Computing resource calling method and related equipment of intelligent camera cluster

Abstract

The present invention provides a method for calling computing power resources of an intelligent camera cluster and related equipment, wherein the method includes: shooting a target video through an intelligent camera cluster, and allocating preliminary computing power resources to perform video image analysis to obtain video image data. The video image data is sent to a computing power resource allocation system, and tasks are assigned to cameras according to the video image data and a task allocation plan for the cluster is determined. Based on the task allocation plan, the allocation system calculates the second computing power resources of each camera and formulates a computing power resource allocation plan for the intelligent camera cluster. This method significantly improves the resource utilization efficiency and processing speed of the entire system by dynamically allocating computing power resources to each camera in the intelligent camera cluster and adjusting resource allocation according to actual task requirements and video content complexity. This solves the problem of resource waste caused by static resource allocation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a computing resource calling method for an intelligent camera cluster and related equipment. Background Art

[0002] With the widespread application of intelligent monitoring systems, such as urban security, traffic monitoring, and retail analysis, there is a growing demand for efficient management and resource scheduling of intelligent camera clusters. In the prior art, the computing resources of intelligent camera clusters are usually statically allocated and managed, which means that each camera is allocated fixed computing resources regardless of its actual task load or the complexity of the video content. Static computing resource allocation cannot adapt to the dynamic changes in camera load, resulting in some cameras being unable to effectively process video data due to insufficient resources, while other cameras may have excess computing resources idle, resulting in a waste of resources. Summary of the invention

[0003] The main purpose of the present invention is to solve the technical problem that the existing intelligent camera cluster uses static computing resource allocation, resulting in waste of computing resources.

[0004] A first aspect of the present invention provides a method for calling computing power resources of an intelligent camera cluster, and the method for calling computing power resources of the intelligent camera cluster includes:

[0005] Using each smart camera in the smart camera cluster to shoot a target, a corresponding target video is obtained respectively;

[0006] Allocate the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video;

[0007] The video image data is sent to a computing power resource allocation system, and the computing power resource allocation system allocates target tasks to each smart camera according to the video image data to determine a task allocation plan for the smart camera cluster, wherein the task allocation plan includes target tasks corresponding to each smart camera;

[0008] The computing power resource allocation system calculates the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera, and obtains a computing power resource allocation plan;

[0009] The computing power resources are called according to the computing power resource allocation scheme, and the corresponding target tasks are executed by each smart camera in the smart camera cluster according to the called second computing power resources.

[0010] Optionally, in a first implementation of the first aspect of the present invention, allocating the same first computing power resource to each smart camera in the smart camera cluster, and performing video image analysis by using the corresponding first computing power resource through the smart camera corresponding to each target video to obtain video image data of each target video includes:

[0011] Allocate the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform frame processing and image preprocessing on the corresponding target video by each smart camera to obtain a plurality of preprocessed images;

[0012] Performing block processing on the plurality of pre-processed images, dividing the pre-processed images into a plurality of image blocks of fixed sizes, and applying linear projection transformation and adding position encoding processing to each image block to obtain an image block embedding vector with spatial information;

[0013] The dependency relationship between each image block and all other blocks in the corresponding preprocessed image is calculated according to the image block embedding vector through a preset multi-head self-attention mechanism to obtain a global vector of each image block.

[0014] Performing nonlinear changes on the global vector through a preset feedforward neural network to obtain a final embedding vector for each image block;

[0015] The embedded vectors of the image blocks in each preprocessed image are vector aggregated to obtain the image feature vector corresponding to each preprocessed image, and the image feature vector corresponding to each preprocessed image is used as the video image data of the corresponding target video.

[0016] Optionally, in a second implementation of the first aspect of the present invention, sending the video image data to a computing power resource allocation system, and allocating target tasks to each smart camera according to the video image data by the computing power resource allocation system, and determining a task allocation scheme for a smart camera cluster includes:

[0017] The image feature vectors corresponding to each pre-processed image are sent to a computing power resource allocation system, and the image feature vectors corresponding to each pre-processed image are input into a preset task complexity prediction model through the computing power resource allocation system;

[0018] The task complexity prediction model is used to evaluate the task complexity of each smart camera in performing various tasks according to the input image feature vector;

[0019] Generate a preliminary complexity evaluation result corresponding to each smart camera according to the task complexity of each smart camera performing various tasks, and adjust and optimize the preliminary complexity evaluation result in combination with the current computing power resource status of the smart camera cluster to obtain a task complexity evaluation result;

[0020] Sorting the various tasks in the task complexity evaluation result by a preset priority ranking algorithm to obtain the execution priority of the various tasks in the task complexity evaluation result;

[0021] According to the execution priority and the computing power resource conditions of each smart camera, target tasks are allocated to each smart camera through a dynamic scheduling algorithm to obtain a task allocation plan for the smart camera cluster.

[0022] Optionally, in a third implementation of the first aspect of the present invention, the task complexity prediction model includes sub-models corresponding to various types of tasks;

[0023] The task complexity prediction model is used to evaluate the task complexity of each smart camera performing various tasks according to the input image feature vector, including:

[0024] Generate time series data corresponding to the input image feature vector according to the time axis of the corresponding target video through the task complexity prediction model;

[0025] Input the time series data into the sub-models corresponding to various tasks, wherein the sub-models are time convolutional networks;

[0026] Calculating and extracting the temporal features of the time series data through the causal convolution layer and the dilated convolution layer in the temporal convolution network;

[0027] The time series features are nonlinearly mapped using a preset activation function to obtain the task complexity of the corresponding task.

[0028] Optionally, in a fourth implementation of the first aspect of the present invention, the preliminary complexity evaluation results corresponding to each smart camera are generated according to the task complexity of each smart camera performing various tasks, and the preliminary complexity evaluation results are adjusted and optimized in combination with the current computing power resource status of the smart camera cluster, and the task complexity evaluation results obtained include:

[0029] Generate a preliminary complexity evaluation result corresponding to each smart camera according to the task complexity of each smart camera performing various tasks;

[0030] Calculating resource usage data of each smart camera according to the current computing power resource status of the smart camera cluster, wherein the resource usage data includes computing power load, resource utilization rate and data transmission delay;

[0031] Conduct a feasibility analysis on various tasks in the preliminary complexity assessment results according to the resource status data to obtain a feasibility analysis result, and adjust and optimize the preliminary complexity assessment results according to the feasibility analysis results to obtain a task complexity assessment result.

[0032] Optionally, in a fifth implementation of the first aspect of the present invention, the target task is assigned to each smart camera through a dynamic scheduling algorithm according to the execution priority and the computing power resource conditions of each smart camera, and the task allocation scheme for the smart camera cluster is obtained, including:

[0033] A group of Harris Hawks are randomly generated through a preset Harris Hawk optimization algorithm, and each Harris Hawk individual represents a task allocation plan;

[0034] Calculating the fitness value of the task allocation scheme represented by each Harris hawk individual according to a preset fitness function, the execution priority and the computing power resource conditions of each smart camera, and selecting the position of the Harris hawk individual with the highest fitness value as the current prey position;

[0035] Calculating the prey escape energy and jump intensity according to a preset prey escape energy and jump intensity calculation formula, and selecting a corresponding predation strategy according to the prey escape energy and the jump intensity;

[0036] updating the position of each Harris hawk individual according to the predation strategy, recalculating the fitness value of the task allocation scheme represented by each Harris hawk individual after the position update, and reselecting the position of the Harris hawk individual with the highest fitness value as the current prey position;

[0037] Determine whether the preset stop condition is met;

[0038] If satisfied, the current prey position is output as the optimal solution;

[0039] If not, return to the step of calculating the prey escape energy and jump intensity according to the preset prey escape energy and jump intensity calculation formula until the preset stop condition is met;

[0040] A task allocation scheme for the smart camera cluster is determined according to the optimal solution.

[0041] Optionally, in a sixth implementation of the first aspect of the present invention, the computing power resource allocation system calculates the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera to obtain the computing power resource allocation plan, including:

[0042] The task parameters and video image data of the target tasks corresponding to each smart camera are input into a preset computing resource detection model through the computing resource allocation system, wherein the computing resource detection model includes an input layer, an attention mechanism layer, a feature fusion layer, a multi-task detection layer and an output layer;

[0043] The task parameters of the target tasks corresponding to the smart cameras are feature-encoded through the output layer to obtain task features;

[0044] Performing feature fusion on the task feature and the corresponding image feature vector through the attention mechanism layer and the feature fusion layer to obtain a fused vector feature;

[0045] Performing multi-task detection according to the fusion vector feature through the multi-task detection layer to obtain a multi-task detection result, integrating the multi-task detection result to obtain a second computing power resource corresponding to the smart camera, and outputting the second computing power resource corresponding to the smart camera through the output layer;

[0046] The second computing power resources of each smart camera in the smart camera cluster are integrated to obtain a computing power resource allocation plan.

[0047] A second aspect of the present invention provides a computing resource calling device for an intelligent camera cluster, wherein the intelligent camera cluster includes a plurality of intelligent cameras; the computing resource calling device for the intelligent camera cluster includes:

[0048] A video acquisition module is used to shoot a target through each smart camera in the smart camera cluster to obtain a corresponding target video;

[0049] An image analysis module is used to allocate the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video;

[0050] A task allocation module is used to send the video image data to a computing power resource allocation system, and the computing power resource allocation system allocates target tasks to each smart camera according to the video image data to determine a task allocation scheme for the smart camera cluster, wherein the task allocation scheme includes target tasks corresponding to each smart camera;

[0051] A resource calculation module, configured to calculate the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera through the computing power resource allocation system, and obtain a computing power resource allocation plan;

[0052] The resource calling module is used to call computing resources according to the computing resource allocation plan, and execute the corresponding target task according to the called second computing resources through each smart camera in the smart camera cluster.

[0053] The third aspect of the present invention provides a computing power resource calling device for an intelligent camera cluster, comprising: a memory and at least one processor, the memory storing instructions, the memory and the at least one processor being interconnected through lines; the at least one processor calling the instructions in the memory so that the computing power resource calling device of the intelligent camera cluster executes the steps of the above-mentioned computing power resource calling method of the intelligent camera cluster.

[0054] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the steps of the above-mentioned method for calling computing power resources of the intelligent camera cluster.

[0055] The computing power resource calling method and related equipment of the above-mentioned intelligent camera cluster shoot the target video through the intelligent camera cluster, and allocate preliminary computing power resources for video image analysis to obtain video image data. The video image data is sent to the computing power resource allocation system, and the camera is tasked according to the video image data and the task allocation plan of the cluster is determined. Based on the task allocation plan, the allocation system calculates the second computing power resources of each camera and formulates the computing power resource allocation plan of the intelligent camera cluster. This method significantly improves the resource utilization efficiency and processing speed of the entire system by dynamically allocating computing power resources to each camera in the intelligent camera cluster and adjusting resource allocation according to actual task requirements and video content complexity. This solves the problem of resource waste caused by static resource allocation.

[0056] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of an embodiment of a method for calling computing resources of a smart camera cluster in an embodiment of the present invention;

[0059] Figure 2 A schematic diagram of an embodiment of a computing resource calling device for a smart camera cluster in an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of an embodiment of a computing resource calling device for a smart camera cluster in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0063] To facilitate understanding of this embodiment, a method for calling computing resources of a smart camera cluster disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0064] 101. Shooting a target through each smart camera in the smart camera cluster to obtain corresponding target videos respectively;

[0065] In one embodiment of the present invention, smart camera clusters can be applied in multiple fields such as public security, urban management, and commercial operations. These clusters can achieve extensive monitoring and data collection by deploying multiple smart cameras at key locations. In different application scenarios, the layout of smart cameras and the shooting methods of target videos are also different. In the field of urban security and public monitoring, smart camera clusters are usually deployed around major traffic arteries, important public places, and key infrastructure. In traffic monitoring systems, smart camera clusters are widely deployed at intersections, highway entrances and exits, and important transportation hubs. In the field of retail and commercial analysis, smart camera clusters are more often used to analyze customer behavior and optimize store layouts. Different smart cameras shoot according to predetermined angles to obtain target videos.

[0066] 102. Allocate the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video;

[0067] In one embodiment of the present invention, the method of allocating the same first computing power resource to each smart camera in the smart camera cluster, and performing video image analysis by using the corresponding first computing power resource through the smart camera corresponding to each target video to obtain the video image data of each target video includes: allocating the same first computing power resource to each smart camera in the smart camera cluster, and performing frame processing and image preprocessing on the corresponding target video by using the corresponding first computing power resource through each smart camera to obtain multiple preprocessed images; performing block processing on the multiple preprocessed images, cutting the preprocessed images into multiple image blocks of fixed size, and applying linear projection transformation and adding position encoding processing to each image block to obtain an image block embedding vector with spatial information; calculating the dependency relationship between each image block and all other blocks in the corresponding preprocessed image according to the image block embedding vector through a preset multi-head self-attention mechanism, and obtaining the global vector of each image block. The global vector is nonlinearly changed through a preset feedforward neural network to obtain the final embedding vector of each image block; performing vector aggregation on the embedding vectors of the image blocks in each preprocessed image to obtain the image feature vector corresponding to each preprocessed image, and using the image feature vector corresponding to each preprocessed image as the video image data of the corresponding target video.

[0068] Specifically, the same primary computing resources, that is, the first computing resources, are allocated to each camera in the smart camera cluster, which ensures that each camera has sufficient computing power to process the captured video data in the early stage of processing. Using these resources, each camera performs frame processing on the video it captures, decomposing the video into separate image frames, providing a basis for further image analysis. Then, these image frames are subjected to image preprocessing, including but not limited to contrast adjustment, noise removal, and color correction, etc., to improve image quality and reduce the computational burden of subsequent processing. Based on the preprocessing, each image is subdivided into multiple image blocks of fixed size. This step ensures the consistency and comparability of subsequent processing by cutting each image according to a predetermined size. Each image block is then transformed by linear projection, and this transformation can be expressed as L(x)=Ax+b, where x represents the original pixel matrix of the image block, A is the transformation matrix, and b is the offset, which aims to map the image block to a new space to better represent the image content. At the same time, in order to retain the position information of the image block in the original image, position encoding is added to each image block, which is achieved by adding an encoding vector related to the position of the image block, enhancing the model's sensitivity to spatial position. Subsequently, the preset multi-head self-attention mechanism is used, which forms a global vector of the image block by calculating the relationship between the image block embedding vectors. Specifically, the multi-head self-attention mechanism allows the model to take into account the correlation with other image blocks when processing each image block. Finally, the global vector of each image block is nonlinearly changed through a feedforward neural network to further extract and strengthen image features and convert the global vector into the final embedding vector. The embedding vectors of all image blocks are aggregated, and a feature vector representing the entire preprocessed image is finally obtained through simple vector addition or more complex pooling operations. Taking the urban traffic monitoring scene as an example, the video data captured by the smart camera may include busy streets and various traffic participants. Through the processing flow described above, each camera first frames the video and extracts continuous image frames. Suppose in a video frame of a busy intersection, after preprocessing to enhance the image quality, the image is divided into blocks of uniform size, each of which contains different visual information, such as pedestrians, vehicles, or road signs. Applying linear projection transformation to these image blocks, such as transforming a 32x32 pixel image block through a specific matrix A and offset b, can map these image blocks to a feature space with a higher dimension to better capture details and texture information. At the same time, by adding position encoding, the model can understand the relative position of each image block in the entire image, for example, the relative relationship between the vehicle image block at the center of the intersection and the pedestrian image block on the side of the road in space.Using a multi-head self-attention mechanism, the model not only processes each patch separately, but also considers its dependencies with all other patches when processing each patch. This means that if one patch contains a traffic light and another contains a pedestrian, the model can recognize that the pedestrian may stop or walk due to the traffic light signal, thereby understanding the dynamic interaction in the scene. By further transforming these global vectors through a feed-forward neural network, we can obtain more complex and abstract representations of each patch, and these final embedding vectors are aggregated to form a comprehensive feature representation of the entire image. In the case of traffic monitoring, this means that the complexity of the entire intersection scene - including the type of vehicle, the behavior of pedestrians, and the status of traffic signals - is encoded into a high-dimensional feature vector. This feature vector can then be used in a variety of applications, such as traffic flow analysis, anomaly detection, or scene understanding in autonomous driving systems.

[0069] 103. Send the video image data to the computing power resource allocation system, and use the computing power resource allocation system to allocate target tasks to each smart camera according to the video image data, and determine a task allocation plan for the smart camera cluster, wherein the task allocation plan includes the target tasks corresponding to each smart camera;

[0070] In one embodiment of the present invention, the video image data is sent to a computing power resource allocation system, and the computing power resource allocation system is used to allocate target tasks to each smart camera according to the video image data, and the task allocation scheme of the smart camera cluster is determined, including: sending the image feature vector corresponding to each preprocessed image to the computing power resource allocation system, and the computing power resource allocation system inputs the image feature vector corresponding to each preprocessed image into a preset task complexity prediction model; evaluating the task complexity of each smart camera performing various tasks according to the input image feature vector through the task complexity prediction model; generating a preliminary complexity evaluation result corresponding to each smart camera according to the task complexity of each smart camera performing various tasks, and adjusting and optimizing the preliminary complexity evaluation result in combination with the current computing power resource status of the smart camera cluster to obtain a task complexity evaluation result; sorting the various tasks in the task complexity evaluation result by a preset priority ranking algorithm to obtain the execution priority of the various tasks in the task complexity evaluation result; and allocating target tasks to each smart camera by a dynamic scheduling algorithm according to the execution priority and the computing power resource status of each smart camera to obtain a task allocation scheme for the smart camera cluster.

[0071] Specifically, after completing the preliminary processing and feature extraction of the video data, each smart camera sends the obtained image feature vector to the central computing resource allocation system. These feature vectors are the result of high-dimensional compression of the original video data, and contain key information about the video content, such as object type, quantity, motion status, etc. After receiving this data, the computing resource allocation system will input it into a preset task complexity prediction model. The task complexity prediction model is built based on machine learning technology and can evaluate the computational complexity required to perform related analysis tasks based on the input image feature vector. The output of the model is a complexity score that reflects the amount of computing resources required to process the current video data. Next, the computing resource allocation system will adjust and optimize the preliminary complexity evaluation results based on the task complexity score of each smart camera and the current computing resource status. This step involves considering factors such as the resource utilization rate of each camera in the system and the length of the queue of pending tasks to ensure the rationality and efficiency of resource allocation. This process can be achieved through an optimization algorithm, the goal of which is to minimize the overall resource idle rate and task delay time, which can be expressed as:

[0072] ;

[0073] in, is the total computing power resource of the ith camera, is the amount of resources currently used, is the estimated processing time of the tasks in its task queue, and λ is the adjustment parameter used to balance resource utilization and task delay. Through the preset priority ranking algorithm, the computing resource allocation system will sort the execution priorities of various tasks according to the task complexity evaluation results. This step ensures that those tasks with high complexity and high urgency can get priority to obtain sufficient computing resources for processing. The determination of priority is not only based on the complexity score of the task, but also considers multiple dimensions such as the urgency and importance of the task to ensure the responsiveness and service quality of the system. Finally, according to the execution priority of the task and the current computing power resources of each smart camera, the system allocates target tasks to each smart camera through a dynamic scheduling algorithm. In this process, the scheduling algorithm continuously iterates to find the optimal resource allocation plan to achieve balanced allocation between tasks and efficient use of computing power resources.

[0074] Furthermore, the task complexity prediction model includes sub-models corresponding to various tasks; the task complexity of each intelligent camera performing various tasks is evaluated according to the input image feature vector through the task complexity prediction model, including: generating time series data corresponding to the input image feature vector according to the time axis of the corresponding target video through the task complexity prediction model; inputting the time series data into the sub-model corresponding to each task, wherein the sub-model is a temporal convolutional network; calculating and extracting the time series features of the time series data through the causal convolutional layer and the dilated convolutional layer in the temporal convolutional network; and using a preset activation function to perform nonlinear mapping on the time series features to obtain the task complexity of the corresponding task.

[0075] Specifically, the task complexity prediction model needs to extract temporal features from the video image data sent by the smart camera. This process begins by decomposing the video image data according to the time axis and extracting the image feature vector of each frame. Subsequently, these image feature vectors are organized into a series of temporal data to form a temporal feature matrix, in which each row represents the image feature vector at a time point. In this way, the original video data is converted into a new representation, namely the temporal feature matrix, and the temporal data is input into a specially designed sub-model based on the Temporal Convolutional Network (TCN). TCN is a neural network structure that is particularly suitable for processing temporal data. It learns the temporal dependency of input data through stacked convolutional layers. The key to TCN is that it uses causal convolution and dilated convolution layers. Causal convolution ensures that when processing data at a given time point, only data before that time point will be used, avoiding the leakage of future information. Dilated convolution increases the receptive field of the convolution layer by introducing holes, enabling the model to capture temporal dependencies over a longer range without significantly increasing the number of parameters and computational complexity. After processing the time series data using the TCN model, the feature vector of each frame of the image is converted into a new feature representation containing time series information. Then, these new feature representations are nonlinearly mapped through a preset activation function, and finally the task complexity score of the corresponding task is output.

[0076] Furthermore, the preliminary complexity assessment results corresponding to each smart camera are generated according to the task complexity of each smart camera performing various tasks, and the preliminary complexity assessment results are adjusted and optimized in combination with the current computing power resource status of the smart camera cluster to obtain task complexity assessment results including: generating preliminary complexity assessment results corresponding to each smart camera according to the task complexity of each smart camera performing various tasks; calculating resource usage status data of each smart camera according to the current computing power resource status of the smart camera cluster, wherein the resource usage status data includes computing power load, resource utilization rate and data transmission delay; performing feasibility analysis on various tasks in the preliminary complexity assessment results according to the resource status data to obtain feasibility analysis results, and adjusting and optimizing the preliminary complexity assessment results according to the feasibility analysis results to obtain task complexity assessment results.

[0077] Specifically, after generating the preliminary complexity evaluation results corresponding to each smart camera, the resource usage data of each smart camera is calculated according to the computing power resource status of the current smart camera cluster. Resource usage data usually includes key indicators such as computing power load, resource utilization rate, and data transmission delay. These indicators can be calculated using the formula, which is as follows:

[0078] ;

[0079] ;

[0080]

[0081] Indicates camera Resource utilization, and They are the used and total computing resources respectively; Indicates computing load, is the number of pending tasks, is the maximum number of tasks that the system can accommodate; is the data transmission delay, is the average time it takes for data to be transmitted from the camera to the processing center. The system will perform a feasibility analysis on the various tasks in the preliminary complexity assessment results based on the resource usage data. This analysis aims to identify which tasks are feasible under the current resource conditions and which tasks may be difficult to complete due to insufficient resources. The results of the feasibility analysis will be used to adjust and optimize the preliminary complexity assessment results to ensure that the system can prioritize the most important or urgent tasks when resources are limited. Specifically, the feasibility analysis can be performed by comparing the complexity score of each task with the available resources of the corresponding smart camera. If the complexity score of the task exceeds the current available resource threshold of the camera, it indicates that the task is difficult to perform under the current conditions and needs to be adjusted. Based on the feasibility analysis results, the system sorts the various tasks in the task complexity assessment results through a preset priority ranking algorithm, and assigns target tasks to each smart camera accordingly. In addition, the system also needs to take into account factors such as data transmission delay and network bandwidth limitations, which will affect the efficiency and real-time performance of task allocation. Therefore, when making task allocation decisions, the system must not only consider the complexity and priority of the task itself, but also the time and network resource consumption required to transmit task data from the camera to the processing center, to ensure that task allocation can meet processing requirements and will not cause processing efficiency to be low due to network delays or bandwidth limitations.

[0082] Furthermore, according to the execution priority and the computing power resources of each smart camera, the target task is assigned to each smart camera through a dynamic scheduling algorithm to obtain a task allocation plan for the smart camera cluster, including: randomly generating a group of Harris Hawks through a preset Harris Hawk optimization algorithm, and each Harris Hawk individual represents a task allocation plan; calculating the fitness value of the task allocation plan represented by each Harris Hawk individual according to the preset fitness function, the execution priority and the computing power resources of each smart camera, and selecting the position of the Harris Hawk individual with the highest fitness value as the current prey position; calculating the prey escape energy and jump strength according to the preset prey escape energy and jump strength calculation formula. degree, and select a corresponding predation strategy according to the prey escape energy and the jump intensity; update the position of each Harris hawk individual according to the predation strategy, recalculate the fitness value of the task allocation scheme represented by each Harris hawk individual after the position update, and reselect the position of the Harris hawk individual with the highest fitness value as the current prey position; judge whether the preset stop condition is met; if so, output the current prey position as the optimal solution; if not, return to the step of calculating the prey escape energy and jump intensity according to the preset prey escape energy and jump intensity calculation formula until the preset stop condition is met; determine the task allocation scheme of the intelligent camera cluster according to the optimal solution.

[0083] Specifically, a group of Harris Hawks are randomly generated through the preset Harris Hawk optimization algorithm, and each Harris Hawk individual represents a possible task allocation scheme. This process is similar to the search space of the initialization algorithm, where each solution represents a specific camera task allocation and resource allocation method. In order to evaluate the fitness of these individuals (i.e., task allocation schemes), it is necessary to define a fitness function that comprehensively considers the execution priority of the task and the computing power resources of each smart camera. The fitness function can be expressed as:

[0084] ;

[0085] in Represents a task allocation scheme The fitness value of Representation scheme The comprehensive score of the task execution priority in Representation scheme Efficiency rating of computing resource utilization, It is a plan Possible data processing delays, , and is a weight coefficient used to balance the influence of different indicators. Subsequently, the algorithm calculates the escape energy and jump strength of each prey (i.e., the task allocation scheme) according to the preset prey escape energy and jump strength calculation formula. This step simulates the process of Harris Hawk observing and evaluating the state of prey and preparing to hunt. Escape Energy The formula can be Calculate, where is the initial energy, is the energy consumption rate, is time, simulating the decrease in energy of the prey over time. Jump strength Then we can use $J= 2( 1- ( ) ), where rand() generates a random number between [0,1], simulating the rapid movement and direction change in Harris's hawk's predation behavior. Based on the escape energy of the prey and the jumping strength of the Harris's hawk, the algorithm selects the corresponding predation strategy to update the position of each Harris's hawk individual, that is, to update the task allocation plan. This process is repeated, and the fitness value of the individual is recalculated in each iteration, and the individual with the highest fitness value is selected as the current prey position, that is, the current optimal task allocation plan. This process continues until the preset stop condition is met, such as reaching the maximum number of iterations or the improvement in the fitness value is less than a certain threshold. At this time, the algorithm outputs the current prey position, that is, the task allocation plan with the highest fitness value, as the optimal solution.

[0086] 104. Calculate the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera through the computing power resource allocation system to obtain a computing power resource allocation plan;

[0087] In one embodiment of the present invention, the computing power resource allocation system calculates the second computing power resources of each smart camera in the smart camera cluster according to the target tasks and video image data corresponding to each smart camera to obtain the computing power resource allocation plan, including: inputting the task parameters and video image data of the target tasks corresponding to each smart camera into a preset computing power resource detection model through the computing power resource allocation system, wherein the computing power resource detection model includes an input layer, an attention mechanism layer, a feature fusion layer, a multi-task detection layer and an output layer; feature encoding the task parameters of the target tasks corresponding to each smart camera through the output layer to obtain task features; feature fusion of the task features and the corresponding image feature vectors through the attention mechanism layer and the feature fusion layer to obtain fusion vector features; multi-task detection is performed according to the fusion vector features through the multi-task detection layer to obtain multi-task detection results, and the multi-task detection results are integrated to obtain the second computing power resources of the corresponding smart camera, and the second computing power resources of the corresponding smart camera are output through the output layer; the second computing power resources of each smart camera in the smart camera cluster are integrated to obtain a computing power resource allocation plan.

[0088] Specifically, the computing resource allocation system collects the video image data captured by each smart camera and the corresponding target task parameters. These task parameters may include the type, priority, expected completion time, etc. of the task, while the video image data directly reflects the complexity of task processing. The system uses this data as input and sends it to the preset computing resource detection model. The design of the model adopts a multi-layer structure, including the input layer, the attention mechanism layer, the feature fusion layer, the multi-task detection layer and the output layer. Each layer is responsible for specific functions and tasks, and works together to achieve accurate prediction and allocation of computing resource requirements. In the input layer, the model first preprocesses the received task parameters and video image data, including standardization and denoising, to ensure the quality and consistency of the data. Subsequently, the task parameters are feature encoded through the output layer. This process involves converting the task parameters into a series of numerical features that can better characterize the nature and requirements of the task. After entering the attention mechanism layer, the model will use the attention mechanism to enhance the focus on the important parts of the task features and video image features. The attention mechanism enables the model to automatically focus on the most critical information by calculating the weight distribution between features. In the feature fusion layer, the model deeply fuses the task features and weighted video image features to generate a comprehensive feature representation that can fully reflect the task requirements and the complexity of the video content. Next, the multi-task detection layer will use the fused vector features to simultaneously execute multiple prediction tasks related to computing resource allocation, such as predicting the execution time of the task, the amount of computing resources required, etc. The design of this layer enables the model to predict the resource requirements of multiple tasks at the same time, improving processing efficiency. The prediction results will be integrated and output through the output layer to form the second computing resource demand prediction for the corresponding smart camera. Finally, the system conducts integrated analysis based on the second computing resource demand prediction results of all smart cameras and formulates the final computing resource allocation plan.

[0089] 105. Call computing resources according to the computing resource allocation plan, and use each smart camera in the smart camera cluster to execute a corresponding target task according to the called second computing resources.

[0090] In one embodiment of the present invention, the corresponding computing resources are dynamically called for each smart camera through the internal resource management and scheduling mechanism. This involves adjusting the resource allocation of the data processing center, such as the computing power allocation of the CPU and GPU, and the configuration of storage and network bandwidth. Each camera adjusts its own task processing strategy according to the received computing resource allocation instruction, such as allocating more computing resources to handle high-complexity video analysis tasks, or optimizing data transmission to reduce delays. In order to ensure that the task can be executed efficiently, the system also needs to monitor the resource usage and task execution status of each smart camera in real time. For example, when a camera is responsible for analyzing the video data of a busy intersection, the system determines that this task requires higher computing resources based on the prediction model. The system then dynamically adjusts to allocate more computing resources to the camera to ensure that the video analysis task can be completed quickly and accurately. At the same time, the system ensures the real-time and accuracy of resource allocation through real-time monitoring, thereby maximizing resource utilization while ensuring efficient execution of tasks.

[0091] In this embodiment, the target video is captured by a smart camera cluster, and preliminary computing resources are allocated for video image analysis to obtain video image data. The video image data is sent to the computing resource allocation system, and the camera is tasked according to the video image data and the task allocation plan of the cluster is determined. Based on the task allocation plan, the allocation system calculates the second computing resources of each camera and formulates a computing resource allocation plan for the smart camera cluster. This method significantly improves the resource utilization efficiency and processing speed of the entire system by dynamically allocating computing resources to each camera in the smart camera cluster and adjusting resource allocation according to actual task requirements and video content complexity. This solves the problem of resource waste caused by static resource allocation.

[0092] The above describes the method for calling the computing power resources of the smart camera cluster in the embodiment of the present invention. The following describes the device for calling the computing power resources of the smart camera cluster in the embodiment of the present invention. Figure 2 , an embodiment of a computing resource calling device of an intelligent camera cluster in an embodiment of the present invention includes:

[0093] The video acquisition module 201 is used to shoot a target through each smart camera in the smart camera cluster to obtain a corresponding target video;

[0094] The image analysis module 202 is used to allocate the same first computing power resources to each smart camera in the smart camera cluster, and use the corresponding first computing power resources to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video;

[0095] The task allocation module 203 is used to send the video image data to the computing power resource allocation system, and the computing power resource allocation system allocates target tasks to each smart camera according to the video image data to determine the task allocation plan of the smart camera cluster, wherein the task allocation plan includes the target tasks corresponding to each smart camera;

[0096] The resource calculation module 204 is used to calculate the second computing power resources of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera through the computing power resource allocation system to obtain a computing power resource allocation plan;

[0097] The resource calling module 205 is used to call computing resources according to the computing resource allocation plan, and execute the corresponding target task according to the called second computing resources through each smart camera in the smart camera cluster.

[0098] In an embodiment of the present invention, the computing power resource calling device of the intelligent camera cluster runs the computing power resource calling method of the intelligent camera cluster. The computing power resource calling device of the intelligent camera cluster shoots the target video through the intelligent camera cluster, and allocates preliminary computing power resources to perform video image analysis to obtain video image data. The video image data is sent to the computing power resource allocation system, and the camera is tasked according to the video image data and the task allocation plan of the cluster is determined. Based on the task allocation plan, the allocation system calculates the second computing power resource of each camera and formulates the computing power resource allocation plan of the intelligent camera cluster. This method significantly improves the resource utilization efficiency and processing speed of the entire system by dynamically allocating computing power resources to each camera in the intelligent camera cluster and adjusting resource allocation according to actual task requirements and video content complexity. This solves the problem of resource waste caused by static resource allocation.

[0099] above Figure 2 The computing power resource calling device of the intelligent camera cluster in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The computing power resource calling device of the intelligent camera cluster in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0100] Figure 3It is a structural diagram of a computing resource calling device for a smart camera cluster provided by an embodiment of the present invention. The computing resource calling device 300 for the smart camera cluster may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the computing resource calling device 300 for the smart camera cluster. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the computing resource calling device 300 for the smart camera cluster to implement the steps of the computing resource calling method for the smart camera cluster.

[0101] The computing resource calling device 300 of the smart camera cluster may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The computing power resource calling device structure of the smart camera cluster shown does not constitute a limitation on the computing power resource calling device of the smart camera cluster provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0102] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the computing power resource calling method of the smart camera cluster.

[0103] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0104] 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 is essentially or the part that contributes to the prior art or the whole 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 to enable a computer device (which can be a personal computer, server, or 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), random access memory (RAM), disk or optical disk and other media that can store program code.

[0105] 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 calling computing resources of a smart camera cluster, characterized in that: The smart camera cluster includes multiple smart cameras; the computing power resource calling method includes: Using each smart camera in the smart camera cluster to shoot a target, a corresponding target video is obtained respectively; Allocate the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video; the allocating the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video includes: allocating the same first computing power resource to each smart camera in the smart camera cluster, and use the corresponding first computing power resource to perform frame processing and image preprocessing on the corresponding target video through each smart camera to obtain multiple preprocessed images; The image is processed in blocks, the preprocessed image is cut into multiple image blocks of fixed size, and linear projection transformation and position encoding processing are applied to each image block to obtain an image block embedding vector with spatial information; the dependency relationship between each image block and all other blocks in the corresponding preprocessed image is calculated according to the image block embedding vector by a preset multi-head self-attention mechanism to obtain a global vector of each image block; the global vector is nonlinearly changed by a preset feedforward neural network to obtain a final embedding vector of each image block; the embedding vectors of the image blocks in each preprocessed image are vector aggregated to obtain an image feature vector corresponding to each preprocessed image, and the image feature vector corresponding to each preprocessed image is used as the video image data of the corresponding target video; The video image data is sent to the computing power resource allocation system, and the computing power resource allocation system is used to allocate target tasks to each smart camera according to the video image data, and determine the task allocation plan of the smart camera cluster, wherein the task allocation plan includes the target tasks corresponding to each smart camera; the video image data is sent to the computing power resource allocation system, and the computing power resource allocation system is used to allocate target tasks to each smart camera according to the video image data, and determine the task allocation plan of the smart camera cluster includes: sending the image feature vector corresponding to each pre-processed image to the computing power resource allocation system, and inputting the image feature vector corresponding to each pre-processed image into a preset task complexity prediction model through the computing power resource allocation system; The task complexity prediction model evaluates the task complexity of each smart camera performing various tasks according to the input image feature vector; generates a preliminary complexity evaluation result corresponding to each smart camera according to the task complexity of each smart camera performing various tasks, and adjusts and optimizes the preliminary complexity evaluation result in combination with the current computing power resource status of the smart camera cluster to obtain the task complexity evaluation result; sorts the various tasks in the task complexity evaluation result through a preset priority ranking algorithm to obtain the execution priority of the various tasks in the task complexity evaluation result; allocates target tasks to each smart camera according to the execution priority and the computing power resource status of each smart camera through a dynamic scheduling algorithm to obtain a task allocation plan for the smart camera cluster; The computing power resource allocation system calculates the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera, and obtains a computing power resource allocation plan; The computing power resources are called according to the computing power resource allocation plan, and the corresponding target tasks are executed by each smart camera in the smart camera cluster according to the called second computing power resources.

2. The computing resource calling method according to claim 1, characterized in that: The task complexity prediction model includes sub-models corresponding to various types of tasks; The task complexity prediction model is used to evaluate the task complexity of each smart camera performing various tasks according to the input image feature vector, including: Generate time series data corresponding to the input image feature vector according to the time axis of the corresponding target video through the task complexity prediction model; Input the time series data into the sub-models corresponding to various tasks, wherein the sub-models are time convolutional networks; Calculating and extracting the temporal features of the time series data through the causal convolution layer and the dilated convolution layer in the temporal convolution network; The time series features are nonlinearly mapped using a preset activation function to obtain the task complexity of the corresponding task.

3. The computing resource calling method according to claim 1, characterized in that: The preliminary complexity evaluation results corresponding to each smart camera are generated according to the task complexity of each smart camera performing various tasks, and the preliminary complexity evaluation results are adjusted and optimized in combination with the current computing power resource status of the smart camera cluster, and the task complexity evaluation results obtained include: Generate a preliminary complexity evaluation result corresponding to each smart camera according to the task complexity of each smart camera performing various tasks; Calculating resource usage data of each smart camera according to the current computing power resource status of the smart camera cluster, wherein the resource usage data includes computing power load, resource utilization rate and data transmission delay; A feasibility analysis is performed on each type of task in the preliminary complexity assessment result according to the resource usage status data to obtain a feasibility analysis result, and the preliminary complexity assessment result is adjusted and optimized according to the feasibility analysis result to obtain a task complexity assessment result.

4. The computing resource calling method according to claim 1, characterized in that: According to the execution priority and the computing power resource conditions of each smart camera, target tasks are assigned to each smart camera through a dynamic scheduling algorithm to obtain a task allocation scheme for the smart camera cluster, including: A group of Harris Hawks are randomly generated through a preset Harris Hawk optimization algorithm, and each Harris Hawk individual represents a task allocation plan; Calculating the fitness value of the task allocation scheme represented by each Harris hawk individual according to a preset fitness function, the execution priority and the computing power resource conditions of each smart camera, and selecting the position of the Harris hawk individual with the highest fitness value as the current prey position; Calculating the prey escape energy and jump intensity according to a preset prey escape energy and jump intensity calculation formula, and selecting a corresponding predation strategy according to the prey escape energy and the jump intensity; updating the position of each Harris hawk individual according to the predation strategy, recalculating the fitness value of the task allocation scheme represented by each Harris hawk individual after the position update, and reselecting the position of the Harris hawk individual with the highest fitness value as the current prey position; Determine whether the preset stop condition is met; If satisfied, the current prey position is output as the optimal solution; If not, return to the step of calculating the prey escape energy and jump intensity according to the preset prey escape energy and jump intensity calculation formula until the preset stop condition is met; A task allocation scheme for the smart camera cluster is determined according to the optimal solution.

5. The computing resource calling method according to claim 1, characterized in that: The computing resource allocation system calculates the second computing resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera to obtain the computing resource allocation plan, which includes: The task parameters and video image data of the target tasks corresponding to each smart camera are input into a preset computing resource detection model through the computing resource allocation system, wherein the computing resource detection model includes an input layer, an attention mechanism layer, a feature fusion layer, a multi-task detection layer and an output layer; The task parameters of the target tasks corresponding to the smart cameras are feature-encoded through the output layer to obtain task features; Performing feature fusion on the task feature and the corresponding image feature vector through the attention mechanism layer and the feature fusion layer to obtain a fused vector feature; Performing multi-task detection according to the fusion vector feature through the multi-task detection layer to obtain a multi-task detection result, integrating the multi-task detection result to obtain a second computing power resource corresponding to the smart camera, and outputting the second computing power resource corresponding to the smart camera through the output layer; The second computing power resources of each smart camera in the smart camera cluster are integrated to obtain a computing power resource allocation plan.

6. A computing resource calling device for a smart camera cluster, characterized in that: The smart camera cluster includes a plurality of smart cameras; the computing resource calling device of the smart camera cluster includes: A video acquisition module is used to shoot a target through each smart camera in the smart camera cluster to obtain a corresponding target video; An image analysis module is used to allocate the same first computing power resources to each smart camera in the smart camera cluster, and use the corresponding first computing power resources to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video; the allocation of the same first computing power resources to each smart camera in the smart camera cluster, and use the corresponding first computing power resources to perform video image analysis through the smart camera corresponding to each target video to obtain video image data of each target video includes: allocating the same first computing power resources to each smart camera in the smart camera cluster, and using the corresponding first computing power resources to perform frame processing and image preprocessing on the corresponding target video by each smart camera to obtain multiple preprocessed images; A plurality of preprocessed images are processed in blocks, the preprocessed images are cut into a plurality of image blocks of fixed sizes, and a linear projection transformation and position encoding processing are applied to each image block to obtain an image block embedding vector with spatial information; the dependency relationship between each image block and all other blocks in the corresponding preprocessed image is calculated according to the image block embedding vector by a preset multi-head self-attention mechanism to obtain a global vector of each image block; a nonlinear change is performed on the global vector by a preset feedforward neural network to obtain a final embedding vector of each image block; the embedding vectors of the image blocks in each preprocessed image are vector aggregated to obtain an image feature vector corresponding to each preprocessed image, and the image feature vector corresponding to each preprocessed image is used as video image data of the corresponding target video; A task allocation module is used to send the video image data to a computing power resource allocation system, and the computing power resource allocation system allocates target tasks to each smart camera according to the video image data to determine the task allocation plan of the smart camera cluster, wherein the task allocation plan includes the target task corresponding to each smart camera; the sending of the video image data to the computing power resource allocation system, and the computing power resource allocation system allocates target tasks to each smart camera according to the video image data to determine the task allocation plan of the smart camera cluster, including: sending the image feature vector corresponding to each pre-processed image to the computing power resource allocation system, and inputting the image feature vector corresponding to each pre-processed image into a preset task complexity prediction model through the computing power resource allocation system. ; The task complexity of each smart camera performing various tasks is evaluated according to the input image feature vector through the task complexity prediction model; the preliminary complexity evaluation results corresponding to each smart camera are generated according to the task complexity of each smart camera performing various tasks, and the preliminary complexity evaluation results are adjusted and optimized in combination with the current computing power resource status of the smart camera cluster to obtain the task complexity evaluation results; the various tasks in the task complexity evaluation results are sorted through a preset priority ranking algorithm to obtain the execution priority of the various tasks in the task complexity evaluation results; according to the execution priority and the computing power resource status of each smart camera, the target tasks are assigned to each smart camera through a dynamic scheduling algorithm to obtain the task allocation plan of the smart camera cluster; A resource calculation module, configured to calculate the second computing power resource of each smart camera in the smart camera cluster according to the target task and video image data corresponding to each smart camera through the computing power resource allocation system, and obtain a computing power resource allocation plan; The resource calling module is used to call computing resources according to the computing resource allocation plan, and execute the corresponding target task according to the called second computing resources through each smart camera in the smart camera cluster.

7. A computing resource calling device for a smart camera cluster, characterized in that: The computing resource calling device of the smart camera cluster includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the computing power resource calling device of the smart camera cluster executes the steps of the computing power resource calling method of the smart camera cluster as described in any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, the steps of the computing resource calling method of the intelligent camera cluster as described in any one of claims 1-5 are implemented.

Images