Highway service area sanitation cleaning system based on vehicle and road cloud picture linkage

Abstract

The invention relates to the technical field of highway service area sanitation and cleaning, in particular to a highway service area sanitation and cleaning system based on vehicle-road cloud atlas linkage, a road end monitoring unit is deployed in a preset area of a service area, collects image data in real time, extracts scene change information through dynamic background modeling, and sends the scene change information to a server; establishing a three-dimensional mapping relation between the image and a world coordinate system based on binocular vision calibration; the cloud monitoring management platform comprises a garbage detection module, a positioning calculation module and a task scheduling module and is used for identifying and classifying garbage, acquiring actual physical coordinates of the garbage, generating a cleaning task and distributing the cleaning task to a cleaning vehicle; the multi-type automatic cleaning vehicle comprises a vehicle special for recoverable garbage and a vehicle special for unrecoverable garbage. Through cooperative linkage of the vehicle end, the road end, the cloud end and the graph end, full-link closed-loop management of detection, positioning, cleaning and verification is achieved, the service area sanitation cleaning task can be efficiently and accurately completed, the cleaning efficiency and the intelligent level are remarkably improved, and the system adapts to the complex and changeable service area environment.

Description

Technical Field

[0001] The present invention relates to the technical field of highway service area sanitation and cleaning, and in particular to a highway service area sanitation and cleaning system based on vehicle-road cloud map linkage. Background Art

[0002] With the increasing traffic volume on highways, the sanitation of service areas, as important places to ensure the rest and supply needs of drivers and passengers, has attracted more and more attention. The traditional sanitation and cleaning methods of highway service areas have the following limitations:

[0003] Manual cleaning relies on a large amount of manpower, which is inefficient and difficult to ensure timeliness; the service area is large and the garbage is widely distributed, so manual inspection and cleaning are difficult to achieve comprehensive coverage, and garbage is prone to being retained for a long time, affecting the overall environmental image of the service area and user experience;

[0004] Some existing automated cleaning attempts often have poor results due to immature technology. For example, some simple automatic cleaning devices lack accurate garbage detection and positioning capabilities. In complex service area environments, they cannot accurately identify various types of garbage, resulting in incomplete cleaning. At the same time, these devices lack effective response mechanisms when facing dynamically changing environments, such as heavy traffic and frequent movement of people, and are difficult to adapt to complex scenarios.

[0005] In terms of garbage detection, common target detection algorithms have insufficient accuracy in service area scenarios. There are many types of garbage in service areas, with different sizes and shapes, and the lighting conditions are complex and changeable. Traditional algorithms are difficult to accurately identify and classify all types of garbage. Moreover, traditional algorithms are usually based only on monocular vision and cannot accurately obtain the three-dimensional location information of garbage, which brings difficulties to the subsequent cleaning task scheduling.

[0006] In terms of positioning technology, binocular vision positioning alone is susceptible to environmental interference, and the positioning error is large in the case of occlusion, light changes, etc.; and relying solely on RTK positioning, the positioning effect is unstable in areas with poor signals, such as areas with severe building occlusion in the service area;

[0007] In addition, the current sanitation and cleaning system lacks effective coordination among the vehicle, road, cloud, and map ends; each link is independent of each other, and real-time sharing and interaction of data cannot be achieved, resulting in a lack of integrity and efficiency in the planning and execution of cleaning tasks.

[0008] To sum up, there is an urgent practical need to develop a sanitation and cleaning system that can adapt to the complex environment of highway service areas and has the capabilities of efficient garbage detection, precise positioning, intelligent task scheduling, and vehicle-road cloud map linkage and coordination. The present invention is developed based on this background. Summary of the invention

[0009] The purpose of the present invention is to provide a highway service area sanitation cleaning system based on vehicle-road cloud map linkage to solve the problems raised in the above background technology.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] A highway service area sanitation cleaning system based on vehicle-road cloud map linkage, comprising:

[0012] The roadside monitoring unit is deployed in the preset area of ​​the service area to collect image data of the monitoring area in real time, extract scene change information through dynamic background modeling, and establish a three-dimensional mapping relationship between the image coordinate system and the world coordinate system based on binocular vision calibration;

[0013] The cloud monitoring management platform communicates with the roadside monitoring unit and includes:

[0014] The garbage detection module uses an improved target detection algorithm to identify and classify garbage in image data;

[0015] The positioning calculation module converts the detected garbage pixel coordinates into actual physical coordinates based on binocular stereo vision and RTK collaborative positioning technology;

[0016] The task scheduling module generates cleaning tasks based on a multi-factor decision model and dynamically allocates them to cleaning vehicles;

[0017] Various types of automatic cleaning vehicles, including special cleaning vehicles for recyclable and non-recyclable garbage. Each cleaning vehicle is equipped with:

[0018] High-precision positioning module, achieving centimeter-level positioning;

[0019] Semantic map navigation module, dynamic path planning based on incrementally updated maps;

[0020] Classification and cleaning device, which can be adapted to the physical characteristics of different types of garbage for sorting or compression;

[0021] Closed-loop verification module verifies the cleaning effect through deep feature comparison.

[0022] As a preferred solution, the improved target detection algorithm is the YOLOv11 improved model, including:

[0023] Based on the size distribution characteristics of garbage in the service area, the K-means++ clustering algorithm is used to optimize the anchor frame size. The anchor frame size is the logarithmic mean of the width and height of the annotation frame. The specific calculation method is: after taking the natural logarithm of the width and height of the annotation frame, the arithmetic mean is calculated as the anchor frame size;

[0024] A binocular vision feature fusion mechanism is introduced into the Backbone network. The left and right eye feature maps are fused through a three-dimensional convolution operation. After the spatial attention weight is generated, it is multiplied element by element with the left eye feature map, and the fused feature map is output.

[0025] The loss function is defined as the weighted sum of the improved Inner-IoU loss function and the RTK positioning coordinate regression loss, where the weight factor of the RTK regression loss is 0.8.

[0026] As a preferred solution, the coordinate conversion method of the positioning calculation module includes:

[0027] The camera internal parameter matrix and external parameter matrix are obtained through binocular positioning, where the camera internal parameter matrix includes the focal length , and the optical center coordinates , the camera external parameter matrix includes the rotation matrix and translation vector , the target three-dimensional coordinates are calculated based on the difference in the horizontal coordinates of the left and right eye images of the target point:

[0028] Depth value It is equal to the product of the equivalent focal length and the binocular baseline distance divided by the parallax value;

[0029] Target actual horizontal coordinate It is equal to the depth value multiplied by the difference between the horizontal coordinate of the left eye image and the horizontal coordinate of the optical center, and then divided by the left eye focal length;

[0030] Target actual vertical coordinate It is equal to the depth value multiplied by the difference between the ordinate of the left eye image and the ordinate of the optical center, and then divided by the left eye focal length;

[0031] When the binocular positioning depth error exceeds 0.1 meters, the coordinates are corrected based on the RTK positioning data. The correction weight is dynamically calculated based on the RTK signal quality. The higher the signal quality, the greater the weight of the RTK data.

[0032] As a preferred solution, in the path planning algorithm of the task scheduling module, the path cost function integrates the following factors:

[0033] The actual length of the path g(n) and the estimated remaining length h(n);

[0034] The priority weight of garbage type, the priority of flammable garbage is increased by 50%;

[0035] Real-time obstacle density factor. The higher the obstacle density, the greater the path cost.

[0036] RTK positioning confidence factor. The lower the confidence, the higher the path cost.

[0037] The weight coefficients of garbage priority, obstacle density and RTK confidence in the cost function are 0.6, 0.3 and 0.1 respectively.

[0038] As a preferred solution, the closed-loop verification module is implemented by the following steps:

[0039] Extract multi-scale features of the images before and after cleaning, perform channel concatenation of the ResNet-50 features of the cleaned image and the coordinate-aware convolutional features of the pre-cleaned image to generate fused features;

[0040] The verification threshold is dynamically adjusted according to the ambient light intensity. The base threshold is 0.8. For every increase of 1 times the base light intensity, the threshold is increased by 0.2 times.

[0041] If the verification fails, task reallocation, semantic map correction and binocular vision calibration parameter adjustment will be triggered. The calibration parameter correction amount is proportional to the difference in fusion features.

[0042] As a preferred solution, the implementation of the high-precision positioning module includes:

[0043] The carrier phase differential technology is used to achieve RTK centimeter-level positioning, with a positioning error of less than 0.05 meters;

[0044] When the RTK signal is lost, it automatically switches to the binocular vision positioning mode. The switching condition is that the deviation between the binocular positioning data of three consecutive frames and the historical trajectory exceeds the threshold.

[0045] As a preferred solution, the working modes of the classification cleaning device include:

[0046] When sorting recyclable garbage, the magnetic strength of the magnetic attraction mechanism is proportional to the target volume, and the proportionality coefficient is 0.5;

[0047] When compressing non-recyclable garbage, the vacuum adsorption negative pressure value is the smaller value between 0.8 times the garbage density value and 20kPa, and the compressed volume is less than 1 / 5 of the original volume.

[0048] As a preferred solution, the incremental update method of the semantic map includes:

[0049] The grayscale difference between the current image and the reference map is dynamically compared through binocular vision to generate a difference mask. The area with a grayscale difference of more than 30 is marked as a changed area.

[0050] After reconstructing the three-dimensional point cloud of the changed area, the updated garbage distribution, obstacle location and charging pile status are integrated into the semantic map. The fusion weight of the new and old maps is 0.7, and a global consistency check is performed once an hour.

[0051] As a preferred solution, it also includes an exception handling mechanism. When the number of cleaning task failures exceeds the preset threshold, the manual intervention protocol is initiated and the fault data is uploaded to the cloud analysis platform, triggering equipment self-inspection and parameter optimization.

[0052] As a preferred solution, the update frequency of dynamic background modeling is positively correlated with the traffic volume. For every increase of 50 vehicles / minute in traffic volume, the update frequency increases by 1 times, and the maximum update delay does not exceed 2 seconds.

[0053] It can be seen from the technical solution provided by the present invention that the highway service area sanitation cleaning system based on vehicle-road cloud map linkage provided by the present invention has the following beneficial effects:

[0054] Efficient and accurate garbage cleaning: Through the real-time collection of image data by the road-side monitoring unit, the improved target detection algorithm, such as the YOLOv11 improved model, can accurately identify and classify garbage; combined with binocular stereo vision and RTK collaborative positioning technology, the actual physical coordinates of the garbage can be accurately obtained, and then the cleaning tasks are generated with the help of a multi-factor decision model and dynamically assigned to cleaning vehicles, which greatly improves the efficiency and accuracy of garbage cleaning and ensures a clean service area environment;

[0055] High degree of intelligence and automation: Multiple types of automatic cleaning vehicles are equipped with high-precision positioning modules, semantic map navigation modules, classification cleaning devices and closed-loop verification modules, which can automatically complete the entire process from locating garbage, planning routes, executing cleaning to verifying results, reducing manual intervention, improving the intelligence and automation level of cleaning work, and reducing labor costs;

[0056] Flexible adaptation to complex environments: The semantic map adopts an incremental update method to promptly reflect changes in the service area environment; the dynamic background modeling update frequency is positively correlated with the traffic volume and the maximum delay does not exceed 2 seconds, so that the system can flexibly adapt to the dynamically changing environment in the service area and ensure that garbage detection and positioning are not excessively interfered by environmental factors;

[0057] Reliable exception handling: The exception handling mechanism monitors all aspects of the system in real time. Once it detects a cleaning task failure, hardware equipment failure, or communication link abnormality, it can automatically try to recover, issue an alarm, and initiate manual intervention. At the same time, it records and uploads data to ensure that the system can still maintain basic operation and recover in time when facing an abnormality, thereby enhancing system reliability.

[0058] Optimal resource allocation: The task scheduling module considers a variety of factors to generate cleaning tasks and plan routes, and rationally allocates cleaning vehicles, achieving the optimal allocation of cleaning resources in time and space, improving resource utilization efficiency, and reducing operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1The present invention is a schematic diagram of the overall structure of a highway service area sanitation cleaning system based on vehicle-road cloud map linkage. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0062] like Figure 1 As shown, an embodiment of the present invention provides a highway service area sanitation cleaning system based on vehicle-road cloud map linkage, including a road-side monitoring unit, a cloud monitoring management platform and multiple types of automatic cleaning vehicles;

[0063] In the highway service area sanitation and cleaning system of the present invention, the vehicle side, road side, cloud side, and map side work closely together to build a full-link closed-loop management system of detection, positioning, cleaning, and verification, ensuring that the sanitation and cleaning work of the service area is carried out efficiently, accurately, and continuously optimized;

[0064] As the frontier of information collection, the roadside monitoring unit plays a key role. It is deployed in the preset area of ​​the service area and uses advanced monitoring equipment to collect image data of the monitoring area in real time. Through dynamic background modeling technology, it can keenly extract scene change information and promptly detect the appearance of foreign objects such as garbage. At the same time, based on binocular vision calibration, it accurately establishes a three-dimensional mapping relationship between the image coordinate system and the world coordinate system, providing a solid data foundation for subsequent garbage detection and positioning. These collected and processed data are quickly transmitted to the cloud through a stable communication link.

[0065] The cloud is the core brain of the entire system, and the cloud monitoring management platform plays a variety of important functions here; the garbage detection module receives image data from the road side, and uses improved target detection algorithms, such as the YOLOv11 improved model, to accurately identify and classify garbage; the positioning calculation module converts the detected garbage pixel coordinates into actual physical coordinates based on binocular stereo vision and RTK collaborative positioning technology, providing accurate location information for cleaning tasks; the task scheduling module comprehensively considers multiple factors such as garbage location, type, cleaning vehicle status, and service area road conditions, generates scientific and reasonable cleaning tasks through a multi-factor decision model, and dynamically and efficiently allocates them to the most suitable cleaning vehicles; in addition, the cloud also receives information such as the execution of cleaning tasks and equipment status uploaded by the vehicle side, as well as map data updated by the map side, for unified analysis and management;

[0066] The vehicle side refers to various types of automatic cleaning vehicles, which are the direct executors of sanitation and cleaning tasks; the cleaning vehicles for recyclable and non-recyclable garbage are equipped with high-precision positioning modules, which can achieve centimeter-level positioning and accurately know their own positions in any complex environment; the semantic map navigation module plans a dynamic and optimal driving path for the vehicle based on the incrementally updated map, and guides the vehicle to quickly reach the garbage location; the classification cleaning device is based on the physical characteristics of the garbage type, such as the magnetic suction mechanism adaptively adjusts the magnetic force to sort recyclable garbage according to the target volume, and the vacuum adsorption device adjusts the negative pressure value according to the garbage density to compress non-recyclable garbage, so as to efficiently complete the cleaning work; the closed-loop verification module verifies the cleaning effect through deep feature comparison and feeds back the verification results to the cloud;

[0067] The map end is mainly responsible for the management and update of semantic maps. The incremental update method of semantic maps generates difference masks through binocular vision to identify the areas with environmental changes in the service area, and then merges the updated semantic layer with the original map. A global consistency check is also performed every hour to ensure that the map is real-time and accurate. The updated map data is transmitted to the cloud and vehicle end in a timely manner to provide the latest environmental information for vehicle navigation and mission planning.

[0068] When the system is started, the road side first collects data and transmits it to the cloud; the cloud side completes garbage detection and positioning, generates cleaning tasks and assigns them to the vehicle side; the vehicle-side cleaning vehicle, based on the task instructions, uses the positioning and navigation modules to go to the garbage location, uses the classification cleaning device to clean it, and then confirms the cleaning effect through the closed-loop verification module and feeds back to the cloud; at the same time, the map side updates the map in real time and provides it to other sides; throughout the process, the vehicle side, road side, cloud side, and map side continuously interact with data, forming a cyclical and continuously optimized full-link closed-loop management process of detection, positioning, cleaning and verification, thereby ensuring that the highway service area always maintains a good sanitary environment.

[0069] In this embodiment, the roadside monitoring unit is deployed in a preset area of ​​the service area to collect image data of the monitoring area in real time, extract scene change information through dynamic background modeling, and establish a stereo mapping relationship between the image coordinate system and the world coordinate system based on binocular vision calibration;

[0070] Furthermore, the roadside monitoring unit is an important component of the highway service area sanitation and cleaning system based on the vehicle-road cloud map linkage. Its main function is to be deployed in the preset area of ​​the service area, collect image data of the monitoring area in real time, extract scene change information through dynamic background modeling, and establish a three-dimensional mapping relationship between the image coordinate system and the world coordinate system based on binocular vision calibration, so as to provide basic data for subsequent garbage detection, positioning and other operations; the following is a detailed and specific description of the unit:

[0071] Overall function overview:

[0072] The roadside monitoring unit collects image data of the preset area of ​​the service area in real time, uses dynamic background modeling technology to capture the change information in the scene, and uses binocular vision calibration technology to establish a three-dimensional mapping relationship between the image coordinate system and the world coordinate system. These operations can accurately identify the distribution of garbage in the service area and its location in the real world, providing accurate data support for garbage detection and positioning calculations of the cloud monitoring management platform, thereby achieving efficient operation of the entire sanitation system;

[0073] Data collection:

[0074] One of the core functions of the roadside monitoring unit is to collect image data, as follows:

[0075] Image data collection: Deploy monitoring equipment in the preset area of ​​the service area to collect image information of the area in real time; these image data contain various scene information in the service area, which is the basis for subsequent garbage identification and positioning;

[0076] Dynamic background modeling:

[0077] The roadside monitoring unit uses dynamic background modeling technology to extract scene change information. Its specific principles and functions are as follows:

[0078] Background modeling principle: By analyzing continuous image data, a background model of the service area scene is established; the model will be dynamically updated as time goes by and the scene changes to adapt to different environmental conditions;

[0079] Change information extraction: Compare the real-time collected image data with the background model to extract the change information in the scene; this change information may include the appearance of garbage, the movement of vehicles, etc., providing important clues for subsequent garbage detection;

[0080] Binocular vision calibration:

[0081] The roadside monitoring unit uses binocular vision calibration technology to establish a three-dimensional mapping relationship between the image coordinate system and the world coordinate system. The specific steps are as follows:

[0082] Obtaining internal and external parameters: Obtain the camera's internal and external parameter matrices through binocular positioning. The camera's internal parameter matrix includes the focal length. , and the optical center coordinates , the camera external parameter matrix includes the rotation matrix and translation vector ; The intrinsic parameter matrix reflects the internal characteristics of the camera, such as focal length, principal point position, etc.; the extrinsic parameter matrix describes the position and posture of the camera in the world coordinate system;

[0083] Establishment of stereo mapping relationship: Based on the acquired intrinsic and extrinsic matrix, combined with the principle of binocular vision, a stereo mapping relationship between the image coordinate system and the world coordinate system is established; this allows the coordinates of garbage pixels detected in the image to be converted into actual physical coordinates, providing accurate location information for subsequent cleaning task scheduling;

[0084] Data output:

[0085] The roadside monitoring unit transmits the collected image data, extracted scene change information, established stereo mapping relationship and other data to the cloud monitoring management platform. The specific contents are as follows:

[0086] Image data transmission: The real-time collected image data is sent to the garbage detection module of the cloud monitoring management platform for garbage identification and classification;

[0087] Transmission of change information and mapping relationship: The extracted scene change information and the established stereo mapping relationship data are transmitted to the positioning calculation module so as to convert the detected garbage pixel coordinates into actual physical coordinates;

[0088] Application value of the module:

[0089] The road-side monitoring unit provides accurate basic data for the sanitation and cleaning system of highway service areas by collecting image data in real time, extracting scene change information and establishing three-dimensional mapping relationships. This data is the key basis for subsequent garbage detection, positioning and cleaning task scheduling, which helps to improve the efficiency and accuracy of the entire sanitation and cleaning system and ensure that the sanitary environment of the service area is effectively maintained. At the same time, the application of dynamic background modeling and binocular vision calibration technology enables the unit to adapt to different environmental conditions and scene changes, and has strong flexibility and adaptability.

[0090] In this embodiment, the cloud monitoring management platform is in communication connection with the road-side monitoring unit, and includes a garbage detection module, a positioning calculation module, and a task scheduling module;

[0091] Furthermore, the garbage detection module uses an improved target detection algorithm to identify and classify garbage in image data; the garbage detection module is an important part of the cloud monitoring management platform in the highway service area sanitation cleaning system based on the vehicle-road cloud map linkage. Its main function is to use an improved target detection algorithm to identify and classify garbage in the image data collected by the road-side monitoring unit, providing a basis for subsequent positioning and cleaning tasks; the following is a detailed and specific description of the module:

[0092] Overall function overview:

[0093] The garbage detection module receives the image data transmitted by the roadside monitoring unit, uses the improved target detection algorithm to accurately identify the garbage in the image, and classifies it into recyclable garbage and non-recyclable garbage. This module improves the accuracy and efficiency of garbage detection by optimizing the algorithm and introducing specific mechanisms, providing key support for the efficient operation of the entire sanitation and cleaning system.

[0094] Data input:

[0095] The operation of the garbage detection module depends on the image data provided by the roadside monitoring unit, as follows:

[0096] Image data: The roadside monitoring unit collects image information of the preset area of ​​the service area in real time and transmits it to the garbage detection module; these image data contain various scene information in the service area and are the basis for garbage identification and classification;

[0097] Improved object detection algorithm:

[0098] The garbage detection module uses the improved YOLOv11 model to identify and classify garbage. The specific improvements are as follows:

[0099] Anchor box size optimization:

[0100] Based on the garbage size distribution characteristics of the service area, the K-means++ clustering algorithm is used to optimize the anchor box size. The anchor box size calculation satisfies the formula: (in, is the number of labeled boxes within the cluster, is the width of the annotation box, is the height of the annotation box). In this way, the size of the anchor box can be more closely matched to the actual size of the garbage in the service area, thus improving the accuracy of target detection.

[0101] Binocular vision feature fusion mechanism:

[0102] In the Backbone network, a binocular visual feature fusion mechanism is introduced to generate spatial attention weights, and the calculation satisfies the formula: (in, It is the feature map output after fusing the visual features of the left and right eyes. is the activation function, is a three-dimensional convolution operation, Input feature map for the left eye, Input feature map for the right eye); This mechanism can make full use of binocular vision information, enhance the feature extraction capability of garbage targets, and improve detection accuracy;

[0103] Loss function definition:

[0104] The loss function is defined as the weighted sum of the improved Inner-IoU loss and the RTK positioning coordinate regression loss, and the weight factor ; By reasonably designing the loss function, the model can be better optimized during the training process and the performance of the model can be improved;

[0105] Garbage identification and classification:

[0106] The garbage detection module uses an improved target detection algorithm to identify and classify garbage in image data. The specific process is as follows:

[0107] Feature extraction: Extract features from the input image data and extract the feature information of junk targets through the improved network structure in the target detection algorithm;

[0108] Target recognition: Based on the extracted feature information, identify the garbage targets in the image and determine their locations and bounding boxes;

[0109] Garbage classification: Classify the identified garbage targets into recyclable garbage and non-recyclable garbage, providing a classification basis for subsequent cleaning tasks;

[0110] Data output:

[0111] The spam detection module outputs the identified and classified spam information to other modules of the cloud monitoring management platform. The specific contents are as follows:

[0112] Garbage location and classification information: The location information of the detected garbage in the image and the classification results (recyclable or non-recyclable) are transmitted to the positioning calculation module for coordinate conversion and cleaning task scheduling;

[0113] Application value of the module:

[0114] The garbage detection module improves the accuracy and efficiency of garbage identification and classification in highway service areas through an improved target detection algorithm. Accurate garbage detection results provide a reliable basis for subsequent positioning and cleaning tasks, help optimize the allocation of cleaning resources, improve the overall operating efficiency of the sanitation and cleaning system, and ensure that the sanitary environment of the service area is effectively maintained. At the same time, the improved algorithm and mechanism of this module enable it to adapt to different environmental conditions and garbage types, and has strong versatility and adaptability.

[0115] Furthermore, the positioning calculation module converts the detected garbage pixel coordinates into actual physical coordinates based on binocular stereo vision and RTK collaborative positioning technology; the positioning calculation module is a key part of the cloud monitoring management platform in the highway service area sanitation cleaning system based on the vehicle-road cloud map linkage. Its main function is to convert the garbage pixel coordinates detected by the garbage detection module into actual physical coordinates based on binocular stereo vision and RTK collaborative positioning technology, providing accurate location information for the scheduling of subsequent cleaning tasks; the following is a detailed description of the module:

[0116] Overall function overview:

[0117] The positioning calculation module receives the garbage pixel coordinates output by the garbage detection module, and uses binocular stereo vision and RTK collaborative positioning technology to accurately calculate the actual physical coordinates of the garbage in the real world. During the calculation process, the characteristics and advantages of binocular vision positioning and RTK positioning are comprehensively considered, and the accuracy and reliability of the positioning results are ensured through operations such as coordinate conversion and error correction, providing strong support for cleaning vehicles to efficiently perform cleaning tasks;

[0118] Data input:

[0119] The operation of the positioning calculation module depends on the data provided by multiple modules, as follows:

[0120] Garbage pixel coordinates: provided by the garbage detection module. This coordinate is the garbage location information detected in the image coordinate system and is the starting data for positioning calculation;

[0121] Binary target calibration parameters: including internal parameter matrix ( ) and the external parameter matrix ,These parameters are obtained through the dual target positioning of the road end monitoring unit and are used to establish the connection between the image coordinate system and the world coordinate system;

[0122] RTK data: The RTK positioning data provided by the high-precision positioning module is used to correct coordinates when the binocular positioning error is large to improve positioning accuracy;

[0123] Coordinate transformation method:

[0124] Binocular vision positioning:

[0125] The internal parameter matrix and external parameter matrix obtained by binocular positioning are used to calculate the target 3D coordinates based on the disparity. The calculation formula is as follows: , , (in, is the binocular baseline distance, that is, the distance between the optical centers of the left and right cameras; is the disparity value, i.e., the horizontal pixel difference between corresponding points of the same object in the left and right images; is the focal length of the camera; is the horizontal pixel coordinate of the target point in the left image; The vertical pixel coordinates of the target point in the target image; and The camera is and focal length of direction; and are the coordinates of the origin of the image coordinate system in the pixel coordinate system respectively); using these formulas, the pixel coordinates of the garbage can be converted into preliminary physical coordinates in the three-dimensional space;

[0126] RTK collaborative positioning and error correction:

[0127] When the binocular positioning error exceeds the threshold Meters, coordinate correction is performed through RTK data; correction weight The RTK confidence is dynamically calculated to ensure that the results of binocular vision positioning and RTK positioning can be reasonably integrated under different positioning accuracy conditions to improve the accuracy of the final positioning;

[0128] Data output:

[0129] The positioning calculation module outputs the converted actual physical coordinates of the garbage to the task scheduling module of the cloud monitoring management platform. The specific contents are as follows:

[0130] Actual physical coordinates of garbage: Accurate actual physical coordinates of garbage provide accurate location basis for the task scheduling module to plan cleaning tasks and allocate cleaning vehicles, so that cleaning vehicles can quickly and accurately reach the garbage location for cleaning operations;

[0131] Application value of the module:

[0132] The positioning calculation module realizes the precise conversion of garbage pixel coordinates to actual physical coordinates through binocular stereo vision and RTK collaborative positioning technology. The accurate positioning results provide a key guarantee for the efficient operation of the sanitation cleaning system, enabling cleaning vehicles to quickly reach the garbage location, improve cleaning efficiency, and reduce the residence time of garbage in the service area, thereby improving the overall sanitation environment quality of the service area. At the same time, the error correction mechanism ensures that reliable positioning results can be obtained under different environments and positioning conditions, enhancing the stability and adaptability of the system.

[0133] Furthermore, the task scheduling module generates cleaning tasks based on a multi-factor decision model and dynamically assigns them to cleaning vehicles. The task scheduling module is the core part of the cloud monitoring management platform in the highway service area sanitation and cleaning system based on the vehicle-road cloud map linkage. Its main responsibility is to generate cleaning tasks based on a multi-factor decision model and dynamically assign these tasks to cleaning vehicles to ensure that the entire sanitation and cleaning process is carried out efficiently and orderly. The following is a detailed description of this module:

[0134] Overall function overview:

[0135] The task scheduling module comprehensively analyzes the garbage information identified by the garbage detection module, the actual physical coordinates of the garbage obtained by the positioning calculation module, and the real-time status of the cleaning vehicles, etc., and uses a multi-factor decision-making model to generate the most optimized cleaning task plan and dynamically and reasonably allocate tasks to each cleaning vehicle. By continuously optimizing task allocation, the utilization efficiency of cleaning resources is improved, the garbage cleaning time is shortened, and the sanitation of the service area is ensured to always remain in good condition.

[0136] Data input:

[0137] The task scheduling module relies on a wide range of data sources, as follows:

[0138] Garbage location and classification information: obtained from the garbage detection and positioning calculation module; the specific location coordinates of the garbage in the service area and the type of garbage (recyclable or non-recyclable) are clarified. This is the basic data for generating cleaning tasks and determines where the cleaning vehicle needs to go and the cleaning method to be adopted;

[0139] Cleaning vehicle status information: including the vehicle's current location, remaining power or fuel, and the operating status of the vehicle's cleaning equipment; real-time understanding of the cleaning vehicle status helps to reasonably allocate tasks, avoid assigning tasks to vehicles that are not working properly or are too far away, and improve the feasibility and efficiency of task execution;

[0140] Service area map and traffic information: covers the layout map of the service area, including location information such as roads, buildings, parking lots, and real-time traffic data, such as congested sections and construction areas. This information is used to plan the driving routes of cleaning vehicles, avoid congested sections, reduce driving time, and ensure efficient completion of cleaning tasks;

[0141] Multi-factor decision model and task generation:

[0142] The task scheduling module uses a multi-factor decision model to generate cleaning tasks. The specific contents are as follows:

[0143] Multi-factor consideration: Comprehensively consider multiple factors such as the priority of garbage, the distance between the cleaning vehicle and the garbage, the current status of the vehicle, and road conditions. For example, hazardous garbage such as flammable garbage is given a higher priority and cleaning vehicles are arranged to clean it up first. Cleaning vehicles that are closer and in good condition are given priority in assigning tasks to reduce the total time for task execution.

[0144] Task generation logic: Based on a comprehensive evaluation of multiple factors, determine the order in which each piece of garbage needs to be cleaned and the corresponding cleaning vehicle; the generated cleaning task contains detailed information such as garbage location, garbage type, estimated cleaning time, assigned cleaning vehicle number, etc.

[0145] Path planning algorithm:

[0146] When assigning tasks, the task scheduling module needs to plan the driving path for the cleaning vehicle. The path cost function is defined as: (in, , , ; From the starting point to the node The actual cost of the vehicle, such as the distance and time traveled by the vehicle; For slave nodes The estimated cost to reach the destination; For Node The priority of garbage disposal, which is calculated as the weighted sum of the garbage flammability weight and the RTK confidence; For Node Traffic factors such as congestion level; For Node The confidence of RTK positioning); through the path cost function, a variety of factors are comprehensively considered to plan the optimal driving path for the cleaning vehicle, reduce the task execution cost, and improve the cleaning efficiency;

[0147] Task allocation and dynamic adjustment:

[0148] Initial task allocation: According to the cleaning task plan generated by the multi-factor decision model, the task is allocated to the most suitable cleaning vehicle; in the allocation process, the matching degree between the vehicle and the task is fully considered. For example, the recyclable garbage task is allocated to the recyclable garbage special cleaning vehicle equipped with corresponding sorting equipment;

[0149] Dynamic adjustment mechanism: During the execution of the cleaning task, if there is an emergency, such as vehicle failure, new garbage being found, etc., the task scheduling module can monitor and re-evaluate the task in real time; dynamically adjust the task allocation according to the new situation, and promptly reallocate the task of the faulty vehicle to other available vehicles to ensure that the cleaning task is not affected and continues to be carried out efficiently;

[0150] Data output:

[0151] The task scheduling module outputs the generated cleaning tasks and corresponding allocation plans to the cleaning vehicles. The specific contents are as follows:

[0152] Cleaning task instructions: contain detailed information of the task, such as garbage location coordinates, garbage type, cleaning requirements, planned driving path, etc., which are sent to the corresponding cleaning vehicles to guide the cleaning vehicles to accurately perform the cleaning tasks;

[0153] Application value of the module:

[0154] The task scheduling module achieves efficient generation and dynamic allocation of cleaning tasks through a scientific multi-factor decision-making model and a reasonable path planning algorithm; this greatly improves the overall operating efficiency of the highway service area sanitation cleaning system, optimizes the allocation of cleaning resources, reduces garbage cleaning time, and improves the sanitary environment quality of the service area; at the same time, the dynamic adjustment mechanism enables the system to have the ability to respond to emergencies, enhances the stability and reliability of the system, and provides strong guarantees for the normal operation of the service area.

[0155] In this embodiment, the multiple types of automatic cleaning vehicles include cleaning vehicles for recyclable and non-recyclable garbage, and each cleaning vehicle is equipped with:

[0156] High-precision positioning module, achieving centimeter-level positioning;

[0157] Semantic map navigation module, dynamic path planning based on incrementally updated maps;

[0158] Classification and cleaning device, which can be adapted to the physical characteristics of different types of garbage for sorting or compression;

[0159] Closed-loop verification module verifies the cleaning effect through deep feature comparison;

[0160] Furthermore, the multi-type automatic cleaning vehicle is the key execution unit of the highway service area sanitation cleaning system based on the vehicle-road cloud map linkage. Its main function is to efficiently clean up different types of garbage in the service area according to the cleaning tasks assigned by the cloud monitoring management platform to ensure a clean service area environment; the following is a detailed description:

[0161] Overall function overview:

[0162] Multi-type automatic cleaning vehicles are divided into special cleaning vehicles for recyclable and non-recyclable garbage. They are equipped with a variety of advanced modules and devices, which can accurately locate the garbage location, plan the optimal cleaning path, and implement targeted cleaning operations according to the garbage type, while verifying the cleaning effect, forming a complete cleaning closed-loop process;

[0163] Vehicle configuration and functional modules:

[0164] High-precision positioning module:

[0165] Application of positioning technology: Carrier phase differential technology is used to achieve RTK centimeter-level positioning, which can reduce the positioning error to less than 0.05 meters. This technology accurately calculates the real-time position of the vehicle in the service area by receiving satellite signals and differential signals from base stations.

[0166] Positioning mode switching: When the RTK signal is lost, it automatically switches to the binocular vision positioning mode. The switching condition is that three consecutive frames of positioning data are out of tolerance. Using the binocular vision principle similar to that of the roadside monitoring unit, the binocular camera on the vehicle is used to obtain image information, calculate the vehicle position, ensure the continuity and accuracy of positioning, and provide stable position support for the execution of cleaning tasks.

[0167] Semantic map navigation module:

[0168] Map construction and updating: Navigation based on incrementally updated semantic maps; Generate difference masks through binocular vision (in, is the image at the current moment, is the reference image, To take the absolute value operation, is the difference mask, ), identify environmental changes within the service area;

[0169] Then the updated semantic layers are fused, and the fusion formula is: (in, is the updated semantic map, is the semantic map before updating, For the updated semantic layer, is the fusion coefficient), and triggers a global consistency check every hour to ensure the map is real-time and accurate;

[0170] Path planning and execution: Plan dynamic paths based on the tasks assigned by the cloud monitoring management platform and the semantic map updated in real time; comprehensively consider the road conditions in the service area, the dynamic information of other vehicles and pedestrians, etc., plan an efficient and safe driving path for the vehicle, and guide the vehicle to quickly reach the garbage location;

[0171] Classification cleaning device:

[0172] Recyclable waste sorting: For recyclable waste, it is equipped with magnetic suction mechanism and other sorting devices; the magnetic strength of the magnetic suction mechanism is Adaptive adjustment (where is the magnetic strength of the magnetic attraction mechanism, the target volume is the volume of the recyclable garbage target, ), can automatically adjust the magnetic force according to the different volumes of recyclable garbage, accurately absorb and sort out ferromagnetic recyclables, such as cans;

[0173] Compression of non-recyclable garbage: For non-recyclable garbage, a vacuum adsorption device is used for compression. The negative pressure value of the vacuum adsorption is calculated according to Adjustment (where is the vacuum adsorption negative pressure value, is the density of non-recyclable garbage), compression ratio ≥5:1; by reasonably adjusting the negative pressure value, the non-recyclable garbage can be efficiently compressed to reduce the volume of garbage, which is convenient for subsequent transportation and treatment;

[0174] Closed-loop verification module:

[0175] Cleaning effect verification principle: Verify the cleaning effect through deep feature comparison; first extract the multi-scale features of the images before and after cleaning and perform channel splicing. The feature fusion formula is: ( This is the image before cleaning. For the cleaned image, is a deep residual network used to extract image features. CoordConv is a coordinate convolution layer. represents channel splicing operation);

[0176] Dynamic adjustment of verification threshold: verification threshold Dynamically adjust according to ambient light, the adjustment formula is: (in, is the initial verification threshold, is the illumination adjustment coefficient, the illumination change is the difference between the current illumination and the reference illumination, and the reference illumination is the preset reference illumination value); in this way, it is possible to accurately judge whether the garbage is effectively cleaned under different illumination conditions to ensure the cleaning quality;

[0177] Collaboration between modules:

[0178] Each module works closely together; the high-precision positioning module provides accurate location information for the semantic map navigation module, allowing the vehicle to accurately drive to the task location; the path planned by the semantic map navigation module guides the vehicle to move efficiently, and at the same time feeds back the real-time road conditions to the task scheduling module; the classification cleaning device performs corresponding cleaning operations based on the type of garbage identified by the garbage detection module and the location where the vehicle arrives; the closed-loop verification module feeds back the cleaning effect to the cloud monitoring management platform, and if it does not meet the standards, it can prompt the platform to re-dispatch the vehicle for secondary cleaning;

[0179] Application value of the vehicle:

[0180] With their advanced configuration and efficient working mode, various types of automatic cleaning vehicles have greatly improved the automation and intelligence level of sanitation and cleaning in highway service areas. They can quickly and accurately clean up various types of garbage, improve cleaning efficiency, reduce labor costs, and ensure cleaning quality, creating a clean and comfortable environment for the service area and improving the overall service quality and image of the service area.

[0181] In this embodiment, the system also includes an exception handling mechanism. When the number of cleaning task failures exceeds a preset threshold, the manual intervention protocol is initiated and the fault data is uploaded to the cloud analysis platform, triggering equipment self-checking and parameter optimization;

[0182] Furthermore, the exception handling mechanism is an indispensable part of the highway service area sanitation and cleaning system based on the vehicle-road cloud map linkage. Its main function is to ensure that the system can maintain basic operation and restore to normal status in time when facing various abnormal situations, and effectively record and analyze abnormal situations to provide a basis for the continuous optimization of the system. The following is a detailed description of the mechanism:

[0183] Overall function overview:

[0184] The exception handling mechanism monitors the execution status of cleaning tasks, the operation of various hardware devices, the stability of system communication links and other key links in real time; once an abnormal situation is detected, the corresponding response measures are immediately initiated, including automatic recovery attempts, alarms, manual intervention, data recording and uploading, etc., to ensure the reliability of the system and the continuity of sanitation and cleaning work;

[0185] Abnormal monitoring range:

[0186] The exception handling mechanism mainly monitors the following aspects:

[0187] Abnormal execution of cleaning tasks: Continuously track the execution of assigned tasks by cleaning vehicles; if the number of cleaning task failures exceeds a preset threshold, such as three consecutive unsuccessful attempts to clean the same garbage, it is considered abnormal. This may be caused by equipment failure of the cleaning vehicle, inaccessible garbage location or other unexpected factors;

[0188] Hardware equipment failure: various hardware equipment of cleaning vehicles, such as high-precision positioning modules, classification cleaning devices, drive systems, etc., as well as hardware facilities of roadside monitoring units and cloud monitoring management platforms, are monitored; for example, when the positioning module of the cleaning vehicle cannot obtain valid positioning data for 5 consecutive minutes, or the CPU usage rate of the cloud platform server exceeds 90% for more than 10 minutes, it is considered that the hardware equipment has failed;

[0189] Communication link abnormality: Monitor the data transmission link between the vehicle, road and cloud. If the data transmission interruption time exceeds 1 minute, or the packet loss rate exceeds 10% for 5 consecutive minutes, it is considered as a communication link abnormality. This may affect the coordination between the various parts of the system, leading to problems in task allocation, data transmission and other links.

[0190] Exception handling process:

[0191] Automatic recovery attempts:

[0192] When an abnormality is detected, the system first attempts to automatically recover; for example, if the communication link is abnormal, the system will automatically reconnect the communication port and try to restore data transmission; if the cleaning vehicle has a task execution abnormality due to a temporary software failure, the system will automatically restart the relevant software module; automatic recovery attempts will be made multiple times within the set time. If the system is successfully recovered within the specified number of times, the system will continue to operate normally and record the abnormality and recovery process;

[0193] Alert issued:

[0194] If the automatic recovery attempt fails, the exception handling mechanism will immediately sound an alarm; the alarm information will be notified to relevant personnel in a variety of ways, such as sending SMS alarms to the system administrator's mobile phone, popping up a striking red alarm prompt on the cloud monitoring management platform interface, etc. The alarm content details the type of abnormality, time of occurrence, location of occurrence (if it involves a cleaning vehicle, it includes the vehicle number and approximate location) and other key information, so that the administrator can quickly understand the situation and take further measures;

[0195] Manual intervention initiated:

[0196] When the number of cleaning task failures exceeds the threshold or other serious abnormal situations occur and automatic recovery is ineffective, the manual intervention protocol is initiated; the system will suspend the automatic scheduling of related abnormal tasks and generate a detailed manual intervention task list on the cloud monitoring management platform, which contains abnormal details, recommended treatment measures and other information; at the same time, the maintenance personnel are notified to go to the site or conduct manual troubleshooting and repairs of the system through remote operation; for example, if the classification cleaning device of the cleaning vehicle fails, the maintenance personnel need to go to the vehicle location for inspection;

[0197] Data recording and uploading:

[0198] During the exception handling process, the system will record all the data related to the exception in detail, including the operating parameters of each module of the system in the period before the exception occurred, the specific status information when the exception occurred, the operation records of automatic recovery attempts, and the various operations and feedback during the manual intervention process. These data will be uploaded to the cloud analysis platform for subsequent abnormal cause analysis, system performance evaluation, and optimization and improvement. By analyzing a large amount of abnormal data, we can find the potential weak links of the system, optimize and upgrade it in a targeted manner, and improve the stability and reliability of the system.

[0199] Application value of the mechanism:

[0200] The exception handling mechanism effectively enhances the robustness and fault tolerance of the highway service area sanitation and cleaning system. When faced with a complex and changeable actual operating environment, it can respond to various abnormal situations in a timely manner, ensure the continuous progress of sanitation and cleaning work, and reduce the deterioration of sanitation conditions in service areas caused by abnormalities. At the same time, through the recording and analysis of abnormal data, it provides a valuable basis for the optimization and improvement of the system, which helps to continuously improve the overall performance and reliability of the system and provide more stable and efficient sanitation and cleaning services for service areas.

[0201] In this embodiment, the update frequency of dynamic background modeling is positively correlated with the traffic volume. For every 50 vehicles / minute increase in traffic volume, the update frequency is doubled, and the maximum update delay does not exceed 2 seconds.

[0202] Specifically, dynamic background modeling plays a key role in the highway service area sanitation cleaning system of the present invention, and a subtle correlation mechanism is established between its update frequency and vehicle flow;

[0203] In a complex environment such as a highway service area, the traffic volume is in a state of continuous change. The roadside monitoring unit is responsible for real-time monitoring of the traffic volume. Through advanced sensor technology and intelligent algorithms, it can accurately count the number of vehicles passing through a specific area per unit time and quantify it in minutes.

[0204] When the traffic volume is relatively low, the dynamic background modeling will be updated at a basic frequency; for example, assuming that the traffic volume is less than 50 vehicles / minute, the update frequency of the background modeling is set to once every 4 seconds; as the traffic volume gradually increases, once the incremental threshold of 50 vehicles / minute is reached, the update frequency will be increased accordingly; specifically, the update frequency will be doubled directly, that is, from the original update every 4 seconds to update every 2 seconds; if the traffic volume continues to increase, every time the increment reaches 50 vehicles / minute again, the update frequency will double again to update every 1 second, and so on;

[0205] However, in order to ensure the stability of the system and the effectiveness of data processing, a maximum update delay limit is set; no matter how rapidly the traffic volume increases, the update delay of dynamic background modeling is strictly controlled to no more than 2 seconds; this means that even if the traffic volume increases significantly, the update frequency may theoretically be extremely high, but the actual update interval will not be shorter than 2 seconds; for example, when the traffic volume continues to increase, the update frequency should reach once every 0.5 seconds according to the positive correlation rule, but due to the maximum update delay limit, the actual update frequency remains at once every 2 seconds; this limitation avoids excessive consumption of system resources due to too fast an update frequency, ensuring that the system can maintain stable operation while efficiently capturing dynamic changes in the environment, thereby providing timely and reliable background data support for subsequent links such as garbage detection, and facilitating the efficient operation of the entire highway service area sanitation and cleaning system.

[0206] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A highway service area sanitation cleaning system based on vehicle-road cloud map linkage, characterized by: include: The roadside monitoring unit is deployed in the preset area of ​​the service area to collect image data of the monitoring area in real time, extract scene change information through dynamic background modeling, and establish a three-dimensional mapping relationship between the image coordinate system and the world coordinate system based on binocular vision calibration; The cloud monitoring management platform communicates with the roadside monitoring unit and includes: The garbage detection module uses an improved target detection algorithm to identify and classify garbage in image data; The positioning calculation module converts the detected garbage pixel coordinates into actual physical coordinates based on binocular stereo vision and RTK collaborative positioning technology; The task scheduling module generates cleaning tasks based on a multi-factor decision model and dynamically allocates them to cleaning vehicles; Various types of automatic cleaning vehicles, including special cleaning vehicles for recyclable and non-recyclable garbage. Each cleaning vehicle is equipped with: High-precision positioning module, achieving centimeter-level positioning; Semantic map navigation module, dynamic path planning based on incrementally updated maps; Classification and cleaning device, which can be adapted to the physical characteristics of different types of garbage for sorting or compression; Closed-loop verification module verifies the cleaning effect through deep feature comparison.

2. A highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1, characterized in that: The improved target detection algorithm is an improved YOLOv11 model, including: Based on the size distribution characteristics of garbage in the service area, the K-means++ clustering algorithm is used to optimize the anchor frame size. The anchor frame size is the logarithmic mean of the width and height of the annotation frame. The specific calculation method is: after taking the natural logarithm of the width and height of the annotation frame, the arithmetic mean is calculated as the anchor frame size; A binocular vision feature fusion mechanism is introduced into the Backbone network. The left and right eye feature maps are fused through a three-dimensional convolution operation. After the spatial attention weight is generated, it is multiplied element by element with the left eye feature map, and the fused feature map is output. The loss function is defined as the weighted sum of the improved Inner-IoU loss function and the RTK positioning coordinate regression loss, where the weight factor of the RTK regression loss is 0.

8.

3. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The coordinate conversion method of the positioning calculation module includes: The camera internal parameter matrix and external parameter matrix are obtained through binocular positioning, where the camera internal parameter matrix includes the focal length , and the optical center coordinates , the camera external parameter matrix includes the rotation matrix and translation vectors , the target three-dimensional coordinates are calculated based on the difference in the horizontal coordinates of the left and right eye images of the target point: Depth value It is equal to the product of the equivalent focal length and the binocular baseline distance divided by the parallax value; Target actual horizontal coordinate It is equal to the depth value multiplied by the difference between the horizontal coordinate of the left eye image and the horizontal coordinate of the optical center, and then divided by the left eye focal length; Target actual vertical coordinate It is equal to the depth value multiplied by the difference between the ordinate of the left eye image and the ordinate of the optical center, and then divided by the left eye focal length; When the binocular positioning depth error exceeds 0.1 meters, the coordinates are corrected based on the RTK positioning data. The correction weight is dynamically calculated based on the RTK signal quality. The higher the signal quality, the greater the weight of the RTK data.

4. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: In the path planning algorithm of the task scheduling module, the path cost function combines the following factors: The actual length of the path g(n) and the estimated remaining length h(n); The priority weight of garbage type, the priority of flammable garbage is increased by 50%; Real-time obstacle density factor. The higher the obstacle density, the greater the path cost. RTK positioning confidence factor. The lower the confidence, the higher the path cost. The weight coefficients of garbage priority, obstacle density and RTK confidence in the cost function are 0.6, 0.3 and 0.1 respectively.

5. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The closed-loop verification module is implemented by the following steps: Extract multi-scale features of the images before and after cleaning, perform channel concatenation of the ResNet-50 features of the cleaned image and the coordinate-aware convolutional features of the pre-cleaned image to generate fused features; The verification threshold is dynamically adjusted according to the ambient light intensity. The base threshold is 0.

8. For every increase of 1 times the base light intensity, the threshold is increased by 0.2 times. If the verification fails, task reallocation, semantic map correction and binocular vision calibration parameter adjustment will be triggered. The calibration parameter correction amount is proportional to the difference in fusion features.

6. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The implementation of the high-precision positioning module includes: The carrier phase differential technology is used to achieve RTK centimeter-level positioning, with a positioning error of less than 0.05 meters; When the RTK signal is lost, it automatically switches to the binocular vision positioning mode. The switching condition is that the deviation between the binocular positioning data of three consecutive frames and the historical trajectory exceeds the threshold.

7. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The working modes of the classification cleaning device include: When sorting recyclable garbage, the magnetic strength of the magnetic attraction mechanism is proportional to the target volume, and the proportionality coefficient is 0.5; When compressing non-recyclable garbage, the vacuum adsorption negative pressure value is the smaller value between 0.8 times the garbage density value and 20kPa, and the compressed volume is less than 1 / 5 of the original volume.

8. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The incremental update method of the semantic map includes: The grayscale difference between the current image and the reference map is dynamically compared through binocular vision to generate a difference mask. The area with a grayscale difference of more than 30 is marked as a changed area. After reconstructing the three-dimensional point cloud of the changed area, the updated garbage distribution, obstacle location and charging pile status are integrated into the semantic map. The fusion weight of the new and old maps is 0.7, and a global consistency check is performed once an hour.

9. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: It also includes an exception handling mechanism. When the number of cleaning task failures exceeds the preset threshold, the manual intervention protocol is initiated and the fault data is uploaded to the cloud analysis platform, triggering equipment self-inspection and parameter optimization.

10. The highway service area sanitation cleaning system based on vehicle-road cloud map linkage according to claim 1 is characterized by: The update frequency of the dynamic background modeling is positively correlated with the traffic volume. For every increase of 50 vehicles / minute in the traffic volume, the update frequency increases by 1 time, and the maximum update delay does not exceed 2 seconds.

Images