Monitoring video edge resource allocation method and system based on cloud edge collaboration

Abstract

The invention discloses a monitoring video edge resource allocation method and system based on cloud edge collaboration. The method comprises the following steps: constructing a cloud edge collaboration framework model consisting of a cloud layer and an edge layer; wherein the edge layer is divided into k monitoring areas, and the interior of each monitoring area is provided with a corresponding edge node; obtaining the monitoring area security level and edge node state information of the edge layer; when a preset constraint condition is satisfied, establishing an optimization problem taking the monitoring video task processing time delay and bandwidth use loss minimization of the edge layer as a target according to the monitoring area security level and the edge node state information; an optimization problem is converted into a Markov decision problem, the Markov decision problem is solved by adopting a DDPG algorithm through a cloud layer, and an optimal task collaborative allocation strategy of an edge layer is generated, so that effective utilization of computing resources of the edge layer in a complex environment of a monitoring area is realized, and collaborative task processing capability among edge nodes is improved.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technologies, and in particular to an edge resource allocation method and system for monitoring videos based on cloud-edge collaboration. Background Art

[0002] Intelligent video surveillance plays an important role in the security system. Taking a production site as an example, when a worker accidentally approaches dangerous high-voltage equipment, the intelligent video surveillance system can respond quickly, issue an alarm or directly start the shutdown mechanism of the equipment, thereby ensuring the safety of personnel. At present, artificial intelligence models based on artificial neural networks are one of the efficient means of processing surveillance videos, but video processing models based on neural networks often require a lot of parameter support and computing power. Traditional intelligent monitoring methods usually transmit surveillance videos to cloud servers for intelligent analysis, and then return the analysis results to the monitoring terminal. This process causes serious delays and occupies network congestion, so that the intelligent processing of surveillance videos loses its real-time response capability and cannot effectively ensure production safety. Intelligent processing of surveillance videos through edge computing can improve the real-time performance of video transmission and reduce the bandwidth occupation caused by video transmission. However, traditional edge computing is limited by the computing power of edge devices and servers and cannot support real-time video processing based on artificial intelligence.

[0003] Cloud-edge collaboration combines the advantages of cloud computing and edge computing. Cloud computing has abundant computing and storage resources, which can support the reasoning and training of artificial intelligence models, but its real-time performance is limited by various environments. Edge computing has fewer computing and storage resources, but deploying lightweight AI models on the edge can improve real-time performance and reduce network bandwidth usage. However, the traditional cloud-edge collaboration framework does not fully consider the effective use of edge layer computing resources, and also ignores the collaborative task processing capabilities between edge nodes, resulting in a waste of computing resources and affecting the real-time response capabilities of intelligent processing. Summary of the invention

[0004] The present invention provides an edge resource allocation method and system for surveillance video based on cloud-edge collaboration, which is used to solve the technical problem that the existing cloud-edge collaboration framework does not fully consider the effective utilization of the computing resources of the edge layer for surveillance video, and also ignores the collaborative task processing capability between edge nodes, thereby causing waste of computing resources and affecting the real-time response capability of intelligent processing.

[0005] A first aspect of the present invention provides an edge resource allocation method for surveillance video based on cloud-edge collaboration, the method comprising:

[0006] Construct a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer; the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node;

[0007] Obtain the security level of the monitoring area and edge node status information of the edge layer;

[0008] Under the preset constraints, an optimization problem is established with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information;

[0009] The optimization problem is converted into a Markov decision problem, and the Markov decision problem is solved by using the DDPG algorithm at the cloud layer to generate the optimal task collaboration allocation strategy for the edge layer.

[0010] More specifically, the edge node status information includes monitoring video processing tasks composed of local monitoring video tasks and edge monitoring video tasks, the percentage of computing power allocated to different tasks by each edge node, and the channel transmission rate; the step of establishing an optimization problem with the goal of minimizing the monitoring video task processing delay and bandwidth usage loss at the edge layer according to the security level of the monitoring area and the edge node status information under the preset constraint conditions includes:

[0011] Based on the local surveillance video task, a local computing model is established according to the percentage of computing power allocated to different tasks by each edge node;

[0012] Based on the edge surveillance video task, a task offloading model is established according to the percentage of computing power allocated to different tasks by each edge node and the channel transmission rate;

[0013] Under the preset constraints, an optimization problem is established based on the security level of the monitored area, the local computing model and the task offloading model, with the goal of minimizing the processing delay and bandwidth loss of the monitoring video tasks at the edge layer.

[0014] More specifically, the step of converting the optimization problem into a Markov decision problem, solving the Markov decision problem by using the DDPG algorithm at the cloud layer, and generating an optimal task collaborative allocation strategy at the edge layer includes:

[0015] The optimization problem is transformed into a Markov decision problem, where the state space is composed of the percentage of computing power allocated to different tasks by each edge node, the average delay of the edge node in processing video frames, the channel transmission rate and the security area level; the percentage of tasks unloaded from each edge node to other edge nodes is used as the action space; and the reward function is set according to the minimization of the delay of monitoring video task processing and bandwidth usage loss at the edge layer;

[0016] Based on the DDPG algorithm, the policy network and value network are deployed in the cloud layer. The policy network learns the deterministic strategy according to the state space and action space to output the action value under a given state. The value network learns according to the reward function, evaluates the value under a given state, and generates the optimal task collaboration allocation strategy for the edge layer.

[0017] More specifically,

[0018] The local computing model establishment process is expressed as:

[0019]

[0020] Where: represents the computational delay of the i-th edge node processing the local surveillance video task, represents the total number of local video frames that the ith edge node needs to process, I represents the amount of byte image data contained in each frame of video, represents the amount of computation required to process each byte of image data, represents the computing power of the ith edge node, Indicates the percentage of computing power used by the i-th edge node when processing local surveillance video tasks.

[0021] More specifically,

[0022] The establishment process of the task offloading model is expressed as:

[0023]

[0024] Where: represents the total delay of each video frame when the jth edge node transmits the video frame to the ith edge node for task calculation in time slot t, represents the average computational delay of the i-th edge node processing the video frame sent by the j-th edge node in time slot t, represents the transmission delay of the jth edge node transmitting each video frame to the ith edge node in time slot t;

[0025] in,

[0026]

[0027] Where: It represents the ratio of the edge monitoring video task volume to be processed from the jth edge node to the ith edge node in the tth time slot to the current total task volume of the ith edge node, represents the total number of local surveillance video frames that the jth edge node needs to process; It represents the percentage of computing power occupied by the i-th edge node when cooperating with the monitoring video task transmitted by the j-th edge node in time slot t;

[0028] in,

[0029]

[0030] Where: Indicates the total number of edge nodes in the edge layer, It represents the information transmission rate between the jth edge node and the ith edge node in time slot t.

[0031] More specifically,

[0032] The monitoring video task processing delay of each edge node in the edge layer is expressed as:

[0033]

[0034] Where: It represents the average delay of edge node i in processing the surveillance video task in time slot t.

[0035] More specifically,

[0036] The optimization problem is expressed as:

[0037]

[0038] Where: represents a binary variable, where It is used to indicate whether the i-th monitoring area is a key area or an idle area in the t-th time slot; Indicates the weight value of the surveillance video processing task in the key area, represents the weight value of the surveillance video processing task in the idle area; Indicates the ratio of video frame transmission bandwidth to the total channel bandwidth; It represents the ratio of the amount of tasks to be processed from the i-th edge node to the j-th edge node in time slot t to the current total amount of tasks of the i-th edge node;

[0039] The preset constraints include:

[0040]

[0041] Where: C1 represents the constraint on the proportion of edge nodes transmitting surveillance video tasks; C2 represents the constraint that the tasks assigned to edge nodes cannot exceed their computing capabilities; C3 represents the constraint on the real-time processing delay of edge nodes for surveillance video tasks; C4 represents that the computing resources used by the edge layer for surveillance video task processing cannot exceed the total computing capacity of the edge layer.

[0042] The second aspect of the present invention further provides an edge resource allocation system for monitoring video based on cloud-edge collaboration, the system comprising:

[0043] The framework model building unit is used to build a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer; wherein the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node;

[0044] An acquisition unit, used to obtain the security level of the monitoring area and the status information of the edge node of the edge layer;

[0045] An optimization problem establishing unit, used to establish an optimization problem with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information under the condition of satisfying the preset constraints;

[0046] The allocation strategy generation unit is used to convert the optimization problem into a Markov decision problem, solve the Markov decision problem by using the DDPG algorithm at the cloud layer, and generate an optimal task collaboration allocation strategy at the edge layer.

[0047] A third aspect of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the edge resource allocation method for monitoring video as described above are implemented.

[0048] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the edge resource allocation method for monitoring video as described above are implemented.

[0049] It can be seen from the above technical solutions that the present invention has the following advantages:

[0050] The present invention provides an edge resource allocation method and system for monitoring videos based on cloud-edge collaboration, wherein the method comprises: constructing a cloud-edge collaboration framework model consisting of a cloud layer and an edge layer; wherein the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node; obtaining the monitoring area security level and edge node status information of the edge layer; under the condition of satisfying preset constraints, establishing an optimization problem with the goal of minimizing the monitoring video task processing delay and bandwidth usage loss of the edge layer according to the monitoring area security level and edge node status information; converting the optimization problem into a Markov decision problem, solving the Markov decision problem by using the DDPG algorithm through the cloud layer, and generating the optimal task collaborative allocation strategy for the edge layer.

[0051] In the present invention, edge nodes of the edge layer are used to realize real-time intelligent reasoning of surveillance videos. In order to effectively utilize the computing resources of the edge layer, a joint model is performed for the complex environment of the monitoring area, combined with the time-varying nature of the monitoring area and the monitoring video task processing delay and bandwidth usage problems, to establish an optimization problem with the goal of minimizing the monitoring video task processing delay and bandwidth usage loss of the edge layer, and convert the optimization problem into a Markov decision problem; the Markov decision problem is solved by the cloud layer to realize non-real-time task processing such as edge resource allocation model training, thereby generating an optimal task collaborative allocation strategy for the edge layer, so that the edge computing resources are adjusted with the location and number of key areas, while maintaining the real-time response processing of idle areas, and realizing the effective utilization of the computing resources of the edge layer, thereby solving the problem that the existing cloud-edge collaborative framework does not fully consider the effective utilization of the computing resources of the monitoring video by the edge layer, and also ignores the collaborative task processing capabilities between edge nodes, thereby causing waste of computing resources and affecting the real-time response capability of intelligent processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0053] Figure 1 A flowchart of the steps of an edge resource allocation method for monitoring video based on cloud-edge collaboration provided by an embodiment of the present invention;

[0054] Figure 2 An architectural diagram of a cloud-edge collaboration framework model provided for an embodiment of the present invention;

[0055] Figure 3 A structural block diagram of an edge resource allocation system for monitoring video based on cloud-edge collaboration provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The embodiments of the present invention provide an edge resource allocation method and system for surveillance videos based on cloud-edge collaboration, which is used to solve the technical problem that the existing cloud-edge collaboration framework does not fully consider the effective utilization of the computing resources of the surveillance video by the edge layer, and also ignores the collaborative task processing capability between edge nodes, thereby causing waste of computing resources and affecting the real-time response capability of intelligent processing.

[0057] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 The first aspect of the present invention provides an edge resource allocation method for monitoring video based on cloud-edge collaboration, the method comprising:

[0059] Step 101, construct a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer.

[0060] See also Figure 2 The cloud-edge collaborative framework model includes a cloud layer and an edge layer. The cloud server in the cloud layer is responsible for processing non-real-time tasks, such as the training of intelligent models such as video inference models and edge resource allocation models. The trained models are transmitted to edge devices through the industrial core network. The edge layer transmits the state and strategy set observed by the interaction with the environment and video training data to the cloud layer through the core network. The edge layer is also mainly responsible for the real-time processing of monitoring videos. The edge layer contains two types of nodes: management nodes and edge nodes. The management node allocates edge resources through the edge resource allocation model. The edge node is responsible for collecting video data through the camera and analyzing it.

[0061] Step 102, obtaining the security level of the monitoring area and the edge node status information of the edge layer.

[0062] Based on the time-varying nature of the complex environment of the monitoring area at the edge layer, the monitoring area is pre-divided into "key" areas and "idle" areas according to the density of personnel gathering. The complexity of different monitoring areas is then used to adjust the cooperation mode between edge nodes, so that edge computing resources are adjusted with the location and number of "key" areas, while maintaining the real-time processing of "idle" areas; in other words, when the computing power of the edge nodes in the "key" areas is not sufficient to meet the real-time requirements, some videos that cannot be processed in time by the edge nodes in the "key" areas can be transmitted to the edge nodes in the "idle" areas via industrial Ethernet, thereby avoiding the waste of computing resources and improving the real-time response processing capabilities of the edge nodes.

[0063] Assume that there are k monitoring areas in the edge layer, and each area is monitored and video processed by an edge node. Therefore, the total number of edge nodes in the edge layer is also k; and each monitoring area will generate corresponding monitoring video tasks that need to be processed by the edge node, that is, each monitoring area will have corresponding local monitoring video tasks and edge monitoring video tasks. For example, the monitoring video analysis task generated by the i-th monitoring area in time slot t can be divided into two parts. The first part is processed by the local edge node (that is, the local monitoring video task), and when the local edge node has insufficient computing power, the remaining part (that is, the edge monitoring video task) will be processed collaboratively by other edge nodes.

[0064] Then the monitoring video task transmission matrix at the edge layer in time slot t It can be expressed as:

[0065]

[0066] Where: It is expressed as the ratio (percentage) of the amount of tasks to be processed transmitted from the i-th edge node to the j-th edge node in time slot t to the current total amount of tasks of the i-th edge node.

[0067] It should be noted that in the present invention, binary variables are used Indicates the different security levels of the monitored area, that is, whether the i-th area is a key area or an idle area in the t time slot, where Indicates that the i-th area is the key area, Indicates that the ith area in time slot t is an idle area. It should be noted that is a time variable, that is, the i-th area will switch between the key area and the idle area in different time slots due to uncontrollable factors, such as the sudden gathering of personnel and changes in engineering tasks.

[0068] It is understandable that the edge node status information includes the monitoring video processing tasks composed of local monitoring video tasks and edge monitoring video tasks, the percentage of computing power allocated to different tasks by each edge node, and channel transmission rate .

[0069] Specifically, we can define the calculation distribution matrix To represent the ratio of the computing power allocated to the video analysis task of each edge node to the total computing power of the node (computing power percentage), as follows:

[0070]

[0071] Where: Indicates the percentage of computing power used by the i-th edge node when processing local surveillance video tasks. Indicates the percentage of computing power used by the i-th edge node to collaboratively process the edge monitoring video task transmitted by the j-th edge node.

[0072] At the same time, the information transmission mode between edge nodes can include wireless transmission and wired transmission, so its transmission rate changes dynamically with different time slots t under different transmission modes. Without loss of generality, the transmission rate matrix can be written as for:

[0073]

[0074] Step 103, under the conditions of satisfying preset constraints, an optimization problem is established with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information.

[0075] Specifically, this step includes the following sub-steps:

[0076] Sub-step 1031, based on the local surveillance video task, establish a local computing model according to the percentage of computing power allocated to different tasks by each edge node.

[0077] The local computing model establishment process is expressed as:

[0078]

[0079] Where: represents the computational delay of the i-th edge node processing the local surveillance video task, represents the total number of local surveillance video frames that the ith edge node needs to process, I represents the amount of byte image data contained in each frame of video, represents the amount of computation required to process each byte of image data, represents the computing power of the ith edge node, Indicates the percentage of computing power used by the i-th edge node when processing local surveillance video tasks.

[0080] Sub-step 1032, based on the edge monitoring video task, establish a task offloading model according to the percentage of computing power allocated to different tasks by each edge node and the channel transmission rate.

[0081] It should be noted that when transmitting video frames to other nodes for intelligent reasoning, not only the network transmission delay and computing delay should be considered, but also the computing tasks of the load node itself. Regardless of whether there are people in the area monitored by the load node, the node needs to calculate and process the local surveillance video and the edge surveillance video unloaded from other nodes.

[0082] Therefore, when processing video frames sent by other nodes, the edge node must first process the video frames that need to be analyzed, and cross-process the local monitoring video frames and other video frames in a certain proportion, so as to ensure the real-time processing of local videos and offloaded videos at the same time.

[0083] use represents the total number of surveillance video frames that the jth edge node needs to process, then the average computational delay of the i-th edge node processing the video frames sent by the j-th edge node can be calculated. for:

[0084]

[0085] Where: It represents the ratio of the amount of edge monitoring video tasks to be processed transmitted from the jth edge node to the ith edge node in time slot t to the current total amount of tasks of the ith edge node.

[0086] From the above, we can see that the transmission delay of each video frame transmitted from the jth edge node to the ith edge node in time slot t is It can be expressed as:

[0087]

[0088] Where: It represents the information transmission rate between the jth edge node and the ith edge node in time slot t.

[0089] Then, when the jth edge node transmits the video frame to the ith edge node for calculation in time slot t, the total delay generated by each video frame is for:

[0090]

[0091] For edge node j, when its own computing power cannot meet the real-time requirements of local video analysis, part of the video in its monitoring area is analyzed and processed locally, and the remaining part needs to be transmitted to other edge nodes for processing.

[0092] Therefore, each edge node needs to process the local surveillance video frames and the surveillance video frames transmitted from other nodes. The average delay of edge node i in processing the video frame in time slot t is:

[0093]

[0094] Then, for the entire edge layer, the total latency state can be expressed as:

[0095]

[0096] Sub-step 1033, under the conditions of satisfying preset constraints, establishes an optimization problem with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss at the edge layer according to the security level of the monitoring area, the local computing model and the task offloading model.

[0097] The present invention cooperates with each edge node to process the surveillance video of the "key" area and the "idle" area in real time. For the "key" area, it is necessary to prioritize the real-time processing of the surveillance video in the area to ensure the safety of personnel. Therefore, the edge layer resources can be prioritized to analyze and process the surveillance video of the area according to the edge node status and network environment. At the same time, for the proctoring video processing in the "idle" area, a certain real-time performance must be maintained so that the area has the ability to respond to emergencies.

[0098] Considering the limited computing power of the edge layer, in order to achieve lower computing latency, processing weights are set for the monitoring video task delays in "key" areas and "idle" areas, and high weights are set for video processing tasks in key areas. , set a lower weight for video processing tasks in the "idle" area .in, Indicates the maximum delay limit for processing a single video task in the i-th area when the i-th area is a key area; It represents the maximum delay limit of processing a single video task in the i-th area when the i-th area is an idle area. In addition, in order to save communication resources, it is also necessary to minimize the transmission amount of video frames.

[0099] Therefore, in time slot t, the resource allocation problem for edge nodes can be described as an optimization problem of jointly minimizing delay and bandwidth usage loss, as shown below:

[0100]

[0101] Where: Indicates the ratio of video frame transmission bandwidth to the total channel bandwidth;

[0102] The preset constraints specifically include:

[0103]

[0104] Where: C1 represents the constraint on the proportion of edge nodes transmitting surveillance video tasks; C2 represents the constraint that the tasks assigned to edge nodes cannot exceed their computing capabilities; C3 represents the constraint on the real-time processing delay of edge nodes for surveillance video tasks; C4 represents that the computing resources used by the edge layer for surveillance video task processing cannot exceed the total computing capacity of the edge layer.

[0105] Step 104, convert the optimization problem into a Markov decision problem, solve the Markov decision problem by using the DDPG algorithm at the cloud layer, and generate the optimal task coordination allocation strategy at the edge layer.

[0106] It is understandable that edge nodes need to determine the task offloading percentage based on their own computing power, channel transmission rate, security area level and other status information, while minimizing task processing delays and communication resource losses, so as to achieve optimal edge resource scheduling and allocation.

[0107] This step specifically includes the following sub-steps:

[0108] Sub-step 1041, converts the optimization problem into a Markov decision problem (MDP problem), wherein the state space is composed of the percentage of computing power allocated to different tasks by each edge node, the average delay of the edge node in processing video frames, the channel transmission rate and the security area level; the percentage of tasks offloaded from each edge node to other edge nodes is used as the action space; and the reward function is set according to the minimization of the delay in processing surveillance video tasks and the bandwidth usage loss at the edge layer.

[0109] It should be noted that the monitoring video task offloading behavior of the edge node will affect the environment state, and the total overhead of the framework model is determined by the current state of the framework model and the joint actions of all edge nodes. In addition, the previous state and previous action of the framework model will transform the environment state into a new state.

[0110] In this case, the optimization problem can be transformed into a multi-agent Markov decision process S is the collection of all states, A is the action space, and P represents the state transition probability. is the current reward, Represents the discount factor.

[0111] 1): State space :For the given cloud-edge collaborative framework model, the state space of each time slot is determined by the percentage of computing power allocated to different tasks by each edge node , the average delay of edge nodes processing video frames , channel transmission rate and security area levels Composition, that is

[0112] 2) Action Space :The action space is the percentage of tasks that each edge node offloads to other nodes, so .

[0113] 3) Reward Function : Reward function In order to solve the above optimization problem, the total delay and bandwidth loss are minimized under certain constraints, so we can get:

[0114]

[0115] If the constraints are not satisfied, then .

[0116] Sub-step 1042, deploys the policy network and the value network in the cloud layer based on the DDPG algorithm, and uses the policy network to learn the deterministic strategy according to the state space and action space to output the action value under a given state; the value network learns according to the reward function, evaluates the value under a given state, and generates the optimal task collaboration allocation strategy for the edge layer.

[0117] In order to solve the above MDP problem, considering the high-dimensional continuous action space of the task offloading optimization problem, the deep deterministic policy gradient algorithm (DDPG algorithm) is adopted to deploy a policy network, a value network, a target policy network, a target value network and an experience revisit cache in the cloud layer using cloud servers.

[0118] In the cloud server, through the Actor-Critic architecture, the policy network learns deterministic policies based on the state-policy set formed by the state space and action space transmitted by the edge layer, that is, outputs the action value under a given state and transmits the action value to the edge layer through the core network; the value network is responsible for learning the value function and evaluating the value of the current state.

[0119] In order to solve the problems of sample correlation and stability, the edge layer transmits the experience data obtained from the interaction with the environment to the cloud server, which stores it in the experience replay cache and then randomly samples it for edge resource allocation model training. In addition, the DDPG algorithm also uses the target policy network and the target value network to improve learning stability. After each iteration, the cache space of the algorithm is updated according to the optimal strategy of the iteration and the cache update mechanism, so as to generate the optimal task collaborative allocation strategy, and then achieve the optimal percentage of tasks unloaded from each edge node to other nodes. Under the condition of minimizing delay and bandwidth loss, the edge computing resources are adjusted with the location and number of key areas, while maintaining the real-time response processing of the idle area, realizing the effective utilization of the computing resources of the edge layer in the complex environment of the monitoring area, and improving the collaborative task processing capabilities between edge nodes.

[0120] See also Figure 3 The second aspect of the present invention further provides an edge resource allocation system for monitoring video based on cloud-edge collaboration, the system comprising:

[0121] The framework model construction unit 201 is used to construct a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer; wherein the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node;

[0122] An acquisition unit 202 is used to acquire the security level of the monitoring area and the edge node status information of the edge layer;

[0123] The optimization problem establishing unit 203 is used to establish an optimization problem with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information under the preset constraint conditions;

[0124] The allocation strategy generation unit 204 is used to convert the optimization problem into a Markov decision problem, solve the Markov decision problem by using the DDPG algorithm at the cloud layer, and generate an optimal task collaboration allocation strategy at the edge layer.

[0125] The third aspect of the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any of the above edge resource allocation methods for monitoring videos are implemented.

[0126] The fourth aspect of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above edge resource allocation methods for monitoring videos when the computer program is executed by a processor.

[0127] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0128] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0130] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0131] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0132] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for allocating edge resources for surveillance video based on cloud-edge collaboration, characterized in that: The method comprises: Construct a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer; the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node; Obtain the security level of the monitoring area and edge node status information of the edge layer; Under the preset constraints, an optimization problem is established with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information; The optimization problem is converted into a Markov decision problem, and the Markov decision problem is solved by using the DDPG algorithm at the cloud layer to generate the optimal task collaboration allocation strategy for the edge layer.

2. The edge resource allocation method for monitoring video according to claim 1, characterized in that: The edge node status information includes monitoring video processing tasks composed of local monitoring video tasks and edge monitoring video tasks, the percentage of computing power allocated to different tasks by each edge node, and the channel transmission rate; the step of establishing an optimization problem with the goal of minimizing the monitoring video task processing delay and bandwidth usage loss at the edge layer according to the security level of the monitoring area and the edge node status information under the preset constraint conditions includes: Based on the local surveillance video task, a local computing model is established according to the percentage of computing power allocated to different tasks by each edge node; Based on the edge surveillance video task, a task offloading model is established according to the percentage of computing power allocated to different tasks by each edge node and the channel transmission rate; Under the preset constraints, an optimization problem is established based on the security level of the monitored area, the local computing model and the task offloading model, with the goal of minimizing the processing delay and bandwidth loss of the monitoring video tasks at the edge layer.

3. The edge resource allocation method for monitoring video according to claim 2, characterized in that: The step of converting the optimization problem into a Markov decision problem, solving the Markov decision problem by using the DDPG algorithm at the cloud layer, and generating an optimal task collaborative allocation strategy at the edge layer includes: The optimization problem is transformed into a Markov decision problem, where the state space is composed of the percentage of computing power allocated to different tasks by each edge node, the average delay of the edge node in processing video frames, the channel transmission rate and the security area level; the percentage of tasks unloaded from each edge node to other edge nodes is used as the action space; and the reward function is set according to the minimization of the delay of monitoring video task processing and bandwidth usage loss at the edge layer; Based on the DDPG algorithm, the policy network and value network are deployed in the cloud layer. The policy network learns the deterministic strategy according to the state space and action space to output the action value under a given state. The value network learns according to the reward function, evaluates the value under a given state, and generates the optimal task collaboration allocation strategy for the edge layer.

4. The edge resource allocation method for monitoring video according to claim 2, characterized in that: The local computing model establishment process is expressed as: Where: represents the computational delay of the i-th edge node processing the local surveillance video task, represents the total number of local video frames that the ith edge node needs to process, I represents the amount of byte image data contained in each frame of video, represents the amount of computation required to process each byte of image data, represents the computing power of the ith edge node, Indicates the percentage of computing power used by the i-th edge node when processing local surveillance video tasks.

5. The edge resource allocation method for monitoring video according to claim 4, characterized in that: The establishment process of the task offloading model is expressed as: Where: represents the total delay of each video frame when the jth edge node transmits the video frame to the ith edge node for task calculation in time slot t, represents the average computational delay of the i-th edge node processing the video frame sent by the j-th edge node in time slot t, represents the transmission delay of the jth edge node transmitting each video frame to the ith edge node in time slot t; in, Where: It represents the ratio of the edge monitoring video task volume to be processed from the jth edge node to the ith edge node in the tth time slot to the current total task volume of the ith edge node, represents the total number of local surveillance video frames that the jth edge node needs to process; It represents the percentage of computing power occupied by the i-th edge node when cooperating with the monitoring video task transmitted by the j-th edge node in time slot t; in, Where: Indicates the total number of edge nodes in the edge layer, It represents the information transmission rate between the jth edge node and the ith edge node in time slot t.

6. The edge resource allocation method for monitoring video according to claim 5, characterized in that: The monitoring video task processing delay of each edge node in the edge layer is expressed as: Where: It represents the average delay of edge node i in processing the surveillance video task in time slot t.

7. The edge resource allocation method for monitoring video according to claim 6, characterized in that: The optimization problem is expressed as: Where: represents a binary variable, where It is used to indicate whether the i-th monitoring area is a key area or an idle area in the t-th time slot; Indicates the weight value of the surveillance video processing task in the key area, represents the weight value of the surveillance video processing task in the idle area; Indicates the ratio of video frame transmission bandwidth to the total channel bandwidth; It represents the ratio of the amount of tasks to be processed from the i-th edge node to the j-th edge node in time slot t to the current total amount of tasks of the i-th edge node; The preset constraints include: Where: C1 represents the constraint on the proportion of edge nodes transmitting surveillance video tasks; C2 represents the constraint that the tasks assigned to edge nodes cannot exceed their computing capabilities; C3 represents the constraint on the real-time processing delay of edge nodes for surveillance video tasks; C4 represents that the computing resources used by the edge layer for surveillance video task processing cannot exceed the total computing capacity of the edge layer.

8. An edge resource allocation system for surveillance video based on cloud-edge collaboration, characterized in that: The system comprises: The framework model building unit is used to build a cloud-edge collaborative framework model consisting of a cloud layer and an edge layer; wherein the edge layer is divided into k monitoring areas, and each monitoring area is equipped with a corresponding edge node; An acquisition unit, used to obtain the security level of the monitoring area and the status information of the edge node of the edge layer; An optimization problem establishing unit, used to establish an optimization problem with the goal of minimizing the processing delay of the monitoring video task and the bandwidth usage loss of the edge layer according to the security level of the monitoring area and the edge node status information under the condition of satisfying the preset constraints; The allocation strategy generation unit is used to convert the optimization problem into a Markov decision problem, solve the Markov decision problem by using the DDPG algorithm at the cloud layer, and generate an optimal task collaboration allocation strategy at the edge layer.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the edge resource allocation method for monitoring video as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the edge resource allocation method for monitoring video as described in any one of claims 1-7 are implemented.

Images