Edge end video analysis method, device, equipment and medium
By optimizing the allocation of CPU and GPU computing time in edge devices through the Bandit algorithm, and selecting appropriate video resolution and neural network models, the problem of low resource utilization in edge video analysis systems is solved, and efficient video analysis and processing is achieved.
Patent Information
- Application Number
- CN202411293239.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing edge video analytics systems suffer from low computing resource utilization under limited resource conditions, resulting in high latency and error rates, as well as significant waste of hardware resources.
The Bandit algorithm is used to divide the computing time of CPU and GPU in edge devices, select appropriate video resolution and neural network model, optimize the utilization of computing resources, and achieve efficient allocation of resources by having the CPU perform instruction-intensive computing and the GPU perform data-intensive computing, combined with FPGA switching.
It improves the utilization of computing resources in the video analytics system, reduces latency and error rates, avoids waste of hardware resources, and enables rapid processing of video analytics tasks.
Smart Images

Figure CN119027894A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of video analysis, and relates to an edge end video analysis method, device, equipment and medium. Background Art
[0002] The video analysis system at the edge supports multiple applications such as video surveillance, vehicle counting, traffic control, and autonomous driving. In order to cope with limited edge resources, the existing system has an execution pipeline to preprocess video frames to filter out unnecessary frames or extract only the region of interest (ROI) in a frame. When sending the preprocessed frames for model inference, the existing system selects a suitable neural network (NN) model to best balance model inference accuracy and latency. In this execution pipeline, video preprocessing involves a large number of operations such as searching, sorting, and matching, and its computational workload belongs to instruction-intensive computing (IIC), which is executed in the CPU of the edge device. The computational load of model inference is data-intensive computing (DIC), which involves simple operations but large amounts of data, and the DIC load is executed in the GPU of the edge device. Real-time video analysis systems need to achieve high video analysis accuracy while meeting latency requirements. Summary of the invention
[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide an edge video analysis method, device, equipment and medium to improve the computing resource utilization of edge video analysis equipment.
[0004] To achieve the above object, the present invention is implemented by adopting the following technical solutions:
[0005] In a first aspect, the present invention provides an edge video analysis method, comprising:
[0006] Get the video to be analyzed and the corresponding video analysis task;
[0007] According to the video analysis task, a video resolution set and a neural network model set to be selected are obtained;
[0008] According to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the computing time of the CPU and the GPU in the edge device is divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set;
[0009] Based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model, the analysis result of the video to be analyzed is obtained.
[0010] Furthermore, the neural network models in the neural network model set are all pre-trained based on the video analysis task.
[0011] Furthermore, the CPU in the edge device is used to pre-process the video to be analyzed, and the CPU performs instruction-intensive calculations; the GPU in the edge device is used to perform model inference on the video input into the neural network model to obtain the analysis results of the video, and the GPU performs data-intensive calculations.
[0012] Furthermore, the CPU and GPU in the edge device are switched via FPGA.
[0013] Further, according to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the CPU and GPU computing time in the edge device are divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set, including:
[0014] For each video frame of the video to be analyzed, under the premise that the set CPU and GPU computing time meets the delay constraint and FPGA time constraint, different CPU and GPU computing time, resolution and neural network model are combined to form multiple actions of the Bandit algorithm, and the parsing accuracy of the video frame under different actions is calculated to obtain the resolution, neural network model and CPU and GPU computing time when the parsing accuracy is maximum;
[0015] The resolution of the video frame is selected from the video resolution set, and the neural network model of the video frame is selected from the neural network model set.
[0016] Furthermore, the delay constraint of the i-th video frame is expressed as:
[0017] ,
[0018] in, Indicates that the i-th video frame has a resolution of The CPU calculation time under , is the video resolution set; Indicates that the i-th video frame has a resolution of , Neural Network Model The GPU computing time under , is the neural network model set; D is the delay requirement;
[0019] The FPGA time constraint for the i-th video frame is expressed as:
[0020] ,
[0021] in, represents the number of FPGAs allocated to the CPU image to pre-process the i-th video frame; represents the number of FPGAs allocated to the GPU image to perform model inference on the i-th video frame; Indicates the frame rate of the video to be analyzed.
[0022] Furthermore, the Bandit algorithm includes:
[0023] At every moment , the Bandit algorithm observes different contexts , select action , give corresponding rewards ;
[0024] in, Moment of action It has three characteristics: CPU and GPU time division, resolution and neural network models ; The context of the moment Includes four dimensions: CPU resource utilization , GPU resource utilization , Delayed demand and the frame rate of the video to be analyzed ;
[0025] right The expected reward and its feature vector The linear relationship between them is modeled as:
[0026] ,
[0027] in, is a weight vector, representing the accuracy model parameters of online learning, and , represents the dimension of the feature vector;
[0028] Use a LMMSE estimator to estimate ,set up and Respectively Input samples of round feature matrices and corresponding reward vectors, is the number of samples;
[0029] Applying the Bayesian Gauss-Markov theorem, we get Estimated coefficients of round :
[0030] ,
[0031] in, express dimensional identity matrix;
[0032] Estimation Error The covariance of is expressed as:
[0033] ;
[0034] For any hyperparameter , ,but:
[0035] ,
[0036] in, represents the hyperparameter that balances exploration and exploitation, The larger the size, the more emphasis is placed on exploration; accordingly, Round Selection The highest upper confidence bound of the estimated reward is calculated as:
[0037] ;
[0038] The action of momentary choice for ,in, Represents a collection of actions.
[0039] In a second aspect, the present invention further provides an edge video analysis device, the device comprising:
[0040] The module for acquiring the video to be analyzed and the analysis task is used to acquire the video to be analyzed and the corresponding video analysis task;
[0041] A resolution and neural network model set acquisition module, used to acquire a video resolution set and a neural network model set to be selected according to a video analysis task;
[0042] A computing resource division module, used for dividing the computing time of the CPU and GPU in the edge device according to the video to be analyzed, the video resolution set and the neural network model set based on the Bandit algorithm, and selecting a video resolution from the video resolution set and a neural network model from the neural network model set;
[0043] The video analysis module is used to obtain the analysis result of the video to be analyzed based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model.
[0044] In a third aspect, the present invention further provides an edge video analysis device, comprising:
[0045] Memory for storing computer programs;
[0046] A processor is used to execute the computer program to implement the steps of the above-mentioned edge video analysis method.
[0047] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the above-mentioned edge video analysis method.
[0048] Compared with the prior art, the present invention has the following beneficial effects:
[0049] The edge video analysis method provided by the present invention divides the CPU and GPU computing time in the edge device through the Bandit algorithm, selects the corresponding video resolution from the video resolution set, and selects the corresponding neural network model from the neural network model set to achieve optimal utilization of computing resources, so that the video analysis system can quickly process the video analysis task on the edge device, thereby minimizing the delay in task processing; by effectively adjusting resource allocation, the error rate caused by resource shortage or overload can be reduced, and at the same time, the hardware can be more fully utilized, and some computing resources will no longer be idle, thereby reducing the waste of hardware resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 A schematic diagram of a flow chart of an edge video analysis method provided by an embodiment of the present invention;
[0051] Figure 2 Schematic diagram of the workflow of the Bandit algorithm according to an embodiment of the present invention;
[0052] Figure 3 A schematic diagram of a workflow for counting vehicles and pedestrians in an embodiment of the present invention;
[0053] Figure 4 A schematic diagram of the structure of an edge video analysis device provided by an embodiment of the present invention;
[0054] Figure 5 An internal structure diagram of an edge video analysis device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0055] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. The embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.
[0056] The term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.
[0057] Embodiment 1:
[0058] like Figures 1 to 3 As shown, an embodiment of the present invention provides an edge video analysis method. Figure 1 The flowchart is a flow chart of the edge video analysis method. This flowchart only shows the logical sequence of the method described in this embodiment. Under the premise of no conflict, in other possible embodiments of the present invention, different Figure 1 The steps shown or described are accomplished in the order shown.
[0059] The edge video analysis method provided in this embodiment can be applied to a terminal and can be performed by an edge video analysis device. The device can be implemented by software and / or hardware, and the device can be integrated in the terminal.
[0060] See also Figure 1 The method of the embodiment of the present invention specifically includes the following steps 1 to 4. Among them:
[0061] Step 1: Get the video to be analyzed and the corresponding video analysis task.
[0062] Step 2: According to the video analysis task, obtain the image resolution set and neural network model set to be selected.
[0063] Among them, the neural network models in the neural network model set have all been pre-trained according to the video analysis task.
[0064] Step 3: According to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the computing time of the CPU and GPU in the edge device is divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set.
[0065] Among them, in an embodiment of the present invention, the CPU in the edge device is used for video preprocessing, which includes filtering out unnecessary video frames or extracting only the region of interest (ROI) in the video frame. Video preprocessing involves a large number of operations such as searching, sorting, and matching, and its computational workload belongs to instruction-intensive computing (IIC). The GPU in the edge device is used to perform model reasoning using a neural network model. The operations involved in model reasoning are simple but the amount of data is large. Therefore, the GPU performs data-intensive computing (DIC). The real-time video analysis system needs to achieve high video analysis accuracy while meeting the latency requirements.
[0066] Step 3 specifically includes: for each video frame of the video to be analyzed, under the premise that the set CPU and GPU computing time meets the delay constraint and FPGA time constraint, different CPU and GPU computing times, resolutions and neural network models are combined to form multiple actions of the Bandit algorithm, and the parsing accuracy of the video frames under different actions is calculated to obtain the resolution, neural network model and CPU and GPU computing time when the parsing accuracy is maximum.
[0067] The resolution of the video frame is selected from the video resolution set, and the neural network model of the video frame is selected from the neural network model set.
[0068] The delay constraint of the i-th video frame is expressed as:
[0069] ,
[0070] in, Indicates that the i-th video frame has a resolution of The CPU calculation time under , is the video resolution set; Indicates that the i-th video frame has a resolution of , Neural Network Model The GPU computing time under , is the neural network model set; D is the delay requirement;
[0071] The FPGA time constraint for the i-th video frame is expressed as:
[0072] ,
[0073] in, represents the number of FPGAs allocated to the CPU image to pre-process the i-th video frame; represents the number of FPGAs allocated to the GPU image to perform model inference on the i-th video frame; Indicates the frame rate of the video to be analyzed.
[0074] like Figure 2 As shown, the Bandit algorithm works well in applications where the context changes smoothly by repeatedly observing a context, selecting an action, and observing the reward of the selected action, which is a good match for video analysis applications.
[0075] In the embodiment of the present invention, the Bandit algorithm is implemented by the LinUCB algorithm, and the specific process includes:
[0076] At every moment , by observing different contexts , select action , give corresponding rewards ; Indicates action A collection of . represents a feature vector that captures all available side information, including the current context and actions selected features.
[0077] in, Moment of action It has three characteristics: CPU and GPU time division, resolution and neural network models ; The context of the moment Includes four dimensions: CPU resource utilization , GPU resource utilization , Delayed demand and the frame rate of the video to be analyzed .
[0078] Will reward It is defined as the accuracy of the analysis observed after the action. Since we cannot directly obtain the analysis results and the truth once the analysis results are online, the uncertainty of the analysis results is used instead, which shows that the lower the uncertainty, the higher the probability of correct analysis.
[0079] right The expected reward and its feature vector The linear relationship between them is modeled as:
[0080] ,
[0081] in, is a weight vector, representing the accuracy model parameters of online learning, and , express The dimension of the eigenvector of ;
[0082] Use a LMMSE estimator to estimate ,set up and Respectively Input samples of round feature matrices and corresponding reward vectors, is the number of samples;
[0083] Applying the Bayesian Gauss-Markov theorem, we get Estimated coefficients of round :
[0084] ,
[0085] in, express dimensional identity matrix;
[0086] Estimation Error The covariance of is expressed as:
[0087] ;
[0088] For any hyperparameter , ,but:
[0089] ,
[0090] in, represents the hyperparameter that balances exploration and exploitation, The larger the size, the more emphasis is placed on exploration; accordingly, Round Selection The highest upper confidence bound of the estimated reward is calculated as:
[0091] ;
[0092] The action of momentary choice for , that is, select the set can make The biggest move, among them, Represents a collection of actions.
[0093] Step 4: Based on the divided computing time of the CPU and GPU, as well as the selected video resolution and neural network model, obtain the analysis result of the video to be analyzed.
[0094] For example, Figure 3 As shown, the video analysis task is to count the number of vehicles and pedestrians in the video, and the Auburn dataset is used as the dataset to be analyzed. The dataset contains 54GB of 24-hour traffic video, and the bounding boxes of vehicles and pedestrians are marked in the video. This embodiment compares Gemini with two state-of-the-art edge video analysis systems and an offline optimal solution as the upper bound of performance in vehicle and pedestrian detection on the DeepLens-Max10 test platform, and transmits the frame to Max10 in response to DeepLens, using FPGA to simulate CPU and GPU resources. Represents video analysis system Assigned and FPGA time is used as CPU and GPU resources. Representation System Optimal fixed CPU and GPU resource allocation.
[0095] The Offline Optimal Scheme (OOS) is calculated using dynamic programming with full workload information. It outputs the optimal computing resource allocation strategy, resolution, and neural network model. The offline optimality is an upper bound on the accuracy of the omniscient strategy when the future IIC and DIC workloads are fully known.
[0096] Three metrics are used to evaluate the method: 1) Accuracy: For the vehicle counting application, the parsing accuracy is calculated as ,in, is the number of vehicles in the analysis result, is the true value. For vehicle and pedestrian detection applications, the resolution accuracy is calculated as ,in, It is the overlapping area between the analysis result positioning box and the correct positioning box. To correctly locate the area of the box; 2) Latency Miss Rate: quantifies the percentage of video analysis tasks that do not meet the latency requirements set by the video analysis application; 3) Hardware Utilization: indicates the effective utilization of computing resources in the video analysis system.
[0097] Embodiment 2:
[0098] Based on the same inventive concept as in Example 1, the embodiment of the present invention also provides an edge video analysis device for implementing the above-mentioned edge video analysis method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiment of the edge video analysis device provided below can refer to the limitations of the edge video analysis method above, and will not be repeated here.
[0099] like Figure 4 As shown, an embodiment of the present invention provides an edge video analysis device, including:
[0100] The module for acquiring the video to be analyzed and the analysis task is used to acquire the video to be analyzed and the corresponding video analysis task;
[0101] A resolution and neural network model set acquisition module, used to acquire a video resolution set and a neural network model set to be selected according to a video analysis task;
[0102] A computing resource division module, used for dividing the computing time of the CPU and GPU in the edge device according to the video to be analyzed, the video resolution set and the neural network model set based on the Bandit algorithm, and selecting a video resolution from the video resolution set and a neural network model from the neural network model set;
[0103] The video analysis module is used to obtain the analysis result of the video to be analyzed based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model.
[0104] Embodiment 3:
[0105] The embodiment of the present invention further provides an edge video analysis device, which includes a computer device. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the edge video analysis method in the aforementioned embodiment is implemented.
[0106] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0107] Embodiment 4:
[0108] The embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the following method are implemented:
[0109] Get the video to be analyzed and the corresponding video analysis task;
[0110] According to the video analysis task, a video resolution set and a neural network model set to be selected are obtained;
[0111] According to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the computing time of the CPU and the GPU in the edge device is divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set;
[0112] Based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model, the analysis result of the video to be analyzed is obtained.
[0113] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products, and therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0114] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0115] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0117] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. An edge video analysis method, characterized in that: include: Get the video to be analyzed and the corresponding video analysis task; According to the video analysis task, a video resolution set and a neural network model set to be selected are obtained; According to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the computing time of the CPU and the GPU in the edge device is divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set; Based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model, the analysis result of the video to be analyzed is obtained.
2. The edge video analysis method according to claim 1, characterized in that: The neural network models in the neural network model set are all pre-trained based on the video analysis task.
3. The edge video analysis method according to claim 1, characterized in that: The CPU in the edge device is used to pre-process the video to be analyzed, and the CPU performs instruction-intensive calculations; the GPU in the edge device is used to perform model inference on the video input into the neural network model to obtain the analysis results of the video, and the GPU performs data-intensive calculations.
4. The edge video analysis method according to claim 3, characterized in that: The CPU and GPU in the edge device are switched via FPGA.
5. The edge video analysis method according to claim 1, characterized in that: According to the video to be analyzed, the video resolution set and the neural network model set, based on the Bandit algorithm, the CPU and GPU computing time in the edge device are divided, and the video resolution is selected from the video resolution set, and the neural network model is selected from the neural network model set, including: For each video frame of the video to be analyzed, under the premise that the set CPU and GPU computing time meets the delay constraint and FPGA time constraint, different CPU and GPU computing time, resolution and neural network model are combined to form multiple actions of the Bandit algorithm, and the parsing accuracy of the video frame under different actions is calculated to obtain the resolution, neural network model and CPU and GPU computing time when the parsing accuracy is maximum; The resolution of the video frame is selected from the video resolution set, and the neural network model of the video frame is selected from the neural network model set.
6. The edge video analysis method according to claim 5, characterized in that: The delay constraint of the i-th video frame is expressed as: , in, Indicates that the i-th video frame has a resolution of The CPU calculation time under , is the video resolution set; Indicates that the i-th video frame has a resolution of , Neural Network Model The GPU computing time under , is the neural network model set; D is the delay requirement; The FPGA time constraint for the i-th video frame is expressed as: , in, represents the number of FPGA times allocated to the CPU image to pre-process the i-th video frame; represents the number of FPGAs allocated to the GPU image to perform model inference on the i-th video frame; Indicates the frame rate of the video to be analyzed.
7. The edge video analysis method according to claim 6, characterized in that: The Bandit algorithm includes: At every moment , the Bandit algorithm observes different contexts , select action , give corresponding rewards ; in, Moment of action It has three characteristics: CPU and GPU time division, resolution and neural network models ; The context of the moment Includes four dimensions: CPU resource utilization , GPU resource utilization , Delayed demand and the frame rate of the video to be analyzed ; right The expected reward and its feature vector The linear relationship between them is modeled as: , in, is a weight vector, representing the accuracy model parameters of online learning, and , represents the dimension of the feature vector; Use a LMMSE estimator to estimate ,set up and Respectively Input samples of round feature matrices and corresponding reward vectors, is the number of samples; Applying the Bayesian Gauss-Markov theorem, we get Estimated coefficients of round : , in, express dimensional identity matrix; Estimation Error The covariance of is expressed as: ; For any hyperparameter , ,but: , in, represents the hyperparameter that balances exploration and exploitation, The larger the size, the more emphasis is placed on exploration; accordingly, Round Selection The highest upper confidence bound of the estimated reward is calculated as: ; The action of momentary choice for ,in, Represents a collection of actions.
8. An edge video analysis device, characterized in that: include: The module for acquiring the video to be analyzed and the analysis task is used to acquire the video to be analyzed and the corresponding video analysis task; A resolution and neural network model set acquisition module, used to acquire a video resolution set and a neural network model set to be selected according to a video analysis task; A computing resource division module, used for dividing the computing time of the CPU and GPU in the edge device according to the video to be analyzed, the video resolution set and the neural network model set based on the Bandit algorithm, and selecting a video resolution from the video resolution set and a neural network model from the neural network model set; The video analysis module is used to obtain the analysis result of the video to be analyzed based on the divided computing time of the CPU and GPU, and the selected video resolution and neural network model.
9. An edge video analysis device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the steps of edge video analysis according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the edge video analysis method described in any one of claims 1 to 7 are implemented.
