Image matching method and system based on edge computing power

Abstract

The invention discloses an image matching method and system based on edge computing power, and the method comprises the steps: collecting a computing power index and a network index of an edge node in real time, and generating a resource state matrix according to the computing power index and the network index; according to the resolution and the feature complexity of the road image, disassembling an image matching task into a preprocessing sub-task, a feature extraction sub-task and a feature matching sub-task, and constructing a node weight function based on the resource state matrix; adopting an improved genetic algorithm, taking maximization of the total weight as a target, and generating an optimal distribution scheme from subtasks to edge nodes in an iteration process; and when the node time delay of the edge node exceeds a threshold value or the task fails, migrating the matched subtask to a suboptimal node based on incremental check point data, and correspondingly updating the resource state matrix. Therefore, a dynamic perception-precise scheduling-rapid self-healing complete link is formed, and the system stability in a complex road scene is guaranteed while the end-to-end time delay is reduced.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image matching method and system based on edge computing power. Background Art

[0002] In an edge computing environment, image matching tasks involve high computing load (such as feature extraction and feature matching) and real-time requirements, and therefore require multi-node collaborative processing.

[0003] In traditional edge computing scenarios, when a task is interrupted due to a node failure or network fluctuation, the full amount of checkpoint data involved in the task needs to be migrated to other nodes. The checkpoint data includes the complete feature library, matching intermediate results, etc.

[0004] However, due to the huge amount of feature data required for autonomous driving road image matching, full data transmission will occupy high bandwidth resources, resulting in a significant increase in migration latency. Especially in bandwidth-constrained edge environments, task recovery time may exceed real-time requirements, and repeated transmission of redundant data will waste bandwidth resources, seriously restricting system reliability and efficiency. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an image matching method and system based on edge computing power, aiming to solve the above-mentioned problems recorded in the prior art.

[0006] A first aspect of the present invention is to provide an image matching method based on edge computing power, the method comprising: Collect computing power indicators and network indicators of edge nodes in real time, and generate a resource status matrix based on the computing power indicators and the network indicators; According to the resolution and feature complexity of the road image, the image matching task is decomposed into preprocessing, feature extraction, and feature matching subtasks, and a node weight function is constructed based on the resource status matrix; An improved genetic algorithm is used to maximize the total weight and generate the optimal allocation scheme of subtasks to edge nodes in the iterative process; When the node delay of the edge node exceeds a threshold or the task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly.

[0007] According to one aspect of the above technical solution, an improved genetic algorithm is used to maximize the total weight, and the steps of generating an optimal allocation scheme of subtasks to edge nodes in an iterative process include: Randomly create multiple task allocation schemes corresponding to subtasks to edge nodes, and determine whether each task allocation scheme meets the pre-set node resource constraint conditions; Calculate the comprehensive scores of the task allocation solutions that meet the node resource constraint conditions according to pre-set calculation rules; The comprehensive scores of each of the task allocation schemes are ranked, and the crossover probability values ​​and mutation probability values ​​between the task allocation schemes are automatically adjusted to iteratively generate an optimal allocation scheme for subtasks to edge nodes.

[0008] According to one aspect of the above technical solution, the step of calculating the comprehensive scores of the task allocation solutions that meet the node resource constraint conditions according to pre-set calculation rules includes: For each subtask corresponding to the task allocation scheme, score it according to the computing power index and network index of the target node to obtain a computing power score and a network score; A weight coefficient is set according to the task type of the subtask, and the computing power score and the network score are weighted and summed to generate a subtask allocation score; For subtasks with serial dependencies, if they are assigned to different nodes, the total score will be deducted based on the communication delay between nodes; The scores of all subtasks in each task allocation scheme are accumulated to obtain a comprehensive score of the corresponding task allocation scheme.

[0009] According to one aspect of the above technical solution, when the node latency of the edge node exceeds a threshold or the task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly, including: Monitor the task processing delay of the edge node in real time. If it exceeds a preset threshold or the task execution fails, mark the node as a faulty node. According to the current resource status matrix, select the node with the highest network score and remaining available computing power greater than or equal to the subtask requirement as the suboptimal node; Extracting incremental checkpoint data from the failed node and transmitting the incremental checkpoint data and task context state to the suboptimal node; After the suboptimal node resumes the task, the computing power index and the network index in the resource status matrix are updated according to the resource occupancy status after the migration.

[0010] According to one aspect of the above technical solution, the step of extracting incremental checkpoint data from the faulty node and transmitting the incremental checkpoint data and task context state to the suboptimal node includes: Extracting the newly added or modified data in the faulty node since the last complete checkpoint, including the difference feature point set, partial matching results, and task progress indicator; Perform timestamp filtering on the difference feature point set to retain only feature data related to the current task; The characteristic data is compressed and transmitted in fragments. After the suboptimal node receives the characteristic data, the data integrity is verified by hash check. If the check fails, retransmission is triggered; And the current execution environment is restored according to the task context state, including the number of completed feature pairs, the current matching progress pointer and the inter-node communication link state.

[0011] According to one aspect of the above technical solution, the step of compressing and transmitting the feature data in segments includes: Using a hybrid compression algorithm of Huffman coding and run-length coding to losslessly compress the incremental checkpoint data; The compressed data is segmented into fixed-size data packets, and the characteristic data is transmitted according to the transmission priority.

[0012] According to one aspect of the above technical solution, the transmission priority is to first transmit the basic feature data and then transmit the incremental result data.

[0013] A second aspect of the present invention is to provide an image matching system based on edge computing power, which is applied to the method described in the above technical solution. The system includes: A matrix generation module is used to collect computing power indicators and network indicators of edge nodes in real time, and generate a resource status matrix based on the computing power indicators and the network indicators; A task decomposition module is used to decompose the image matching task into preprocessing, feature extraction, and feature matching subtasks according to the resolution and feature complexity of the road image, and to construct a node weight function based on the resource status matrix; The task iteration module is used to generate the optimal allocation scheme of subtasks to edge nodes during the iteration process with the goal of maximizing the total weight using an improved genetic algorithm; The task migration module is used to migrate the matching subtask to a suboptimal node based on incremental checkpoint data when the node delay of the edge node exceeds a threshold or the task fails, and to update the resource status matrix accordingly.

[0014] The third aspect of the present invention is to provide a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described in the above technical solution.

[0015] The fourth aspect of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the above technical solution when executing the program.

[0016] Compared with the prior art, the image matching method based on edge computing power shown in the present invention has the following beneficial effects: The present invention generates a resource status matrix by collecting node computing power and network indicators in real time, globally perceives resource dynamics, and eliminates resource mismatches caused by static allocation. Then, in view of the high-resolution and complex features of road images in complex traffic scenarios, the task is decomposed into preprocessing, feature extraction, and matching subtasks. A differentiated weight function is designed to prioritize the allocation of computing power-intensive tasks to high-performance nodes and network-sensitive tasks to high-bandwidth nodes. An improved genetic algorithm is used to accelerate the search for the optimal allocation solution by dynamically adjusting the crossover / mutation probability values. At the same time, dependent tasks are forced to be allocated nearby to reduce the cumulative delay of cross-node communication. Finally, when a node fails, tasks are quickly migrated to the suboptimal node based on incremental checkpoints, and the resource status is updated in a closed loop to prevent the spread of the failure's impact. This forms a complete chain of dynamic perception, precise scheduling, and rapid self-healing, which reduces end-to-end latency while ensuring system stability in complex road scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which: Figure 1 Schematic diagram of the process of an image matching method based on edge computing power in one embodiment of the present invention; Figure 2 This is a structural block diagram of an image matching system based on edge computing power in one embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, features, and advantages of the present invention more readily apparent, the following detailed description of specific embodiments of the present invention is provided in conjunction with the accompanying drawings. The accompanying drawings illustrate several embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0019] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly attached to the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1 See also Figure 1 The first embodiment of the present invention provides an image matching method based on edge computing power, the method comprising steps S10 to S40: Step S10: collecting computing power indicators and network indicators of edge nodes in real time, and generating a resource status matrix according to the computing power indicators and the network indicators.

[0022] First of all, it should be noted that in the road image matching and processing system based on edge computing power, in order to achieve low-latency response for tasks such as traffic sign recognition and obstacle detection, it is necessary to dynamically select the optimal edge node.

[0023] Specifically, when performing image matching and subsequent image processing of road images based on edge computing power, in order to match and determine the best edge node for image processing among multiple edge nodes, it is first necessary to collect the computing power indicators and network indicators corresponding to the edge nodes used for image processing in real time, and then generate the resource status matrix corresponding to the edge nodes used for road image processing based on the above computing power indicators and network indicators.

[0024] By generating a resource status matrix, the road image processing system can dynamically perceive the resource status of edge nodes, provide a quantitative basis for subsequent task allocation, reduce task latency, and reduce the matching failure rate caused by node overload.

[0025] In step S20 , the image matching task is decomposed into preprocessing, feature extraction, and feature matching subtasks according to the resolution and feature complexity of the road image, and a node weight function is constructed based on the resource status matrix.

[0026] In this embodiment, in the road image matching task, the core goal of decomposing the image matching task into preprocessing, feature extraction, and feature matching subtasks is to adapt the complex image processing process to heterogeneous edge node resources through task decomposition and dynamic weight allocation.

[0027] It should be noted that road images typically have high resolution, such as 4K images output by traffic surveillance cameras, and dynamically changing feature complexity. For example, the number of feature points increases dramatically when vehicles are densely packed. Therefore, in this embodiment, before performing formal image matching, the task needs to be broken down into the following subtasks: Preprocessing subtask: This includes image denoising, resizing, and ROI cropping. This subtask relies on network bandwidth and requires fast transmission of the original image or preprocessing results. Feature extraction subtask: Extract key points and descriptors using algorithms such as SIFT and ORB. In road scenes, the number of feature points varies with vehicle / pedestrian density. This subtask relies heavily on GPU computing power, especially memory capacity and parallel computing capabilities. Feature matching subtask: This task compares features across consecutive frames or across multiple cameras, such as vehicle re-identification. The matching complexity is proportional to the square of the number of feature points, requiring a balanced allocation of computing and communication resources to avoid overloading a single node.

[0028] In this embodiment, when constructing node weight functions based on the resource status matrix, differentiated weight functions are designed based on subtask type and road scenario requirements. Specifically, these weight functions include a preprocessing subtask weight function, a feature extraction subtask weight function, and a feature matching subtask weight function. The preprocessing subtask weight function prioritizes network metrics to ensure rapid transmission of preprocessed images to the next step. The feature extraction subtask weight function emphasizes GPU computing power, prioritizing nodes with sufficient computing power for intensive computations. The feature matching subtask weight function balances computational efficiency with communication quality to avoid overall delays in returning matching results due to network congestion.

[0029] In addition, if there is a serial dependency between the preprocessing and feature extraction subtasks (for example, the ROI cropping result needs to be used for feature extraction immediately), these two subtasks are bound and assigned to the same node or a low-latency adjacent node to avoid additional delays introduced by cross-node communication.

[0030] Step S30 , using an improved genetic algorithm with the goal of maximizing the total weight, generates an optimal allocation scheme of subtasks to edge nodes during an iterative process.

[0031] In this embodiment, an improved genetic algorithm is used to maximize the total weight. The steps of generating an optimal allocation scheme of subtasks to edge nodes in an iterative process include: Randomly create multiple task allocation schemes corresponding to subtasks to edge nodes, and determine whether each task allocation scheme meets the pre-set node resource constraint conditions; Calculate the comprehensive scores of the task allocation solutions that meet the node resource constraint conditions according to pre-set calculation rules; The comprehensive scores of each of the task allocation schemes are ranked, and the crossover probability values ​​and mutation probability values ​​between the task allocation schemes are automatically adjusted to iteratively generate an optimal allocation scheme for subtasks to edge nodes.

[0032] Specifically, the purpose of this embodiment in generating the optimal allocation scheme of subtasks to edge nodes during the iteration process is to create diversified task allocation schemes and filter out invalid schemes that do not meet resource constraints. The core of this is to generate diversified task allocation schemes through genetic coding and to filter out invalid solutions based on resource constraints. More specifically, each allocation scenario is represented as a sequence of integers, where the sequence length equals the number of subtasks, and each position corresponds to the node number to which the subtask is assigned. For example, the preprocessing subtask is assigned to node 1, feature extraction to node 2, and so on. The initial population randomly generates multiple such sequences to ensure a wide range of possible allocations. Each node is checked for its ability to meet the real-time resource requirements of the subtask, where: Computing power constraint: The node's current computing load (such as CPU / GPU utilization) plus the estimated consumption of subtasks must not exceed a safety threshold (such as 90%) to avoid task failure caused by overload. Memory / GPU memory constraints: Verify whether the node's remaining memory and GPU memory meet the minimum requirements of the subtask. For example, feature extraction requires GPU memory ≥ the model requirement value.

[0033] For solutions that don't meet constraints, nodes assigned to conflicting subtasks are prioritized for replacement. For example, nodes with insufficient video memory are replaced with available nodes in the same region, rather than directly eliminated, to preserve potentially high-quality genes. This embodiment directly maps task allocation relationships through gene encoding, simplifying subsequent optimization operations. By combining hard constraint filtering with dynamic correction, the effectiveness and diversity of the initial population are improved.

[0034] The step of calculating the comprehensive scores of the task allocation schemes that meet the node resource constraint conditions according to pre-set calculation rules includes: For each subtask corresponding to the task allocation scheme, score it according to the computing power index and network index of the target node to obtain a computing power score and a network score; A weight coefficient is set according to the task type of the subtask, and the computing power score and the network score are weighted and summed to generate a subtask allocation score; For subtasks with serial dependencies, if they are assigned to different nodes, the total score will be deducted based on the communication delay between nodes; The scores of all subtasks in each task allocation scheme are accumulated to obtain a comprehensive score of the corresponding task allocation scheme.

[0035] Specifically, the purpose of calculating the comprehensive score of the task allocation scheme that meets the node resource constraint condition in this embodiment is to quantitatively evaluate the comprehensive quality of the allocation scheme based on the subtask characteristics and the node status.

[0036] More specifically, the computing power score selects node computing power metrics based on the subtask type. For example, feature extraction tasks focus on available GPU computing power (remaining GPU utilization) and video memory compatibility (remaining video memory / task required video memory). Preprocessing tasks focus on available CPU computing power and memory headroom. The network score, on the other hand, is based on the data transmission requirements of the subtask. For example, preprocessing requires uploading the original image, while feature matching requires returning the results. The score evaluates node bandwidth, latency, and packet loss rate, among other metrics.

[0037] Also, dynamically adjust the computing power and network score weights based on the subtask type. For example: Preprocessing tasks (network-sensitive): network weight accounts for 70%, computing power accounts for 30%; Feature extraction task (computing power sensitive): computing power accounts for 80% of the weight, and the network accounts for 20%.

[0038] If dependent subtasks (e.g., preprocessing to feature extraction) are assigned to different nodes, the total score is penalized proportionally to the inter-node communication latency, forcing dependent tasks to be assigned to the nearest node. Finally, the weighted scores of all subtasks are accumulated to form a composite score for the allocation scheme, which reflects overall resource utilization and task execution efficiency.

[0039] Step S40: When the node delay of the edge node exceeds a threshold or the task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly.

[0040] In this embodiment, the purpose of migrating the matching subtask and updating the resource status matrix accordingly is to solve the problem of inefficient task recovery caused by edge node failure or network fluctuation, and to ensure the continuity and low latency of the image matching task through incremental checkpoint migration and dynamic resource update mechanism.

[0041] Compared with the prior art, the image matching method based on edge computing power shown in this embodiment has the following beneficial effects: This embodiment generates a resource status matrix by collecting node computing power and network metrics in real time, achieving global awareness of resource dynamics and eliminating resource mismatches caused by static allocation. Furthermore, in response to the high-resolution and complex features of road images in complex traffic scenarios, tasks are broken down into preprocessing, feature extraction, and matching subtasks. Differentiated weighting functions are designed to prioritize computing-intensive tasks to high-performance nodes and network-sensitive tasks to high-bandwidth nodes. An improved genetic algorithm dynamically adjusts crossover / mutation probabilities to accelerate the search for the optimal allocation solution. Dependent tasks are forced to be allocated nearby, reducing the cumulative latency of cross-node communication. Finally, in the event of a node failure, tasks are rapidly migrated to suboptimal nodes based on incremental checkpoints, and resource status is updated in a closed-loop manner to prevent the spread of the failure's impact. This complete chain of dynamic awareness, precise scheduling, and rapid self-healing is formed, reducing end-to-end latency while ensuring system stability in complex road scenarios.

[0042] Example 2 The second embodiment of the present invention also provides an image matching method based on edge computing power. The image matching method based on edge computing power shown in this embodiment is basically similar to the image matching method based on edge computing power shown in the first embodiment, except that: In this embodiment, when the node latency of the edge node exceeds a threshold or a task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly, including: Monitor the task processing delay of the edge node in real time. If it exceeds a preset threshold or the task execution fails, mark the node as a faulty node. According to the current resource status matrix, select the node with the highest network score and remaining available computing power greater than or equal to the subtask requirement as the suboptimal node; Extracting incremental checkpoint data from the failed node and transmitting the incremental checkpoint data and task context state to the suboptimal node; After the suboptimal node resumes the task, the computing power index and the network index in the resource status matrix are updated according to the resource occupancy status after the migration.

[0043] Specifically, in road image processing, the feature points of consecutive frames have strong spatiotemporal continuity. For example, vehicle trajectories have extremely strong coherence. Therefore, only new or modified feature points since the last complete checkpoint are extracted, such as ORB feature points added by vehicle movement, to avoid repeated transmission of the full set of features.

[0044] The step of extracting incremental checkpoint data from the faulty node and transmitting the incremental checkpoint data and the task context state to the suboptimal node includes: Extracting the newly added or modified data in the faulty node since the last complete checkpoint, including the difference feature point set, partial matching results, and task progress indicator; Perform timestamp filtering on the difference feature point set to retain only feature data related to the current task; The characteristic data is compressed and transmitted in fragments. After the suboptimal node receives the characteristic data, the data integrity is verified by hash check. If the check fails, retransmission is triggered; And the current execution environment is restored according to the task context state, including the number of completed feature pairs, the current matching progress pointer and the inter-node communication link state.

[0045] Specifically, based on the task progress identifier, such as the timestamp interval, only feature data related to the current task window, such as the image frames within the last 5 seconds, is retained. For example, if a fault occurs at the 10th second, all historical feature points before the 5th second are discarded to focus on the key data before and after the fault.

[0046] More specifically, after receiving data, the suboptimal node calculates the corresponding MD5 hash value for each packet and compares it with the source node. If the check fails, only the specific fragment is retransmitted. For example, if the third fragment is damaged, only that fragment is retransmitted. Furthermore, based on the number of completed feature matches, repeated calculations are skipped and execution resumes from the last successful match point before the failure, restoring the data transmission connection between the nodes.

[0047] The steps of compressing and fragmenting the feature data include: Using a hybrid compression algorithm of Huffman coding and run-length coding to losslessly compress the incremental checkpoint data; The compressed data is segmented into fixed-size data packets, and the characteristic data is transmitted according to the transmission priority.

[0048] In addition, the transmission priority is to transmit the basic feature data first and then transmit the incremental result data.

[0049] Specifically, the Huffman coding in the hybrid compression algorithm improves the compression rate by performing frequency statistics and encoding on the feature point coordinates, that is, integer data. Continuous repeated value compression is performed on the binary feature descriptor to further improve the compression rate.

[0050] More specifically, when data is segmented and transmitted after data compression, the compressed data is first segmented into data packets with a preset data capacity, such as 512KB, and the feature point coordinates and descriptors are transmitted first to ensure that the suboptimal node quickly resumes the feature matching process, and then the incremental result data is delayed.

[0051] Compared with the first embodiment, the method shown in this embodiment has the following beneficial effects: This embodiment significantly improves the efficiency, accuracy, and robustness of road image matching in fault migration scenarios through a closed-loop design that combines data simplification, transmission optimization, precise recovery, and resource regulation, providing reliable technical support for real-time decision-making in smart transportation.

[0052] Example 3 See also Figure 2 A third embodiment of the present invention provides an image matching system based on edge computing power, which is applied to the method described in any of the above embodiments. The system includes: The matrix generation module 10 is used to collect computing power indicators and network indicators of edge nodes in real time, and generate a resource status matrix based on the computing power indicators and the network indicators; A task decomposition module 20 is used to decompose the image matching task into preprocessing, feature extraction, and feature matching subtasks according to the resolution and feature complexity of the road image, and to construct a node weight function based on the resource status matrix; The task iteration module 30 is used to generate an optimal allocation scheme of subtasks to edge nodes during the iteration process by using an improved genetic algorithm with the goal of maximizing the total weight; The task migration module 40 is used to migrate the matching subtask to a suboptimal node based on incremental checkpoint data when the node delay of the edge node exceeds a threshold or the task fails, and to update the resource status matrix accordingly.

[0053] Compared with the prior art, the image matching method based on edge computing power shown in this embodiment has the following beneficial effects: This embodiment generates a resource status matrix by collecting node computing power and network metrics in real time, achieving global awareness of resource dynamics and eliminating resource mismatches caused by static allocation. Furthermore, in response to the high-resolution and complex features of road images in complex traffic scenarios, tasks are broken down into preprocessing, feature extraction, and matching subtasks. Differentiated weighting functions are designed to prioritize computing-intensive tasks to high-performance nodes and network-sensitive tasks to high-bandwidth nodes. An improved genetic algorithm dynamically adjusts crossover / mutation probabilities to accelerate the search for the optimal allocation solution. Dependent tasks are forced to be allocated nearby, reducing the cumulative latency of cross-node communication. Finally, in the event of a node failure, tasks are rapidly migrated to suboptimal nodes based on incremental checkpoints, and resource status is updated in a closed-loop manner to prevent the spread of the failure's impact. This complete chain of dynamic awareness, precise scheduling, and rapid self-healing is formed, reducing end-to-end latency while ensuring system stability in complex road scenarios.

[0054] Example 4 A fourth embodiment of the present invention provides a readable storage medium having computer instructions stored thereon, which implement the steps of the method described in the above embodiment when executed by a processor.

[0055] Example 5 A fifth embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the above embodiment when executing the program.

[0056] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0057] Those skilled in the art will appreciate that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or for use in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, a "computer-readable storage medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device.

[0058] More specific examples (a non-exhaustive list) of computer-readable storage media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable storage medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0059] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0060] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0061] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. An image matching method based on edge computing, characterized in that: The method comprises: Collect computing power indicators and network indicators of edge nodes in real time, and generate a resource status matrix based on the computing power indicators and the network indicators; According to the resolution and feature complexity of the road image, the image matching task is decomposed into preprocessing, feature extraction, and feature matching subtasks, and a node weight function is constructed based on the resource status matrix; An improved genetic algorithm is used to maximize the total weight and generate the optimal allocation scheme of subtasks to edge nodes in the iterative process; When the node delay of the edge node exceeds a threshold or the task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly.

2. The image matching method based on edge computing according to claim 1, characterized in that: The improved genetic algorithm is used to maximize the total weight. The steps of generating the optimal allocation scheme of subtasks to edge nodes in the iterative process include: Randomly create multiple task allocation schemes corresponding to subtasks to edge nodes, and determine whether each task allocation scheme meets the pre-set node resource constraint conditions; Calculate the comprehensive scores of the task allocation solutions that meet the node resource constraint conditions according to pre-set calculation rules; The comprehensive scores of each of the task allocation schemes are ranked, and the crossover probability values ​​and mutation probability values ​​between the task allocation schemes are automatically adjusted to iteratively generate an optimal allocation scheme for subtasks to edge nodes.

3. The image matching method based on edge computing according to claim 2, characterized in that: The steps of respectively calculating the comprehensive scores of the task allocation schemes that meet the node resource constraint conditions according to a pre-set calculation rule include: For each subtask corresponding to the task allocation scheme, score it according to the computing power index and network index of the target node to obtain a computing power score and a network score; A weight coefficient is set according to the task type of the subtask, and the computing power score and the network score are weighted and summed to generate a subtask allocation score; For subtasks with serial dependencies, if they are assigned to different nodes, the total score will be deducted based on the communication delay between nodes; The scores of all subtasks in each task allocation scheme are accumulated to obtain a comprehensive score of the corresponding task allocation scheme.

4. The image matching method based on edge computing according to any one of claims 1 to 3, characterized in that: When the node latency of the edge node exceeds a threshold or the task fails, the matching subtask is migrated to a suboptimal node based on incremental checkpoint data, and the resource status matrix is ​​updated accordingly, including: Monitor the task processing delay of the edge node in real time. If it exceeds a preset threshold or the task execution fails, mark the node as a faulty node. According to the current resource status matrix, select the node with the highest network score and remaining available computing power greater than or equal to the subtask requirement as the suboptimal node; Extracting incremental checkpoint data from the failed node and transmitting the incremental checkpoint data and task context state to the suboptimal node; After the suboptimal node resumes the task, the computing power index and the network index in the resource status matrix are updated according to the resource occupancy status after the migration.

5. The image matching method based on edge computing according to claim 4, characterized in that: The step of extracting incremental checkpoint data from the faulty node and transmitting the incremental checkpoint data and the task context state to the suboptimal node comprises: Extracting the newly added or modified data in the faulty node since the last complete checkpoint, including the difference feature point set, partial matching results, and task progress indicator; Perform timestamp filtering on the difference feature point set to retain only feature data related to the current task; The characteristic data is compressed and transmitted in fragments. After the suboptimal node receives the characteristic data, the data integrity is verified by hash check. If the check fails, retransmission is triggered; And the current execution environment is restored according to the task context state, including the number of completed feature pairs, the current matching progress pointer and the inter-node communication link state.

6. The image matching method based on edge computing according to claim 5, characterized in that: The steps of compressing and fragmenting the characteristic data include: Using a hybrid compression algorithm of Huffman coding and run-length coding to losslessly compress the incremental checkpoint data; The compressed data is segmented into fixed-size data packets, and the characteristic data is transmitted according to the transmission priority.

7. The image matching method based on edge computing according to claim 6, characterized in that: The transmission priority is to transmit the basic feature data first and then transmit the incremental result data.

8. An image matching system based on edge computing, characterized in that: The method according to any one of claims 1 to 7, wherein the system comprises: A matrix generation module is used to collect computing power indicators and network indicators of edge nodes in real time, and generate a resource status matrix based on the computing power indicators and the network indicators; A task decomposition module is used to decompose the image matching task into preprocessing, feature extraction, and feature matching subtasks according to the resolution and feature complexity of the road image, and to construct a node weight function based on the resource status matrix; The task iteration module is used to generate the optimal allocation scheme of subtasks to edge nodes during the iteration process with the goal of maximizing the total weight using an improved genetic algorithm; The task migration module is used to migrate the matching subtask to a suboptimal node based on incremental checkpoint data when the node delay of the edge node exceeds a threshold or the task fails, and to update the resource status matrix accordingly.

9. A readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images