the merging network 6 is determined based on the constraint conditions, and the respective first image information is distributed to the target detection networks 4, 5, and the merging network 6.

[0100] For example, if the frame rate of the merged network 6 is 10 FPS, the first image information of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames can be used as the first image information of the merged network 6, the first image information of the 5th, 8th, 11th, 14th, 17th, 20th, 23rd, and 26th frames can be used as the first image information of the target detection network 4, and the first image information of the 6th, 9th, 12th, 15th, 18th, 21st, 24th, and 27th frames can be used as the first image information of the target detection network 5. It can be understood that at this time, the number of overlaps of the first image information of the merged network 6, the target detection network 4, and the target detection network 5 is 0.

[0101] The target detection method provided by the embodiment of the present disclosure responds to the consistent frame rates of the Q target detection networks and the number of overlaps of the first image information of the Q target detection networks determined based on the constraint conditions reaches the set conditions, where Q is a positive integer greater than 1, and the parameters of the Q target detection networks are merged to obtain a merged network, the Q target detection networks are replaced with the merged network, and the first image information of the merged network is determined based on the constraint conditions from the first image information set, and the first image information of the merged network is distributed to the merged network. Thus, when the number of overlaps reaches the set conditions, the parameters of multiple target detection networks with consistent frame rates can be merged to obtain a merged network, and the multiple target detection networks with consistent frame rates can be replaced with the merged network, and the constraint conditions can be reconsidered to distribute the first image information to the merged network, so that the distributed first image information meets the set conditions.

[0102] Based on any of the above embodiments, the target detection model includes inference networks 1 to N and post-processing networks 1 to N, wherein the inference networks correspond one to one to the post-processing networks.

[0103] In some examples, the object detection network includes at least one of an inference network and a post-processing network.

[0104] In some examples, the i-th inference network includes the i-th object detection network. Figure 5 As shown, the i-th target detection network is a subnetwork of the i-th reasoning network (not shown in the figure). Alternatively, the i-th target detection network is the i-th reasoning network.

[0105] In some examples, the target detection model also includes a shared network, which is a network shared by N target detection networks. The shared network can be the previous network of each target detection network or the next network of each target detection network, and there is no excessive limitation here. There is no excessive limitation on the shared network, for example, it can include a backbone network (Backbone), a feature extraction network, etc.

[0106] For example, Figure 5 As shown, the i-th inference network (not shown in the figure) includes the i-th target detection network, shared network A, and shared network B. It can be understood that shared network A is the previous network of each target detection network, and shared network B is the next network of each target detection network. At this time, the frame rate of any target detection network is less than or equal to the frame rate of shared network A. For example, the frame rate of shared network A is 30FPS, and the frame rate of each target detection network is less than or equal to 30FPS.

[0107] Figure 6 is a flow chart of a target detection method according to another exemplary embodiment. Figure 6As shown, the target detection method of the embodiment of the present disclosure includes the following steps.

[0108] S601, obtaining a second image information set, wherein the second image information set includes P frames of second image information, where P is a positive integer.

[0109] It should be noted that there are no excessive restrictions on the second image information, for example, it may include the original image, the result of the original image processing, etc. The result of the original image processing may include the image after the original image processing, the image features of the original image, etc.

[0110] S602, inputting the second image information set into a shared network, and outputting the first image information set from the shared network, wherein the shared network is a network shared among N target detection networks.

[0111] Exemplarily, taking the second image information as the original image, the second image information set is the original image set, the original image set can be input into the shared network, and the feature extraction of the j-th frame original image is performed through the shared network to obtain the j-th frame image feature as the j-th frame first image information, where j is a positive integer not greater than M.

[0112] For example, Figure 5 As shown, the original image set can be input into the sharing network A, and the sharing network A outputs a first image information set.

[0113] S603: Determine, from the first image information set, first image information of each of N target detection networks based on a preset constraint condition, wherein the constraint condition is that the number of overlaps of the first image information of any two target detection networks is minimal, and N is a positive integer.

[0114] S604: Distribute respective first image information to N object detection networks.

[0115] The relevant contents of steps S603-S604 can be found in the above embodiment and will not be described again here.

[0116] S605: Use the K frames of initial detection results generated by the i-th inference network as the initial detection results of the i-th post-processing network, where K is a positive integer.

[0117] S606: Distribute the respective initial detection results to N post-processing networks.

[0118] In this embodiment, the i-th reasoning network includes the i-th target detection network.

[0119] It is understandable that the initial detection result of the post-processing network refers to the initial detection result to be processed by the post-processing network, that is, the initial detection result distributed to the post-processing network. Different post-processing networks may have different initial detection results, and different target detection networks may have different amounts of first image information.

[0120] For example, Figure 5 As shown, taking the distribution of post-processing networks 1 to 3 as an example, the initial detection result of the K1 frame generated by the inference network 1 can be used as the initial detection result of the post-processing network 1, the initial detection result of the K2 frame generated by the inference network 2 can be used as the initial detection result of the post-processing network 2, and the initial detection result of the K3 frame generated by the inference network 3 can be used as the initial detection result of the post-processing network 3.

[0121] For example, taking M=30 as an example, the frame rates of target detection networks 1 to 3 are all 10FPS. The initial detection results of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th, and 28th frames generated by the inference network 1 can be used as the initial detection results of the post-processing network 1, the initial detection results of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th, and 29th frames generated by the inference network 2 can be used as the initial detection results of the post-processing network 2, and the initial detection results of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th, and 30th frames generated by the inference network 3 can be used as the initial detection results of the post-processing network 3.

[0122] The initial detection results of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th and 28th frames are distributed to the post-processing network 1, and the initial detection results of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th and 28th frames are post-processed by the post-processing network 1 to obtain the final detection results of the 1st, 4th, 7th, 10th, 13th, 16th, 19th, 22nd, 25th and 28th frames.

[0123] The initial detection results of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th and 29th frames are distributed to the post-processing network 2, and the initial detection results of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th and 29th frames are post-processed by the post-processing network 2 to obtain the final detection results of the 2nd, 5th, 8th, 11th, 14th, 17th, 20th, 23rd, 26th and 29th frames.

[0124] The initial detection results of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th and 30th frames are distributed to the post-processing network 3, and the initial detection results of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th and 30th frames are post-processed by the post-processing network 3 to obtain the final detection results of the 3rd, 6th, 9th, 12th, 15th, 18th, 21st, 24th, 27th and 30th frames.

[0125] The object detection method provided by an embodiment of the present disclosure obtains a second image information set, where the second image information set includes second image information of P frames, and inputs the second image information set into a shared network, and the shared network outputs a first image information set. The shared network is a network shared among N object detection networks, making a single object detection network more lightweight. In addition, taking the K-frame initial detection results generated by the i-th inference network as the initial detection results of the i-th post-processing network, and distributing the respective initial detection results to the N post-processing networks can achieve staggered distribution of the initial detection results, which helps to improve the object detection efficiency.

[0126] Figure 7 It is a block diagram of an object detection device shown according to an exemplary embodiment. Refer to Figure 7 The object detection device 100 according to an embodiment of the present disclosure includes: an acquisition module 110, a determination module 120, and a distribution module 130.

[0127] The acquisition module 110 is configured to acquire a first image information set, where the first image information set includes M frames of first image information, and M is a positive integer;

[0128] The determination module 120 is configured to determine, from the first image information set, the first image information of each of the N object detection networks based on a preset constraint condition, where the constraint condition is that the overlapping quantity of the first image information of any two object detection networks is minimized, and N is a positive integer;

[0129] The distribution module 130 is configured to distribute the respective first image information to the N object detection networks.

[0130] In an embodiment of the present disclosure, the determination module 120 is further configured to: acquire the frame rates of the N object detection networks respectively; determine the number of frames K of the first image information of the i-th object detection network based on the frame rate of the i-th object detection network, where i is a positive integer not greater than N, and K is a positive integer; and determine the K-frame first image information of the i-th object detection network from the first image information set based on the constraint condition.

[0131] In an embodiment of the present disclosure, the determination module 120 is further configured to: in response to K < M, extract K-frame first image information from the first image information set at an interval of target frames based on the constraint condition as the K-frame first image information of the i-th object detection network.

[0132] In one embodiment of the present disclosure, the determination module 120 is further configured to: in response to M being an integer multiple of K, use (M / K-1) as the target number of frames; in response to M not being an integer multiple of K, use the maximum positive integer not greater than (M / K-1) as the target number of frames.

[0133] In one embodiment of the present disclosure, the determination module 120 is further configured to: in response to the consistent frame rates of the Q target detection networks, extract Q consecutive frames of first image information from the first image information set as the first frame of first image information of the Q target detection networks, where Q is a positive integer greater than 1.

[0134] In one embodiment of the present disclosure, the determination module 120 is further configured to: consecutively number the M frames of first image information in the first image information set according to numbers 1 to M; in response to M being an integer multiple of K, perform a modulo operation on the number of the first image information and M / K; and use the K frames of first image information with a remainder of zero as the K frames of first image information of the i-th target detection network.

[0135] In one embodiment of the present disclosure, the distribution module 130 is further configured to: in response to the frame rates of the Q target detection networks being consistent and the number of overlaps of the first image information of the Q target detection networks determined based on the constraint condition reaching a set condition, where Q is a positive integer greater than 1; perform parameter merging on the Q target detection networks to obtain a merged network; replace the Q target detection networks with the merged network, and determine the first image information of the merged network from the first image information set based on the constraint condition; and distribute the first image information of the merged network to the merged network.

[0136] In one embodiment of the present disclosure, the distribution module 130 is further configured to: in response to the first image information of each of the Q target detection networks determined based on the constraint condition being consistent, and / or the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaching a set threshold, determine that the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaches the set condition.

[0137] In one embodiment of the present disclosure, the acquisition module 110 is further configured to: acquire a second image information set, wherein the second image information set includes P frames of second image information, and P is a positive integer; input the second image information set into a shared network, and the shared network outputs the first image information set, wherein the shared network is a network shared among N target detection networks.

[0138] In one embodiment of the present disclosure, the i-th inference network includes the i-th target detection network, i is a positive integer not greater than N, and the inference network corresponds one-to-one to the post-processing network. The distribution module 130 is also configured to: use the K frames of initial detection results generated by the i-th inference network as the initial detection results of the i-th post-processing network, where K is a positive integer; and distribute the respective initial detection results to the N post-processing networks.

[0139] In one embodiment of the present disclosure, the target detection processes of the multiple sub-networks within the target detection network are serial, and the frame rates of the multiple sub-networks within the target detection network are consistent.

[0140] In one embodiment of the present disclosure, the object detection processes of N object detection networks are performed in parallel.

[0141] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0142] The target detection device provided by the embodiment of the present disclosure obtains a first image information set, wherein the first image information set includes M frames of first image information, M is a positive integer, and the first image information of each of N target detection networks is determined from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps between the first image information of any two target detection networks is minimized, N is a positive integer, and the first image information of each of the N target detection networks is distributed. Therefore, the first image information set can be distributed taking into account the constraints, so that the distributed first image information satisfies the constraint that the number of overlaps of the first image information of any two target detection networks is minimized, that is, the first image information processed by any two target detection networks try not to overlap, and the staggered distribution of the first image information set can be achieved, which helps to improve the target detection efficiency. In addition, multiple target detection networks can be used to process the first image information set, which helps to improve the resource utilization of the target detection network, and can achieve multi-classification target detection, which is suitable for traffic signal detection scenarios. In addition, each target detection network only needs to process the distributed first image information, and there is no need to process the entire first image information set, which helps to reduce the amount of data that a single target detection network needs to process, helps to improve the target detection efficiency, and a single target detection network is more lightweight, which is suitable for application scenarios where multiple lightweight target detection networks are deployed at the same time.

[0143] Figure 82 is a block diagram of a vehicle according to an exemplary embodiment. For example, vehicle 200 may be a hybrid vehicle, a non-hybrid vehicle, an electric vehicle, a fuel cell vehicle, or other types of vehicles. Vehicle 200 may be an autonomous vehicle, a semi-autonomous vehicle, or a non-autonomous vehicle.

[0144] Reference Figure 8 , the vehicle 200 may include various subsystems, such as an infotainment system 210, a perception system 220, a decision control system 230, a drive system 240, and a computing platform 250. The vehicle 200 may also include more or fewer subsystems, and each subsystem may include multiple components. In addition, each subsystem and each component of the vehicle 200 may be interconnected by wire or wireless means.

[0145] In some embodiments, the infotainment system 210 may include a communication system, an entertainment system, and a navigation system, among others.

[0146] The perception system 220 may include several sensors for sensing information about the environment around the vehicle 200. For example, the perception system 220 may include a global positioning system (the global positioning system may be a GPS system, or a Beidou system or other positioning systems), an inertial measurement unit (IMU), a laser radar, a millimeter wave radar, an ultrasonic radar, and a camera.

[0147] The decision control system 230 may include a computing system, a vehicle controller, a steering system, a throttle, and a braking system.

[0148] The drive system 240 may include components that provide power movement for the vehicle 200. In one embodiment, the drive system 240 may include an engine, an energy source, a transmission system, and wheels. The engine may be one or a combination of an internal combustion engine, an electric motor, and an air compression engine. The engine can convert energy provided by the energy source into mechanical energy.

[0149] Some or all functions of the vehicle 200 are controlled by a computing platform 250. The computing platform 250 may include at least one processor 251 and a memory 252, and the processor 251 may execute instructions 253 stored in the memory 252.

[0150] The processor 251 may be any conventional processor, such as a commercially available CPU. The processor may also include a graphics processor (Graphic Process Unit, GPU), a field programmable gate array (Field Programmable Gate Array, FPGA), a system on chip (System on Chip, SOC), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or a combination thereof.

[0151] The memory 252 may be implemented by any type of volatile or nonvolatile memory device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0152] In addition to instructions 253 , memory 252 may also store data, such as road maps, route information, and data such as the location, direction, and speed of the vehicle. The data stored in memory 252 may be used by computing platform 250 .

[0153] In the embodiment of the present disclosure, the processor 251 may execute the instruction 253 to implement all or part of the steps of the target detection method provided by the present disclosure.

[0154] The vehicle of the embodiment of the present disclosure obtains a first image information set, wherein the first image information set includes M frames of first image information, M is a positive integer, and the first image information of each of N target detection networks is determined from the first image information set based on a preset constraint condition, wherein the constraint condition is that the number of overlaps between the first image information of any two target detection networks is minimal, N is a positive integer, and the first image information of each of the N target detection networks is distributed to each of the N target detection networks. Therefore, the first image information set can be distributed taking into account the constraints, so that the distributed first image information satisfies the constraint that the number of overlaps of the first image information of any two target detection networks is minimized, that is, the first image information processed by any two target detection networks try not to overlap, and the staggered distribution of the first image information set can be achieved, which helps to improve the target detection efficiency. In addition, multiple target detection networks can be used to process the first image information set, which helps to improve the resource utilization of the target detection network, and can achieve multi-classification target detection, which is suitable for traffic signal detection scenarios. In addition, each target detection network only needs to process the distributed first image information, and there is no need to process the entire first image information set, which helps to reduce the amount of data that a single target detection network needs to process, helps to improve the target detection efficiency, and a single target detection network is more lightweight, which is suitable for application scenarios where multiple lightweight target detection networks are deployed at the same time.

[0155] In order to implement the above embodiments, the present disclosure further proposes a server, comprising a processor; a memory for storing processor executable instructions; wherein the processor is configured to implement the steps of the target detection method provided by the present disclosure.

[0156] In order to implement the above embodiments, the present disclosure further proposes a computer-readable storage medium on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the target detection method provided by the present disclosure are implemented.

[0157] Alternatively, the computer readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, or the like.

[0158] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0159] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A target detection method, characterized in that: Adapted to multi-class object detection, the method includes: Obtain a first image information set, where the first image information set includes M frames of first image information, and M is a positive integer; Obtain the frame rate of each of the N object detection networks. Based on the frame rate of the i-th object detection network, determine the number of frames K of the first image information of the i-th object detection network. From the first image information set, determine K frames of first image information of the i-th object detection network based on a preset constraint condition, where i is a positive integer not greater than N, K is a positive integer, the constraint condition is that the overlapping quantity of the first image information of any two object detection networks is minimized, N is a positive integer, the object detection processes of multiple sub-networks within each object detection network are in series, and the frame rates of the respective sub-networks are the same; Distribute the respective first image information to the N object detection networks.

2. The method according to claim 1, characterized in that The determining, from the first image information set, K frames of first image information of the i-th object detection network based on the constraint condition includes: In response to K < M, from the first image information set, extract K frames of first image information at intervals of the target number of frames based on the constraint condition as K frames of first image information of the i-th object detection network.

3. The method according to claim 2, characterized in that The method further includes: In response to M being an integer multiple of K, use (M / K - 1) as the target number of frames; In response to M not being an integer multiple of K, use the largest positive integer not greater than (M / K - 1) as the target number of frames.

4. The method according to claim 2, characterized in that: The method further includes: In response to the frame rates of Q object detection networks being the same, extract Q consecutive frames of first image information from the first image information set as the first frame of first image information of each of the Q object detection networks, where Q is a positive integer greater than 1.

5. The method according to claim 1, characterized in that The determining, from the first image information set, K frames of first image information of the i-th object detection network based on the constraint condition includes: Number the M frames of first image information in the first image information set continuously according to numbers 1 to M; In response to M being an integer multiple of K, perform a modulo operation on the numbers of the first image information with M / K; Use the K frames of first image information with a remainder of zero as K frames of first image information of the i-th object detection network.

6. The method according to claim 1, characterized in that The method further includes: In response to the frame rates of Q object detection networks being the same, and the overlapping quantity of the first image information of each of the Q object detection networks determined based on the constraint condition reaching a set condition, where Q is a positive integer greater than 1; Perform parameter merging on the Q object detection networks to obtain a merged network; Replace the Q object detection networks with the merged network, and determine the first image information of the merged network from the first image information set based on the constraint condition; Distribute the first image information of the merged network to the merged network.

7. The method according to claim 6, characterized in that The method further includes: In response to the first image information of each of the Q target detection networks determined based on the constraint condition being consistent, and / or the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaching a set threshold, it is determined that the number of overlaps of the first image information of each of the Q target detection networks determined based on the constraint condition reaches the set condition.

8. The method according to any one of claims 1 to 7, characterized in that The obtaining of the first image information set comprises: Acquire a second image information set, wherein the second image information set includes P frames of second image information, where P is a positive integer; The second image information set is input into a shared network, and the shared network outputs the first image information set, wherein the shared network is a network shared among N target detection networks.

9. The method according to any one of claims 1 to 7, characterized in that: The i-th inference network includes the i-th target detection network, i is a positive integer not greater than N, the inference network corresponds to the post-processing network one by one, and the method further includes: Using the K frames of initial detection results generated by the i-th inference network as the initial detection results of the i-th post-processing network, where K is a positive integer; Distribute their initial detection results to N post-processing networks.

10. The method according to any one of claims 1 to 7, characterized in that The object detection process of N object detection networks is parallel.

11. A target detection device, characterized in that: Adapted to multi-classification target detection, the device comprises: An acquisition module is configured to acquire a first image information set, wherein the first image information set includes M frames of first image information, where M is a positive integer; A determination module is configured to obtain the frame rates of the N target detection networks, determine the number of frames K of the first image information of the i-th target detection network based on the frame rate of the i-th target detection network, and determine the K frames of first image information of the i-th target detection network from the first image information set based on a preset constraint condition, wherein i is a positive integer not greater than N, K is a positive integer, the constraint condition is that the number of overlaps of the first image information of any two target detection networks is the smallest, N is a positive integer, the target detection processes of the multiple sub-networks in each of the target detection networks are serial, and the frame rates of the sub-networks are consistent; The distribution module is configured to distribute respective first image information to N target detection networks.

12. A vehicle, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to: Implement the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 10 are implemented.

Images