Search

CN-121982921-A - Intelligent parking space guiding system based on edge calculation

CN121982921ACN 121982921 ACN121982921 ACN 121982921ACN-121982921-A

Abstract

The invention belongs to the technical field of intelligent guidance of parking, and particularly relates to an intelligent guidance system of a parking space based on edge calculation, which comprises a multi-source real-time data acquisition system, a distributed edge calculation architecture module and an intelligent decision core operation module; the system of the invention can help the driver to quickly find the parking space and optimize the management efficiency of the parking lot. The system realizes the qualitative change promotion of the operation efficiency of the parking lot through the full-chain optimization of perception-decision-execution. The invention realizes the transformation of parking lot operation from 'experience driving' to 'data intelligent driving' through full chain innovation of accurate perception-intelligent decision-humanized interaction, establishes a new industrial standard pole in three dimensions of economic benefit, user experience and social value, and provides a replicable example for building of smart city infrastructure.

Inventors

  • HAN WEI
  • HU TING
  • CHANG HAIBO
  • SONG XUANPEI
  • JU YAN
  • WANG XUANZHE

Assignees

  • 航天科工智能运筹与信息安全研究院(武汉)有限公司

Dates

Publication Date
20260505
Application Date
20251230

Claims (10)

  1. 1. The intelligent parking stall guiding system based on edge calculation is characterized by comprising a multi-source real-time data acquisition system, a distributed edge calculation architecture module and an intelligent decision core operation module; The multisource real-time data acquisition system is a basic layer of the parking space intelligent guide system, and high-precision and low-delay parking space state sensing and global environment modeling are realized through three modules of Internet of things sensor deployment, computer vision enhancement and auxiliary data source fusion to cooperatively work; The distributed edge computing architecture module is organically combined with cloud cooperative computing through a layered processing architecture, so that efficient processing and real-time response of mass data of a parking lot are realized, and the architecture fully utilizes cloud computing power to perform global optimization while guaranteeing low delay; the intelligent decision core operation module is a central center of the intelligent parking space guiding system, and is composed of three modules, namely a space-time prediction model, dynamic path planning and resource allocation optimization, and high-efficiency decision is realized through cooperation of multiple algorithms.
  2. 2. The intelligent guidance system for a parking space based on edge calculation according to claim 1, wherein the multi-source real-time data acquisition system comprises: (1) The sensor deployment module of the Internet of things is used for installing a low-power consumption geomagnetic sensor on the ground of a parking space, detecting the occupancy state of the parking space in real time; Based on geomagnetic sensing technology, detecting the occupation state of a parking space by utilizing disturbance characteristics of vehicles to the earth magnetic field; firstly, performing magnetic signal baseline calibration, namely continuously sampling for 60 seconds in an empty vehicle state, and calculating the magnetic field intensity average value ; The real-time detection logic is as follows: If it is Lasting for 5 seconds, and judging that the parking space is occupied; The sensor deployment module of the Internet of things can periodically self-check, and meanwhile, adjacent sensors can be cross-verified.
  3. 3. The intelligent guidance system for a parking space based on edge calculation according to claim 2, wherein the multi-source real-time data acquisition system further comprises: (2) The computer vision enhancement module utilizes the existing camera and lightweight YOLOv model of the parking lot, analyzes video stream in real time through edge computing equipment, detects the parking space state, tracks the vehicle track, and fuses the vehicle track with sensor data to improve accuracy; Firstly, a light YOLOv model is utilized to detect the state of a parking space, and the input resolution is reduced to The method comprises the steps of adopting a channel pruning technology, compressing the model volume from 73MB to 18MB, outputting a parking space ID and an occupied state thereof, fusing RelD characteristics based on DeepSORT algorithm to achieve cross-camera target association, adopting NTP protocol to perform time stamp alignment, setting a mapping relation between image pixel coordinates and physical coordinates by a checkerboard calibration method, and performing data synchronization, wherein the clock error of a sensor and a camera is less than 10 ms.
  4. 4. The intelligent guidance system for a parking space based on edge calculation according to claim 3, wherein the multi-source real-time data acquisition system further comprises: (3) The auxiliary data source fusion module is connected with the parking lot gate data, the payment system and the third party map API to construct multidimensional data input, integrates the multidimensional data inside and outside the parking lot, and improves the prediction capability of the system through space-time alignment and feature fusion, wherein the fusion method comprises the following steps: first, a unified space-time coordinate system is defined: , CAD model of parking lot For asynchronous data, linear interpolation is adopted to generate continuous time sequence, and then a space-time characteristic matrix is constructed: Wherein, the In order to make the number of time slices, Dimension features, multimodal fusion network employing attention mechanism: , Wherein, the , Is a global context vector; the fusion step mainly comprises the steps of firstly receiving multi-source heterogeneous data through a Kafka message queue, then cleaning the data, namely removing abnormal values, and performing space-time alignment again, namely mapping all the data to a unified space-time network, performing feature fusion again, generating a combined feature vector, finally outputting a state, and updating a global parking space state map; the multisource real-time data acquisition system realizes multisource complementation, improves detection accuracy, has end-to-end delay from data acquisition to state update of <350ms, supports thousand-parking-space-level deployment, and can expand capacity by linearly increasing edge nodes.
  5. 5. The edge computing-based intelligent guidance system of claim 4, wherein the distributed edge computing architecture module comprises: (1) Hierarchical processing architecture module Adopting an edge node-regional gateway-cloud computing architecture, layering and unloading a data processing task according to real-time requirements and computing complexity, and reducing end-to-end delay and bandwidth consumption, namely data nearby processing and computing load balancing, wherein edge computing nodes are deployed in every 10-20 parking spaces (real-time processing of the regional sensor/camera data, and response delay is less than 200ms; the layering process adopts the following steps: firstly, filtering data at an edge layer, removing sensor noise data such as transient magnetic field fluctuation, performing light reasoning, running a compression plate YOLOv model, outputting a parking space state and vehicle coordinates, performing an emergency obstacle avoidance instruction, triggering an audible and visual alarm when a reverse vehicle is detected, and then performing multi-node data fusion at a gateway layer: , Based on improvement A The method comprises the steps of generating an optimal path of a subarea by an algorithm, converting LoRaWAN data packets into an MQTT protocol mode, finally carrying out global state management of a cloud, integrating multi-gateway data to generate a parking lot digital twin body, updating STGNN prediction model parameters by using historical data, formulating dynamic pricing, equipment maintenance plans and the like according to a long-period strategy, wherein the processing delay of an edge node is less than 150ms, the local decision proportion is not needed to be intervened by the cloud in normal operation above 85%, the communication bandwidth is also saved, and the uplink data quantity is reduced by 72% compared with that of a traditional architecture.
  6. 6. The edge computing-based intelligent guidance system of claim 5, wherein the distributed edge computing architecture module further comprises: (2) Cloud collaborative computing module The balance of low-delay response and high-precision modeling is realized through the labor division mode of edge preprocessing and cloud deep analysis, and the federal learning framework is adopted to ensure the data privacy, the data after edge node preprocessing is uploaded to the cloud for global state analysis and large-scale and learning reasoning, the daily data processing amount can reach 10TB level, and the task unloading strategy is adopted when a collaborative mechanism is designed, namely: Wherein, the Representing the computational complexity (FLOPs), Representing the upstream bandwidth (Mbps), The task deadline is indicated as being the time at which the task is to be terminated, Representing local computing efficiency; Uploading metadata every 5 minutes, such as parking space state change statistics, so that the period synchronization and the event triggering synchronization are realized, and the metadata is immediately and fully uploaded when an abnormal event is detected; the AES-256 is adopted to encrypt the sensor data, the TLS1.3 ensures a communication link, the data are encrypted and transmitted, homomorphic encryption processing is carried out on license plate information of a user, the cloud only reserves hash values, privacy protection is carried out, and authority management and access control are carried out on the basis of RBAC of a blockchain; When a vehicle enters a guide, scene analysis is carried out, firstly, an edge layer responds, at the moment, an entrance camera identifies a license plate, an edge node A inquires reservation information in a local database, and if no reservation exists, a local D is called Secondly, uploading vehicle characteristics to the cloud by the node A, returning a possible target area of the vehicle by the cloud prediction engine, such as the vicinity of an elevator hoistway of an office building, and the like, dynamically adjusting again, triggering global re-planning at the moment when the regional gateway B detects sudden congestion of the target area, issuing a new path instruction to related edge nodes through the MQTT, and finally carrying out feedback optimization, uploading process data to the cloud by each node after the guidance is finished, and updating STGNN model parameters by a federal learning framework to improve the precision of next prediction; Performance optimization is performed by using a dynamic task allocation algorithm: When (when) In the aspect of communication protocol optimization, in the normal state in self-adaptive compression, the system adopts Protobuf binary coding, and in the emergency state, the system is switched into a lossless Zstandard compression mode, and in the aspect of energy source, the power consumption model is utilized: The CPU frequency energy saving rate is regulated in real time according to the load to reach 38 percent.
  7. 7. The intelligent guidance system for a parking space based on edge calculation according to claim 6, wherein the intelligent decision core operation module comprises: (1) Space-time prediction model A transducer+ Graph Neural Network mixed model is adopted to capture the space-time dependency relationship of parking space occupation, 50+ characteristic dimensions such as historical occupation data, real-time flow, peripheral events and the like are input, the parking space availability of each region in the future 30 minutes is predicted, the topological association among parking space nodes spatially has space dependency including adjacent parking spaces and regional parking spaces, the periodicity and the trend of the historical occupation mode temporally generate time dependency, and a space-time diagram is defined as follows: Graph structure Wherein the method comprises the steps of Node collection, wherein each parking space is a node; the method comprises the steps of collecting edges, and connecting parking spaces with the distance of less than 20 m; the adjacency matrix is weighted and reflects the influence intensity among the parking spaces; Spatiotemporal attention mechanism: For the space mask matrix, the local attention range is constrained, and the space-time convolution module is as follows: Wherein the method comprises the steps of For normalizing the adjacency matrix; capturing a time pattern for a time convolution network; the whole prediction process is that firstly, data preprocessing is carried out, and a feature matrix is input History T time slices, N parking spaces and D dimension characteristics, then carrying out Z-score normalization on each parking space, and then carrying out multi-mode fusion: External features Including weather, holiday marks, etc., and then performing joint training with a loss function of Wherein KL divergence term constraint prediction distribution And true distribution Is the consistency of (3); output is future Probability of occupation of each parking space of time slice ; The system updates model parameters in an increment per hour, adapts to dynamic changes, and compresses a teacher model to a student model by adopting a knowledge distillation technology.
  8. 8. The intelligent guidance system for a parking space based on edge calculation according to claim 7, wherein the intelligent decision core operation module further comprises: (2) Dynamic path planning module Based on an improved D Lite algorithm, the dynamic factors of the real-time parking space state, the vehicle running speed and the elevator/stair position are combined, the global optimal path is updated every 5 seconds, multi-objective optimization, namely the shortest path, the minimum turning and the low congestion path, the system fuses a potential field method and multi-objective optimization, real-time path planning under the dynamic environment is realized, incremental search is used, only the change area is updated, global re-planning is avoided, the path length, turning times and congestion degree are balanced, and the dynamic cost function is that Wherein the method comprises the steps of European distance; Steering angle penalty (units: radians); Congestion coefficients; the potential field function is designed as follows, rejecting potential fields: the attraction potential field (target parking space) is as follows: the direction of resultant force: ; the system adopts a B spline curve to fit the original path, meets the kinematic constraint of the vehicle, and can perform real-time collision detection: Wherein the method comprises the steps of In order for the vehicle to occupy an area, Is a collection of obstacles.
  9. 9. The intelligent guidance system for a parking space based on edge calculation according to claim 8, wherein the intelligent decision core operation module further comprises: (3) Resource allocation optimization module The intelligent parking space allocation is realized by adopting a combined auction algorithm, the parking space allocation is modeled as a multi-agent bidding problem, the global resource optimal allocation is realized, and the optimal parking space can be locked 15 minutes in advance for reserved vehicles; The auction model is shown below, with the valuation function of vehicle i for parking j being: Wherein the method comprises the steps of The distance from the current position of the vehicle to the parking space j is the distance; Scoring the convenience of the parking space j; Winner determination rules: s.t. Wherein the method comprises the steps of To assign variables; The special vehicle (such as an electric vehicle) is provided with the following weight lifting steps: Wherein the method comprises the steps of Is a priority gain coefficient; the distribution process comprises preprocessing the system, dividing the type of the parking space, reserving vehicles to lock the parking space 15 minutes in advance, and sequencing according to Priority when collision is encountered, namely when multiple vehicles bid on the same parking space, wherein the priority=0.6 Valuation +0.4 Reverse order of arrival time; The system can reserve elasticity, namely reserve 5% of parking space capacity for the VIP vehicle which arrives temporarily, and can also perform inverse-engine punishment, namely apply bidding cost punishment to the vehicle with frequent change targets.
  10. 10. The edge-calculation-based intelligent parking stall guiding system of claim 9, wherein the system belongs to the technical field of intelligent parking guiding.

Description

Intelligent parking space guiding system based on edge calculation Technical Field The invention belongs to the technical field of intelligent guidance of parking, and particularly relates to an intelligent guidance system of a parking space based on edge calculation. Background With the acceleration of the urbanization process and the rapid increase in the amount of car maintenance, parking difficulty has become a common challenge facing large cities worldwide. Traditional parking lot management relies on manual guidance or static indication systems, and has the problems of low efficiency, resource waste, poor user experience and the like. According to International Parking Institute (IPI) 2022 reports, drivers take on average 4-7 minutes to find a parking space in a large parking lot, resulting in increased congestion in the field, increased carbon emissions, and indirectly affecting peripheral traffic flow. Meanwhile, the control difficulty is high, the inner roads are staggered, the number of the fork is large, and a driver is difficult to quickly find the parking space. The parking management mode is not scientific enough, lacks intelligent guiding means, and the parking stall supply and demand contradiction is outstanding, and the parking stall disperses, and the overground parking stall disperses in office area, family member district, and building overall arrangement is complicated in addition, exists to find situations such as parking stall difficulty, can't the high-efficient limited parking stall of utilization, improves work efficiency. Disclosure of Invention First, the technical problem to be solved The invention aims to solve the technical problem of how to provide an intelligent parking space guiding system based on edge calculation. (II) technical scheme In order to solve the technical problems, the invention provides an intelligent parking stall guiding system based on edge calculation, which comprises a multi-source real-time data acquisition system, a distributed edge calculation architecture module and an intelligent decision core operation module; The multisource real-time data acquisition system is a basic layer of the parking space intelligent guide system, and high-precision and low-delay parking space state sensing and global environment modeling are realized through three modules of Internet of things sensor deployment, computer vision enhancement and auxiliary data source fusion to cooperatively work; The distributed edge computing architecture module is organically combined with cloud cooperative computing through a layered processing architecture, so that efficient processing and real-time response of mass data of a parking lot are realized, and the architecture fully utilizes cloud computing power to perform global optimization while guaranteeing low delay; the intelligent decision core operation module is a central center of the intelligent parking space guiding system, and is composed of three modules, namely a space-time prediction model, dynamic path planning and resource allocation optimization, and high-efficiency decision is realized through cooperation of multiple algorithms. (III) beneficial effects Compared with the prior art, the intelligent parking space guiding system based on the edge calculation is provided, becomes a key link of smart city construction, and has the core aim of realizing optimal allocation and path guiding of parking resources through real-time sensing, dynamic decision and multi-mode interaction. The invention provides a more convenient, quick and safe vehicle management mode, achieves intelligent guidance of parking, improves the passing efficiency of vehicles, solves the problem of difficult finding of parking spaces, and provides a fine and intelligent management means for vehicles. The system of the invention can help the driver to quickly find the parking space and optimize the management efficiency of the parking lot. The system realizes the qualitative change promotion of the operation efficiency of the parking lot through the full-chain optimization of perception-decision-execution. The invention realizes the transformation of parking lot operation from 'experience driving' to 'data intelligent driving' through full chain innovation of accurate perception-intelligent decision-humanized interaction, establishes a new industrial standard pole in three dimensions of economic benefit, user experience and social value, and provides a replicable example for building of smart city infrastructure. Drawings Fig. 1 is a flow chart of intelligent guidance of parking spaces in a parking lot. FIG. 2 is a workflow diagram of an intelligent decision ensemble. Detailed Description For the purposes of clarity, content, and advantages of the present invention, a detailed description of the embodiments of the present invention will be described in detail below with reference to the drawings and examples. In order to solve the problems in the prior art, th