Intelligent street lamp data analysis system based on edge calculation

Abstract

The invention relates to the technical field of edge computing and intelligent traffic, and discloses an intelligent street lamp data analysis system based on edge computing, which comprises the steps of realizing NPU / GPU computing power unit hardware-level isolation control through physical pin mapping of an edge chip; generating a cellular topology network based on geographic coordinates, and dynamically calling cameras, communication modules and computing power units of adjacent nodes to construct a micro-service chain; binary codes are generated through illumination, rain and fog and people flow parameters, and lightweight model plug-ins are triggered to be loaded as required; a bus protocol is improved to realize pin-level computing power cross-node exclusive occupation and microsecond-level response; and the event priority and the elastic release rule are defined to ensure abnormal self-healing. According to the method, the contradiction between resource scheduling stiffness and privacy delay of a traditional edge system is broken through, algorithm self-adaption in an emergency scene is achieved through cooperation of hardware isolation and dynamic coding, data safety is guaranteed through decentralized topology and geographical driving verification, and a complete solution which is high in real-time performance, high in reliability and low in power consumption is provided for an intelligent street lamp.

Description

Technical Field

[0001] The present invention relates to a smart street lamp data analysis system based on edge computing, belonging to the technical field of smart street lamps. Background Art

[0002] As an important part of the intelligentization of urban infrastructure, the core requirement of smart street lamps is to process multi-source heterogeneous data (such as videos, environmental sensors, traffic flow) in real time and quickly respond to emergency scenarios (such as traffic accidents, extreme weather). Traditional technical solutions usually adopt the following methods: 1. Cloud centralized processing: Uploading street lamp node data to the cloud server for analysis, but limited by network latency (usually >500ms) and data privacy risks (such as camera video streams being exposed to the public network), it is difficult to meet the real-time and security requirements of emergency scenarios. 2. Edge virtualized resource pool: Deploying a virtualization platform (such as Docker / Kubernetes) at the edge node to achieve computing power sharing, but the context switching and resource contention of the virtualization layer lead to latency fluctuations (10 - 50ms), and malicious tasks may steal data through shared memory, unable to meet the requirements of high-security scenarios. 3. Fixed algorithm model: Presetting a unified AI model at the street lamp node, but static algorithms cannot adapt to dynamic environmental changes (such as the video misjudgment rate >40% due to rain and fog interference), and the model needs to be updated remotely manually, with a lag in response and high operation and maintenance costs.

[0003] To alleviate the above problems, the industry's attempts at improvement solutions include: data desensitization and compression: reducing privacy risks by blurring or downsampling, but the data availability drops significantly (such as blurred videos being unable to identify license plate details); lightweight model deployment: using techniques such as pruning and quantization to reduce the model size, but still relying on the central node to update parameters regularly and unable to achieve real-time switching of environmental adaptability; local federated learning: adjacent nodes sharing model parameters, but traditional protocols require multiple global aggregations, with a large communication overhead (single update >100ms). However, these improvement solutions still have an essential contradiction: it is difficult to balance resource elasticity and real-time performance, data privacy and algorithm accuracy restrict each other, and dynamic environments and static models cannot be adapted. Summary of the Invention

[0004] The present invention provides a smart street lamp data analysis system based on edge computing, and its main purpose is to solve the problems of rigid resource scheduling, difficulty in balancing privacy and real-time performance, and mismatch between dynamic environments and static algorithms in traditional edge computing systems.

[0005] To achieve the above object, a smart street lamp data analysis system based on edge computing provided by the present invention is characterized in that it includes: Node computing power unit pin mapping configuration: Map the computing power units of the edge computing chips of each street lamp node to physical pin numbers, and the computing power units include but are not limited to NPU cores and GPU stream processors; Define pin status: High level indicates that the computing power unit is occupied, and low level indicates that it is idle and can be called.

[0006] Event triggering and microservice chain dynamic construction: When any node detects a target event, generate a trigger signal including the event type and geographical coordinates; Based on the geographical coordinates and the preset topological relationship of adjacent nodes, call the following hardware modules to construct a cross-node microservice chain: a. The camera module of adjacent node A, used to perform video enhancement analysis; b. The communication module of adjacent node B, used to generate traffic guidance instructions; c. The GPU computing power unit of adjacent node C, used to accelerate model inference.

[0007] Pin-level computing power cross-node physical exclusive control: Specify the pins of the target node through the bus signal position, so that the master node exclusively occupies its corresponding computing power unit, including: a. The master node sends a computing power request signal to the adjacent node; b. After the adjacent node verifies the consistency of the geographical coordinates of the master node and the preset topological relationship, set the target pin to high level; c. During the occupancy of the computing power unit, the local tasks of the adjacent node are prohibited from accessing the computing power unit corresponding to the set pin.

[0008] Environment parameter-driven model adaptive loading: Real-time collect environment parameters, including but not limited to light intensity, rain and fog concentration, and population density; Generate a binary code according to the environment parameters to trigger the loading of a preset lightweight model plugin, including: a. When the rain and fog concentration coding bit is 1, load an infrared enhancement feature extraction plugin; b. When the light intensity coding bit is 0, load a low-light image restoration plugin.

[0009] Service chain task termination and resource release: When any of the following conditions is met, reset all the positioned pins to low level and disassemble the microservice chain: a. The target event is processed; b. The inter-node communication anomaly lasts for more than the preset threshold time; c. The environment parameter coding triggers a higher-priority event type.

[0010] Preferably, the preset method of the adjacent node topological relationship includes: Input the geographical coordinates of each node into a preset grid division algorithm to generate a honeycomb topological network; Define the communication priority of the nodes within the honeycomb unit as: When the Euclidean distance between two nodes satisfies d≤50 meters, a direct communication link is preferentially established; When 50 meters < d≤100 meters, signal relay needs to be performed through an intermediate node.

[0011] Preferably, the specific method for generating the binary encoding of the environmental parameters includes: dividing the light intensity into three threshold intervals: low illuminance < 50 Lux, normal illuminance 50 - 1000 Lux, strong light interference > 1000 Lux, corresponding to the encoding bits 00, 01, 10 respectively; correlating the rain and fog concentration with the visibility, defining that the encoding bit is 1 when the visibility < 30 meters, otherwise 0; the final encoding format is [light encoding][rain and fog encoding][pedestrian flow density encoding], where the pedestrian flow density encoding is dynamically generated by statistically counting the number of pedestrians in real time and the preset density threshold.

[0012] Preferably, the method for loading the model plug-in includes: pre-storing the plug-in library in the L2 cache of the edge chip, each plug-in containing a head identifier and a functional code segment; when the environmental encoding triggers the plug-in loading, loading the specified plug-in from the cache to the NPU instruction register through direct memory access (DMA), and the loading delay is less than 2 μs.

[0013] Preferably, the method for verifying the computing power request signal includes: after receiving the request, the neighboring node performs the following verification steps: verifying whether the geographical coordinates of the master node are in the preset list of trusted nodes; verifying the conflict between the event type in the request signal and the current environmental encoding, and rejecting the response if the conflict probability > 30%; checking whether the remaining computing power units of itself meet the request requirements, and returning the alternative pin number if not.

[0014] Compared with the problems described in the background art, the beneficial effects of the present invention are:

[0015] 1. By directly mapping the physical pins of the edge chip to the computing power units (such as binding the NPU core to Pin23 - 26), hardware-level isolation and control of cross-node computing power resources are achieved. Combining with the dynamic model assembly mechanism triggered by environmental parameters (such as rain and fog concentration, light intensity), the system can switch lightweight plug-ins (such as infrared enhancement modules) in real time and automatically adapt to the optimal algorithm in complex environments. The synergistic effect of hardware isolation and dynamic models realizes the integrated control of "resource exclusivity - environmental adaptation" in edge computing for the first time, breaking through the response bottleneck of traditional systems in sudden scenarios.

[0016] 2. Based on the cellular network generated by geographical coordinates (such as a 50-meter radius cell), communication and computing power invocation between nodes strictly follow the geographical proximity rule (such as direct connection within 50 meters, relay within 100 meters). Through the preset list of trusted nodes (such as nodes authorized by the traffic management center) and geographical signature verification, a decentralized trusted collaboration network is constructed. This design abandons the single-point dependence on the traditional central server and utilizes the fixed topology characteristics of street lights to achieve a double improvement in the localization of resource scheduling and anti-attack capabilities.

[0017] 3. Encode multi-source environmental data (light, rain / fog, pedestrian flow density) into a compact binary signal (such as 01-1-1) to directly drive the on-demand loading of pre-set model plugins. For example, the rain / fog encoding triggers the infrared enhancement plugin to solve the misjudgment problem of cameras in low visibility conditions. The encoding mechanism deeply binds the physical environment with the algorithm logic, enabling the system to achieve scene adaptation of algorithm functions without complex calculations, significantly reducing the computing load of edge nodes.

[0018] 4. By improving the bus protocol (such as expanding the computing power control field in CAN 2.0B), compress operations such as computing power requests and pin state switching into single-frame instruction transmission. The node directly sets the target pin level based on the event priority (such as traffic accidents being the highest priority), skipping the context switching and resource scheduling of the traditional virtualization layer. This mechanism reduces the latency of cross-node computing power collaboration to the microsecond level, providing near-real-time response capabilities for emergency scenarios.

[0019] 5. Define event priorities and exception detection rules (such as automatically releasing resources when the communication interruption exceeds 50 ms), and through dynamic priority calculation (such as high-priority events preempting low-priority tasks) and elastic release logic, ensure that the system can still maintain its core functions in extreme environments. The rapid resource recovery and log recording in abnormal states enable the system to have self-healing capabilities, avoiding the risk of global collapse caused by local failures in traditional solutions. Brief Description of the Drawings

[0020] Figure 1 It is a schematic diagram of the hardware structure of the intelligent street lamp data analysis system based on edge computing of the present invention.

[0021] Figure 2 It is a schematic diagram of the cellular topology network of the intelligent street lamp data analysis system based on edge computing of the present invention.

[0022] The implementation, functional features, and advantages of the present invention will be further described in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Embodiments

[0023] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0024] The embodiments of the present application provide an intelligent street lamp data analysis system based on edge computing. An intelligent street lamp data analysis system based on edge computing, characterized by including:

[0025] Pin mapping configuration of the node computing power unit: Map the computing power units of the edge computing chips of each street lamp node to physical pin numbers, and the computing power units include but are not limited to NPU cores and GPU stream processors; Define the pin state: High level indicates that the computing power unit is occupied, and low level indicates idle and callable;

[0026] Event Triggering and Dynamic Construction of Microservice Chain: When any node detects a target event, a trigger signal containing the event type and geographical coordinates is generated; based on the geographical coordinates and the preset topological relationship of neighboring nodes, the following hardware modules are called to construct a cross-node microservice chain: a. The camera module of neighboring node A, which is used to perform video enhancement analysis; b. The communication module of neighboring node B, which is used to generate traffic guidance instructions; c. The GPU computing power unit of neighboring node C, which is used to accelerate model inference;

[0027] Pin-Level Computing Power Cross-Node Physical Exclusive Control: The specified pin of the target node is located through the bus signal, enabling the master node to exclusively occupy its corresponding computing power unit, including: a. The master node sends a computing power request signal to the neighboring node; b. After the neighboring node verifies the consistency between the geographical coordinates of the master node and the preset topological relationship, it sets the target pin to high level; c. During the occupancy of the computing power unit, the local tasks of the neighboring node are prohibited from accessing the computing power unit corresponding to the set pin;

[0028] Environment Parameter-Driven Model Adaptive Loading: Environment parameters are collected in real time, including but not limited to light intensity, rain and fog concentration, and pedestrian flow density; binary codes are generated according to the environment parameters to trigger the loading of preset lightweight model plugins, including: a. When the rain and fog concentration coding bit is 1, an infrared enhancement feature extraction plugin is loaded; b. When the light intensity coding bit is 0, a low-illumination image restoration plugin is loaded;

[0029] Service Chain Task Termination and Resource Release: When any of the following conditions is met, all the positioned pins are reset to low level and the microservice chain is disassembled: a. The target event is processed; b. The inter-node communication anomaly lasts for more than the preset threshold time; c. The environment parameter coding triggers a higher-priority event type.

[0030] Preferably, the preset method of the neighboring node topological relationship includes: inputting the geographical coordinates of each node into a preset grid division algorithm to generate a cellular topological network; defining the communication priority of nodes within a cellular unit as: when the Euclidean distance between two nodes satisfies d ≤ 50 meters, a direct communication link is preferentially established; when 50 meters < d ≤ 100 meters, signal relay needs to be carried out through an intermediate node.

[0031] Preferably, the specific method for generating binary codes from environment parameters includes: dividing the light intensity into three threshold intervals: low illumination < 50 Lux, normal illumination 50 - 1000 Lux, strong light interference > 1000 Lux, corresponding to the coding bits 00, 01, 10 respectively; associating the rain and fog concentration with visibility, defining that the coding bit is 1 when the visibility < 30 meters, otherwise 0; the final coding format is [light coding][rain and fog coding][pedestrian flow density coding], where the pedestrian flow density coding is dynamically generated by statistically counting the number of pedestrians in real time and the preset density threshold.

[0032] Preferably, the method for loading the model plugin includes: pre-storing a plugin library in the L2 cache of the edge chip, where each plugin contains a header identifier and a functional code segment; when the environment encoding triggers plugin loading, the specified plugin is loaded from the cache to the NPU instruction register through direct memory access (DMA), and the loading latency is less than 2 μs.

[0033] Preferably, the method for verifying the computing power request signal includes: after receiving the request, the neighboring node performs the following verification steps: verifying whether the geographical coordinates of the master node are in the preset trusted node list; verifying the conflict between the event type in the request signal and the current environment encoding, and if the conflict probability > 30%, rejecting the response; checking whether its remaining computing power units meet the request requirements, and if not, returning an alternative pin number.

[0034] Embodiment 1: In the pin mapping configuration of the node computing power unit, an edge computing chip is used, and the physical pin mapping rules between its NPU core and GPU stream processors are as follows: NPU_Core1 → Pin23 - 26 (computing power units 1 - 4), GPU_Stream1 → Pin27 - 30 (computing power units 5 - 8, 0.5 TFLOPS per unit). Pin status control: high / low level signals are output through the chip's GPIO interface. A high level (3.3V) indicates that the computing power unit is occupied, and a low level (0V) indicates it is idle.

[0035] In its event triggering and microservice chain construction, event detection: The camera of node A identifies a traffic accident (vehicle collision), and the trigger signal format is: {event type: traffic accident, geographical coordinates: (116.40°E, 39.90°N)}.

[0036] Topological relationship generation: Based on the honeycomb division algorithm (Voronoi diagram), with a honeycomb radius of 50 meters, a neighboring node list of node A is generated: Node B (distance 42 meters): Deploy a 5G communication module for issuing traffic guidance instructions; Node C (distance 48 meters): An idle GPU computing power unit (Pin28). In its microservice chain invocation, node A invokes the communication module of node B to send a guidance instruction to the traffic management center; node A occupies the GPU computing power unit of node C (Pin28 is set to high level) to accelerate accident video analysis.

[0037] In terms of cross-node control of pin-level computing power, such as bus protocol improvement: a computing power control field is newly added to the extended frame of the CAN 2.0B protocol, with the following structure: | Priority (4 bits) | Target pin (12 bits) | Level (1 bit) | CRC (8 bits); Example instruction: Priority = 3 (traffic accident), Target pin = 28, Level = 1; Verification and response: After receiving the request, Node C executes: Verify the coordinates of Node A in the trusted list (preset traffic management center authorized node); Conflict probability calculation: Based on historical data statistics, the conflict probability between traffic accidents and environmental monitoring is <5% (lower than the 30% threshold), and response is allowed; Return the occupancy confirmation signal for pin 28. In its environment parameter-driven model loading, parameter collection and encoding, such as light intensity: 85 Lux (normal illuminance, encoded 01); Rain and fog concentration: visibility of 25 meters (encoded 1); Pedestrian flow density: 2 people per square meter (encoded 1); Final encoding: 01-1-1 (decimal value 13). And in plugin loading, the infrared enhancement plugin is loaded according to the encoding 13 (rain and fog encoding = 1); DMA loads the plugin from the L2 cache (4KB aligned address 0x8000) to the NPU instruction register, taking 1.8 μs (meeting the requirement of <2 μs); Its service chain termination and resource release, such as termination condition: After the accident is handled, Node A sends a reset instruction to Node C (Pin28 is set to low level); Exception handling: If the communication is interrupted for more than 50 ms (preset threshold), all occupied pins are automatically released and a log is recorded.

[0038] Embodiment 2: Aiming at the fact that the computing power scheduling of existing smart street lights depends on cloud decisions, resulting in an event response delay exceeding 500 milliseconds and being unable to quickly reorganize local computing power resources according to sudden environmental changes (such as heavy rain and a sudden increase in the number of pedestrians at night). This embodiment realizes microsecond-level precise scheduling of hardware-level computing power resources through geographical grid dynamic division and encoding rule design.

[0039] In a specific implementation, geographical grid division and communication link establishment are first carried out. Based on the Beidou positioning coordinates (longitude, latitude, accuracy ±0.1 m) of each street lamp node, the Thiessen polygon algorithm is used to divide honeycomb grid units with a coverage radius of 50 m. If the centroid distance between two grid units is less than 70 m, a relay communication link is established, and the improved CAN bus protocol is used to transmit the computing power request signal. The protocol frame includes a 4-bit priority field (for example, the high-priority code is 0001), a 12-bit target hardware pin address (such as NPU pin 0x3A1), and a 1-bit level activation status. Then, environmental parameter encoding and computing power mapping rules are constructed, and binary encoding rules for three types of environmental parameters are defined: Illumination intensity: 2-bit encoding, 00 represents night mode (<50 Lux, triggering the image enhancement model), 01 represents cloudy day mode (50 - 1000 Lux, triggering the target detection model), 10 represents strong light mode (1000 Lux, triggering the anti-glare model), and the threshold is set according to the ISO 8995-2023 lighting standard. Rain and fog level: 1-bit encoding, 0 indicates no rain or fog, 1 indicates the presence of rain or fog (triggering the rain and fog removal algorithm, and the model is pre-loaded to the NPU address 0x3A1); Pedestrian flow density: Dynamic bit-width encoding. The number of pedestrians is counted by an infrared sensor every 10 seconds, and the bit-width is determined by "taking the integer part of the logarithm to the base 2 and adding 1" (for example, 5 people correspond to 3-bit encoding 101). In terms of computing power conflict decision-making and priority control, when multiple nodes request the same NPU pin simultaneously, the system calculates the ratio of the historical conflict times of this pin to the total request times. If the conflict probability exceeds 30% (determined by regression analysis of historical data), the current request is rejected, and the system automatically switches to the idle pin of the adjacent grid node.

[0040] Embodiment 3: In this embodiment, we adopt a computing power allocation method based on physical pin mapping to further optimize the dynamic scheduling and cross-node control of computing power. Specifically, the computing units of each smart street lamp node, such as NPU and GPU, achieve hardware-level resource isolation through mapping with physical pins, ensuring that the computing power resources between nodes do not interfere with each other. The pin state indicates the occupancy state of the computing power unit through high and low levels: when the pin is at a high level, it means that the computing power unit is occupied, and a low level means that the computing power unit is idle and available. When an event occurs, such as the detection of a traffic accident, the master node generates a trigger signal and, based on the preset geographical topology relationship, dynamically schedules the resources of adjacent nodes. After the event is triggered, the master node collaborates with adjacent nodes, calls the camera module of adjacent nodes for video enhancement analysis, the communication module generates traffic instructions, and at the same time calls the GPU computing power unit to accelerate model inference. This optimization method ensures the maximum utilization of computing power, avoids resource waste, and enables the computing power scheduling to quickly respond to different emergencies.

[0041] To improve the system's adaptability in complex environments, we introduce a dynamic model loading mechanism based on environmental parameters (such as light, rain and fog concentration, and pedestrian flow density). Specifically, the collection of environmental parameters is monitored in real time through sensors on street lamp nodes, and the collected light, rain and fog, and pedestrian flow data are encoded into binary signals. For example, the light intensity is divided into three cases: low illuminance (<50Lux), normal illuminance (50 - 1000Lux), and strong light interference (1000Lux), and each interval corresponds to a binary code. An association is established between the rain and fog concentration and visibility. If the visibility is less than 30 meters, it is encoded as 1, otherwise as 0. The encoding of pedestrian flow density is dynamically calculated by counting the number of pedestrians in the street lamp area and according to a preset density threshold, and finally a binary code [light encoding][rain and fog encoding][pedestrian flow density encoding] is generated.

[0042] When the system receives these environmental parameter encodings, according to the preset rules, it loads the corresponding model plug-ins. For example, when the rain and fog concentration encoding is 1, the system will automatically load the infrared enhancement feature extraction plug-in; when the light intensity is below 50Lux, the low illuminance image restoration plug-in is loaded. This mechanism can effectively reduce the system's computing load and achieve real-time response.

[0043] To further improve the efficiency of cross-node computing power collaboration, this embodiment optimizes the transmission and verification of computing power request signals. When requesting computing power between nodes, the master node first sends a computing power request signal to adjacent nodes through a bus signal, including the event type and geographical coordinates. The adjacent nodes verify the master node's request according to the preset geographical topology relationship. After receiving the computing power request signal, the adjacent nodes perform the following verification: check whether the geographical coordinates of the master node are in the trusted node list; determine whether there is a conflict between the current environmental encoding and the requested event type. If the conflict probability is greater than the set threshold (e.g., 30%), the response is rejected; if the computing power unit of this node meets the requirements, the target pin is set to high level, indicating that the computing power unit has been allocated to the master node. By optimizing this process, the communication delay between nodes is further reduced, and resource waste caused by computing power conflicts is avoided.

[0044] When dealing with multi - event concurrency, to avoid low - priority events occupying a large amount of resources and causing delays in high - priority tasks, we designed a resource scheduling mechanism based on event priorities. Each event type has a corresponding priority value. For example, the priority of traffic accident events is higher than that of environmental monitoring events. When multiple events occur, the system automatically schedules resources according to event priorities to ensure that high - priority tasks are processed first. In addition, during the task execution process, if the communication anomaly between nodes exceeds a set threshold (such as 50 ms), the system will automatically terminate the current task, reset all occupied pins, and release resources. At the same time, the system will record the anomaly log and perform self - diagnosis to ensure that the core functions can still operate continuously in extreme environments.

[0045] To reduce the impact of plugin loading delay on system response, this embodiment optimizes the plugin loading process. Each plugin library is pre - stored in the L2 cache of the edge chip. The plugin includes a functional code segment and an identifier. When the environmental parameter encoding triggers a certain plugin, the system loads the plugin from the cache to the NPU instruction register through the direct memory access (DMA) mechanism. The delay of this process is controlled within 2 microseconds, ensuring that the plugin can be quickly loaded and executed in a dynamic environment. When the event processing is completed or the task is interrupted, the system will immediately release the occupied computing power unit resources and restore the initial state. Through this mechanism, the system can recycle resources in a timely manner when resources are scarce or the task is terminated, avoiding system crashes or deadlocks.

[0046] Embodiment 4: This embodiment provides an optimized intelligent street lamp data analysis system, which can effectively improve the real - time response ability, resource management efficiency, and security of the system. By improving and optimizing the key modules in the system, it is ensured that the solution can operate efficiently in actual deployment and has strong scalability and operability. In this embodiment, to solve the rigid problem of traditional system resource scheduling, a computing power resource scheduling method based on physical pin mapping is adopted. The computing units (such as NPU and GPU) of each street lamp node are managed through physical pin mapping, and each computing power unit corresponds to a corresponding pin. When a node needs to call computing power resources, the system first calculates the real - time requirements and available resources of the current environment according to the event type and the geographical location of the node, and then selects a suitable computing power unit for allocation. In this way, the occupation and release of computing power units are controlled through physical signals, avoiding the context switching and resource contention problems existing in traditional virtualized resource pools.

[0047] Implementation steps: The computing power units of each node are bound to the corresponding physical pins, such as the NPU core to a specific pin number. When an event is detected, the system calculates the required resources based on the event priority and controls the pin status through physical signals to ensure the exclusivity of the target computing power unit. If multiple nodes request the same computing power unit, the system will dynamically adjust the resource allocation through a preset priority mechanism to give priority to high-priority tasks and ensure the real-time response of the system.

[0048] To address the problem that the traditional model deployment method cannot cope with complex environmental changes, this embodiment adopts a dynamic model loading mechanism based on environmental parameters. By real-time collecting environmental parameters such as light intensity, rain and fog concentration, and pedestrian flow density, these information are converted into binary codes and used as trigger signals to load the corresponding lightweight model plugins. For example, when a high rain and fog concentration is detected, the system automatically loads an infrared enhancement feature extraction plugin to handle video analysis tasks under low visibility conditions; when the light intensity is low, a low-light image restoration plugin is loaded to improve the image quality. Implementation path: Each node real-time collects environmental parameters through built-in sensors. Map environmental parameters such as light intensity, rain and fog concentration, and pedestrian flow density into binary codes. Each code corresponds to different environmental conditions. For example, the code for light intensity is divided into three levels (low light, normal light, strong light interference). According to the generated code, the system automatically loads the appropriate algorithm model. For example, when the rain and fog concentration is high, an infrared enhancement plugin is loaded; when the light is low, a low-light image restoration plugin is loaded. The model plugins are pre-stored in the cache of the edge chip, and the plugins are quickly loaded into the NPU instruction register through the DMA (Direct Memory Access) mechanism, and the loading process delay is controlled within 2 microseconds.

[0049] To ensure timely response when an event occurs, this embodiment optimizes the verification of computing power requests and the cross-node resource management method. When a node requests the computing power resources of a neighboring node, the system first verifies the request to confirm the legitimacy and urgency of the request. If the request conflicts with the current environmental parameters (such as the conflict between a traffic accident request and rainy and foggy weather), the system will judge whether to allow the response based on the set conflict probability. If the conflict probability is too high, the system will automatically select alternative computing power resources or postpone the request processing to ensure the accuracy and priority of the task are guaranteed. Step description: The master node sends out a computing power request signal, carrying event type and geographical coordinate information. After receiving the request, the neighboring node first verifies whether the geographical location of the master node conforms to the preset trusted topological relationship. If the event type in the request conflicts with the current environmental code (for example, a traffic accident conflicts with a rainy and foggy environment), the system evaluates whether to continue processing the request based on the conflict probability. If the request meets the conditions, the target node sets the pin status of the corresponding computing power unit to high level, and the master node obtains exclusive use rights; if the conditions are not met, an alternative pin number is returned or the response is delayed.

[0050] This embodiment introduces a dynamic resource scheduling mechanism based on event priority. In the case of multiple concurrent events, the system will ensure that high-priority tasks get priority use of computing resources according to the priority of the events. For low-priority tasks, the system will dynamically adjust their execution order according to the available resources to avoid the delay of low-priority tasks on high-priority tasks. In addition, the system also introduces an exception handling mechanism. When the communication between nodes is abnormal, the occupied resources are automatically released, and the abnormal events are recorded through logs to ensure that the system has self-healing capabilities and avoid system crashes due to partial failures. Optimization steps: define event priorities to ensure that emergency events such as traffic accidents and environmental monitoring are given priority. Dynamic priority calculation is performed on events to ensure that high-priority events are given priority when resources are tight. Set an abnormal threshold (such as 50ms communication delay). Once the threshold is reached, the resources are automatically released, the pin status is reset, and logs are recorded. Self-detect and repair the abnormal state of the system to avoid global crashes caused by node failures or resource conflicts.

[0051] In order to ensure the operability and scalability of the system in large-scale deployment, this embodiment describes in detail how to achieve distributed resource scheduling through edge computing technology to support large-scale smart street light networks. Through technical means such as physical pin mapping, environmental perception, and dynamic scheduling, the system can flexibly respond to different application scenario requirements while maintaining high efficiency and low latency performance. Implementation path: Generate a cellular topology network through geographic coordinates, and prioritize communication and resource calls between nodes based on Euclidean distance. Nodes within 50 meters will prioritize establishing direct communication links, and data will be forwarded through relay nodes if they are over 50 meters. Each node supports automatic detection and response to emergencies to ensure efficient operation of the system in complex environments. The system can adapt to environmental changes, optimize algorithms and resource allocation, so that the stability and reliability of the system can be guaranteed even in large-scale networks.

[0052] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limiting. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent street lamp data analysis system based on edge computing, characterized in that, Including: Node computing power unit pin mapping configuration: Map the computing power units of the edge computing chips of each street lamp node to physical pin numbers. The computing power units include, but are not limited to, NPU cores and GPU stream processors; Define pin states: High level indicates that the computing power unit is occupied, and low level indicates idle and callable; Event triggering and microservice chain dynamic construction: When any node detects a target event, generate a trigger signal including the event type and geographical coordinates; Based on the geographical coordinates and the preset adjacent node topology relationship, call the following hardware modules to construct a cross-node microservice chain: a. The camera module of adjacent node A, which is used to perform video enhancement analysis; b. The communication module of adjacent node B, which is used to generate traffic guidance instructions; c. The GPU computing power unit of adjacent node C, which is used to accelerate model inference; Pin-level computing power cross-node physical exclusive control: Target the specified pins of the target node through the bus signal, so that the master node exclusively occupies its corresponding computing power unit, including: a. The master node sends a computing power request signal to the adjacent node; b. After the adjacent node verifies the consistency between the geographical coordinates of the master node and the preset topology relationship, set the target pin to high level; c. During the occupation of the computing power unit, the local tasks of the adjacent node are prohibited from accessing the computing power unit corresponding to the set pin; Environment parameter-driven model adaptive loading: Real-time collect environment parameters, including but not limited to light intensity, rain and fog concentration, and pedestrian flow density; Generate a binary code according to the environment parameters to trigger the loading of preset lightweight model plugins, including: a. When the rain and fog concentration coding bit is 1, load the infrared enhancement feature extraction plugin; b. When the light intensity coding bit is 0, load the low illuminance image restoration plugin; Service chain task termination and resource release: When any of the following conditions is met, reset all the positioned pins to low level and disassemble the microservice chain: a. The target event is processed; b. The inter-node communication anomaly lasts for more than the preset threshold time; c. The environment parameter coding triggers a higher-priority event type.

2. The intelligent street lamp data analysis system based on edge computing according to claim 1, characterized in that, The preset method of the adjacent node topology relationship includes: Input the geographical coordinates of each node into a preset grid division algorithm to generate a honeycomb topology network; Define the communication priority of the nodes within the honeycomb unit as: When the Euclidean distance between two nodes satisfies d≤50 meters, establish a direct communication link preferentially; When 50 meters < d≤100 meters, signal relay is required through an intermediate node.

3. The intelligent street lamp data analysis system based on edge computing according to claim 1, characterized in that, The specific method for generating the binary code from the environment parameters includes: Divide the light intensity into three threshold intervals: low illuminance < 50 Lux, normal illuminance 50 - 1000 Lux, strong light interference > 1000 Lux, corresponding to the coding bits 00, 01, and 10 respectively; Correlate the rain and fog concentration with visibility, and define that the coding bit is 1 when the visibility < 30 meters, otherwise it is 0; The final coding format is [light coding][rain and fog coding][pedestrian flow density coding], where the pedestrian flow density coding is dynamically generated by real-time counting the number of pedestrians and the preset density threshold.

4. The edge computing-based intelligent street lamp data analysis system according to claim 1, wherein, The method for loading the model plug-in includes: pre-storing a plug-in library in the L2 cache of the edge chip, where each plug-in contains a header identifier and a functional code segment; when the environment encoding triggers plug-in loading, the specified plug-in is loaded from the cache to the NPU instruction register through direct memory access (DMA), and the loading latency is less than 2 μs.

5. The intelligent street lamp data analysis system based on edge computing according to claim 1, characterized in that, The method for verifying the computing power request signal includes: after receiving the request, the neighboring node performs the following verification steps: verifying whether the geographical coordinates of the master node are in the preset trusted node list; verifying the conflict between the event type in the request signal and the current environment encoding, and if the conflict probability > 30%, rejecting the response; checking whether its remaining computing power units meet the request requirements, and if not, returning an alternative pin number.

Images