Urban garbage cleaning system and path planning method of unmanned aerial vehicle and sweeper truck cooperation

Abstract

The application discloses an unmanned aerial vehicle and sweeper cooperative urban garbage cleaning system and path planning method, and belongs to the technical field of intelligent driving. The unmanned aerial vehicle subsystem is provided with an image recognition unit based on a space-time fusion semantic segmentation model, which is used for over-the-horizon cruising of urban roads, and realizes accurate identification, positioning and coordinate conversion of garbage in the driving lane. The sweeper subsystem is provided with a visual perception unit based on multi-sensor fusion, which is used for perceiving obstacles and garbage, and switching operation modes according to the garbage type. The cloud control platform is used for receiving coordinate information, coordinating and scheduling in combination with real-time traffic signal data, and issuing optimal cleaning path instructions. The above-mentioned unmanned aerial vehicle and sweeper cooperative urban garbage cleaning system and path planning method solve the problems of limited coverage, rigid path planning and poor air-ground cooperation of traditional environmental sanitation operation, and significantly improve the operation efficiency of urban garbage cleaning and road traffic safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to an urban waste cleaning system and path planning method that integrates unmanned aerial vehicles (UAVs) and sweepers. Background Technology

[0002] With the acceleration of urbanization, the amount of urban waste generated in my country continues to rise. According to statistics from the China Urban Environmental Sanitation Association, nearly 1 billion tons of waste are generated nationwide each year. To improve waste collection efficiency, existing technologies have begun to introduce drones and intelligent sweepers, but there are still multiple technical bottlenecks: drone image recognition mostly uses single-frame semantic segmentation without combining flight spatiotemporal information, resulting in high rates of missed detections and duplicate recognitions, making it difficult to output accurate waste locations; path planning lacks dynamic adaptation capabilities, and existing sweepers mostly travel according to preset routes without considering real-time traffic signals and other multi-dimensional factors, resulting in significant energy waste.

[0003] Furthermore, the operating range of existing unmanned sweeping vehicles is mostly limited to the roadside area. They cannot promptly detect and quickly clean up litter left behind by traffic accidents, cargo dropped from freight vehicles, or scattered debris blown onto the road surface. This type of road litter easily causes traffic accidents and poses significant road safety hazards. In summary, the existing sanitation system suffers from low recognition accuracy, poor operational efficiency, weak air-ground coordination, and insufficient environmental protection and safety. There is an urgent need for an intelligent sanitation system that integrates spatiotemporal perception, dynamic planning, and cloud-based collaboration to meet the development needs of intelligent, efficient, and safe urban waste management. Summary of the Invention

[0004] The purpose of this invention is to provide an urban waste cleaning system and path planning method that coordinates drones and sweeping vehicles, thereby solving the aforementioned technical problems.

[0005] To achieve the above objectives, the present invention provides an urban waste cleaning system that coordinates unmanned aerial vehicles (UAVs) and sweepers, including an UAV subsystem, a sweeper subsystem, and a cloud control platform; The unmanned aerial vehicle (UAV) subsystem is equipped with an image recognition unit based on a spatiotemporal fusion semantic segmentation model, which is used for beyond-line-of-sight (BLOS) cruising of urban roads to achieve accurate identification, positioning, and coordinate transformation of garbage in the driving lanes; The sweeper system is equipped with a vision perception unit based on multi-sensor fusion to detect obstacles and garbage, and switch the operation mode according to the type of garbage; The cloud-based control platform receives the ground coordinates of the garbage uploaded by the drone subsystem, performs coordinated scheduling in conjunction with real-time traffic signal data, and issues the optimal cleaning path instruction to the sweeper subsystem.

[0006] Preferably, the spatiotemporal fusion semantic segmentation model of the UAV subsystem includes a spatiotemporal feature extraction module, a feature enhancement module, and a semantic segmentation module; The spatiotemporal feature extraction module is used to simultaneously extract visual features from images and spatiotemporal information from UAVs; The feature enhancement module is used to fuse visual features with spatiotemporal information to improve the distinguishability of waste features; The semantic segmentation module is used to output single-frame garbage segmentation results.

[0007] Preferably, the visual perception unit of the sweeper system includes an on-board camera, millimeter-wave radar, and lidar for real-time acquisition of environmental point cloud data and image data; the sweeper system also includes a sweeping operation device, which consists of a water spraying device, an adjustable speed disc brush structure, and a lightweight robotic arm. The sweeping operation device automatically switches sweeping modes based on the perceived type and size of the garbage.

[0008] Preferably, the cloud control platform and the drone subsystem and sweeper vehicle subsystem use a communication network for data interaction to ensure the real-time transmission of garbage location information and traffic signal data.

[0009] A path planning method for an urban waste collection system that combines drones and sweeping vehicles includes the following steps: S1. Drone patrol and identification: The drone patrols beyond visual range over the cleaning area, and after preprocessing the collected images, it inputs them into the spatiotemporal fusion semantic segmentation model to obtain the pixel coordinates of the garbage. S2. Coordinate Transformation and Upload: Based on the principle of flight path calculation, the coordinates of garbage pixels are converted into geodetic coordinates in real time using complex function affine transformation, and then uploaded to the cloud control platform. S3, Cloud-based Collaborative Scheduling: The cloud-based control platform receives garbage location information and, in conjunction with real-time green wave signal parameters at each intersection and sweeper operating data, establishes a cost matrix that includes energy consumption and time. S4. Dynamic Path Planning: The sweeper receives instructions from the cloud, uses the green wave decision algorithm to calculate the traffic efficiency at the intersection, and solves the cost matrix through dynamic programming to obtain the optimal movement path with low energy consumption and efficient garbage recycling.

[0010] Preferably, the affine transformation formula for the complex function in S2 is: ; in, , where is the complex number representation of the garbage point in the image pixel coordinate system. , The x and y coordinates of the garbage points are the pixel coordinates. The imaginary unit of complex numbers. Let be the complex number representation of the garbage collection point in the geodetic coordinate system. , The horizontal and vertical coordinates of the garbage collection point are: , where are the complex coefficients of the transformation. This represents the scaling ratio from pixel coordinates to geodetic coordinates. The angle between the pixel coordinate system and the geodetic coordinate system. , where is the translational complex constant, corresponding to the UAV's GPS geodetic coordinates at the time of image acquisition. , The horizontal and vertical coordinates of the UAV at the moment of image acquisition.

[0011] Preferably, the real-time green wave signal parameters in S3 include the green light duration. Signal period and remaining green light time The sweeper's operating data includes the current remaining battery power, driving speed, and location; The rule for dynamically adjusting the priority of path nodes in the cost matrix is: if the remaining green light time at the intersection corresponding to a certain path node is... Exceeding the threshold If the remaining battery power is sufficient, then the overall cost of that node will be reduced; if the intersection corresponding to a path node is red, that is... The overall cost of this node will be increased.

[0012] Preferably, the intersection traffic efficiency formula in S4 is: ; in, For traffic efficiency, For the observation time The number of vehicles passing through, This represents the number of vehicles that can pass through during the green light period.

[0013] Preferably, when performing spatiotemporal fusion on the multi-frame continuous image segmentation results output by the semantic segmentation module, the spatial transformation relationship between adjacent frames is calculated using the UAV's GPS track, attitude angle, and flight speed to fill in the missing garbage regions in a single frame, and duplicate recognition results of the same garbage target in multiple frames are deduplicated. Assuming the UAV maintains a constant altitude, the process includes the following steps: Step 1: Input the preprocessed standardized image into the spatiotemporal fusion semantic segmentation model, and output the garbage confidence map for each frame. threshold At that time, it was determined to be the initial garbage area; Step 2: Obtain the UAV GPS coordinates corresponding to two adjacent frames. ) 、 ( ) and attitude angles, calculate the spatial transformation matrix Its formula is: ; in, , which is the difference in the horizontal coordinate. , is the difference in the ordinate. Yaw angle UAV stands for Unmanned Aerial Vehicle. For the number of frames; Step 3: Based on the spatial transformation matrix , will the Pixel coordinates of the initial garbage region of the frame ( Mapped to the first The corresponding pixel coordinates of the frame ( ); Step 4: Compare the first The initial garbage region of the frame and the mapped first garbage region Frame garbage region, if the first No garbage identification result at the corresponding frame location, i.e., confidence level. threshold But the first Frame corresponding region confidence threshold This area is added as the first Garbage areas of frames; Step 5: Calculate the... Frame mapping region and the first The overlapping area of ​​the initial region of the frame, if the overlap degree If the same garbage target is identified, the area with higher confidence level will be retained; Step 6: After completion and deduplication, output the spatiotemporally consistent garbage segmentation results and accurate pixel coordinates, and upload them to the cloud control platform.

[0014] Preferably, the UAV subsystem sequentially performs grayscale correction, Gaussian denoising based on the MMSE (Minimum Mean Square Error) criterion, and size normalization operations during the image preprocessing stage to generate a standardized image input model.

[0015] Therefore, the urban waste cleaning system and path planning method using the above-mentioned drone and sweeper vehicle collaboration have the following beneficial effects: 1. By adopting a spatiotemporal fusion semantic segmentation model and combining UAV flight spatiotemporal information, the problem of high false negative rate and duplicate recognition rate in existing UAV single-frame semantic segmentation is solved, and accurate garbage identification and location are achieved.

[0016] 2. By using cloud-based collaborative scheduling and combining real-time traffic signals and other multi-dimensional factors, the optimal cleaning path is planned, replacing the traditional preset route operation mode and effectively reducing the energy consumption of the sweeper.

[0017] 3. The drone's beyond-visual-range patrol, combined with the ground sweeper, covers areas that are difficult for traditional sweepers to reach, such as driving lanes, and promptly cleans up garbage left on driving lanes, avoiding road safety risks.

[0018] 4. Construct an "air-ground-cloud" collaborative system to solve the problems of weak air-ground coordination and poor operation efficiency in existing technologies, improve the level of intelligence and efficiency of cleaning, optimize operation mode to reduce energy waste and eliminate road safety hazards, and meet the development needs of intelligent, efficient and safe urban waste management.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is an overall architecture diagram of the urban waste cleaning system that combines drones and sweeping vehicles according to the present invention; Figure 2 This is an overall flowchart of the urban waste path planning method for the collaboration between drones and sweeping vehicles according to the present invention; Figure 3 This is a flowchart illustrating the spatiotemporal fusion process of the urban waste path planning method for the collaboration between drones and sweeping vehicles according to the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0022] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] like Figures 1-3As shown, this invention provides an urban waste cleaning system and path planning method that coordinates drones and sweepers. The system includes a drone subsystem, a sweeper subsystem, and a cloud control platform. The drone subsystem is equipped with an image recognition unit based on a spatiotemporal fusion semantic segmentation model, which overcomes the shortcomings of traditional single-vision recognition, such as susceptibility to lighting, occlusion, and limited field of view. It is used for beyond-line-of-sight cruising on urban roads to achieve accurate identification, positioning, and coordinate transformation of waste in the driving lane. The sweeper subsystem is equipped with a visual perception unit based on multi-sensor fusion, which compensates for the insufficient perception accuracy of a single sensor under complex road conditions (rain, backlight, obstacle occlusion). It is used to perceive obstacles and waste and switch operating modes according to waste type. The cloud control platform, as the core scheduling hub of the system, receives the waste geodetic coordinate information uploaded by the drone subsystem, combines it with real-time traffic signal data for coordinated scheduling, and issues dynamic optimal cleaning path instructions to the sweeper subsystem to achieve coordinated adaptation between waste cleaning and urban traffic.

[0025] The spatiotemporal fusion semantic segmentation model of the UAV subsystem includes a spatiotemporal feature extraction module, a feature enhancement module, and a semantic segmentation module. The spatiotemporal feature extraction module is used to simultaneously extract image visual features and UAV spatiotemporal information, such as image visual texture and contour features, and UAV GPS coordinates, flight time, attitude, and other spatiotemporal information. The feature enhancement module is used to fuse visual features and spatiotemporal information, enhance the feature differences between garbage and road backgrounds, improve the distinguishability of garbage features, and reduce the misidentification rate of similar debris such as paper scraps and fallen leaves. The semantic segmentation module is used to output single-frame garbage segmentation results, providing an accurate data foundation for subsequent garbage location and coordinate transformation.

[0026] The sweeper system's visual perception unit includes an onboard camera, millimeter-wave radar, and lidar, used to acquire environmental point cloud data and image data in real time, enabling all-weather, all-condition environmental perception. The sweeper system also includes a sweeping operation device, which consists of a water spraying device, an adjustable-speed disc brush structure, and a lightweight robotic arm. The water spraying device suppresses dust and moistens stubborn garbage, the adjustable-speed disc brush is suitable for sweeping scattered garbage, and the lightweight robotic arm can grab and transport block / large garbage. The sweeping operation device automatically switches sweeping modes based on the perceived garbage type and size, achieving refined and intelligent sweeping, improving garbage recycling efficiency and road surface cleanliness.

[0027] The cloud control platform interacts with the drone subsystem and the sweeper subsystem using communication networks such as 5G and cellular vehicle-to-everything (C-V2X) to ensure low-latency and high-reliability transmission of garbage location information and traffic signal data, meeting the real-time requirements of collaborative scheduling.

[0028] The path planning method for a collaborative urban waste cleaning system using drones and sweepers includes the following steps: S1, Drone cruising and identification: The drone conducts beyond-line-of-sight (BLOS) cruising over the cleaning area, preprocesses the collected images, and inputs them into a spatiotemporal fusion semantic segmentation model to obtain the pixel coordinates of the waste; S2, Coordinate transformation and uploading: Based on the principle of trajectory extrapolation, the pixel coordinates of the waste are converted into geodetic coordinates in real time using complex variable function affine transformation, and then uploaded to the cloud control platform to achieve accurate mapping from pixel positioning to geospatial positioning; S3, Cloud-based collaborative scheduling: The cloud control platform receives the waste location information and combines it with real-time data from various intersections... Green wave signal parameters (green wave signal refers to the coordinated timing of continuous intersections in the urban road network, which enables vehicles to pass through continuously green lights at a constant speed) and sweeper operating data are used to construct multi-objective optimization constraints and establish a cost matrix that includes energy consumption and time. The cost matrix comprehensively considers sweeping costs and traffic efficiency, providing data support for solving the optimal path. S4, Dynamic Path Planning: The sweeper receives instructions from the cloud and uses the green wave decision algorithm to calculate the intersection traffic efficiency. By solving the cost matrix through dynamic programming, the optimal movement path with lower energy consumption, higher traffic efficiency, and more efficient garbage collection is obtained, which greatly reduces the sweeper's waiting time at intersections and ineffective detours.

[0029] The formula for the affine transformation of complex functions in S2 is: ; in, , where is the complex number representation of the garbage point in the image pixel coordinate system. , The x and y coordinates of the garbage points are the pixel coordinates. The imaginary unit of complex numbers. Let be the complex number representation of the garbage collection point in the geodetic coordinate system. , The horizontal and vertical coordinates of the garbage collection point are: , where are the complex coefficients of the transformation. This represents the scaling ratio from pixel coordinates to geodetic coordinates. The angle between the pixel coordinate system and the geodetic coordinate system. , where is the translational complex constant, corresponding to the UAV's GPS geodetic coordinates at the time of image acquisition. , The transformation uses complex variable functions to simplify the calculations for the UAV's horizontal and vertical coordinates at the time of image acquisition. Compared with traditional two-dimensional affine transformation, it is faster to calculate and has higher coordinate transformation accuracy.

[0030] The real-time green wave signal parameters in S3 include the green light duration. Signal period and remaining green light time The green wave signal parameters can be obtained using a navigation system (such as Gaode Maps), while the sweeper's operating data includes the current remaining battery power, driving speed, and location; the rule for dynamically adjusting the priority of path nodes in the cost matrix is: if the remaining green light time at the intersection corresponding to a certain path node is... Exceeding the threshold If the remaining battery power is sufficient, then the overall cost of that node will be reduced; if the intersection corresponding to a path node is red, that is... The overall cost of this node is increased, and the cost weight is dynamically adjusted to guide the sweeper to prioritize green light intersections and avoid waiting at red lights.

[0031] The formula for intersection traffic efficiency in S4 is: ; in, For traffic efficiency, For the observation time The number of vehicles passing through, The number of vehicles that can pass through during the green light period is quantified by measuring the traffic capacity of the intersection per unit time, providing a quantitative evaluation indicator for green wave decision-making.

[0032] When performing spatiotemporal fusion on the multi-frame continuous image segmentation results output by the semantic segmentation module, the spatial transformation relationship between adjacent frames is calculated using the UAV's GPS track, attitude angle, and flight speed. This completes the missing garbage regions in a single frame and deduplicates the repeated identification results of the same garbage target in multiple frames. Assuming the UAV maintains a constant altitude to eliminate altitude change interference, simplifying spatial transformation calculations, and improving real-time performance, the process includes the following steps: Step 1: Preprocessing the images acquired by the UAV by sequentially performing grayscale correction, Gaussian denoising, and size normalization operations to eliminate interference from uneven lighting, airflow turbulence, sensor noise, etc., improving the quality of the model input data and generating standardized images that can be imported into the model. Grayscale correction uses a grayscale transformation method in the image spatial domain, with the following formula: ; in, The image to be processed. The processed image, This defines a grayscale transformation operation in image space, used to map the pixel grayscale values ​​of an input image to the pixel grayscale values ​​of an output image. Gaussian denoising based on the MMSE criterion is employed to denoise additive white Gaussian noise in UAV aerial images. Denoising is achieved through minimum mean square error estimation. Specifically, let a noisy image be... , For noisy images, For noise-free images, To follow a normal distribution with variance of Gaussian noise, when represented in the transform domain of the image, has ,in, The coefficients are those of the transformed image with noise. The coefficients are the transformed values ​​of the noise-free image. The coefficients after noise transformation (which still follow a variance of ) (Gaussian distribution). In image denoising, sometimes the variance of the noise... It is known, and sometimes unknown, and needs to be estimated from noisy images. The formula is: ,in, This indicates taking the median value, while The variance estimate is: If the variance is obtained , According to the MMSE criterion, the formula for estimating a noise-free image is: The preprocessed standardized images are input into the spatiotemporal fusion semantic segmentation model, which outputs a garbage confidence map for each frame. When the confidence level... threshold At that time, the threshold value ranges from 0 to 1. We set the confidence level to 0.7. At 0.7, it is determined to be the initial garbage area; Step 2: Obtain the UAV GPS coordinates corresponding to two adjacent frames. ), ( ) and attitude angles, calculate the spatial transformation matrix Its formula is: ; in, , which is the difference in the horizontal coordinate. , is the difference in the ordinate. Yaw angle UAV stands for Unmanned Aerial Vehicle. For the number of frames; Step 3, based on the spatial transformation matrix , will the Pixel coordinates of the initial garbage region of the frame ( Mapped to the first The corresponding pixel coordinates of the frame ( Step 4: Compare the spatial locations of the same garbage target across different frames. The initial garbage region of the frame and the mapped first garbage region Frame garbage region, if the first No garbage identification result at the corresponding frame location, i.e., confidence level. 0.7, but the first Frame corresponding region confidence Threshold 0.7, fill in the area as the first The garbage region of the frame is identified. This step fills in the gaps caused by occlusion or blurring in a single frame, improving the completeness of the identification; Step 5: Calculate the garbage region of the frame. Frame mapping region and the first The overlapping area of ​​the initial region of the frame, if the overlap degree Here, y is set to 80%, meaning that if the overlap is above 80%, it is determined to be the same garbage target, and the area with higher confidence is retained. By removing duplicate identification data from multiple frames, the processing pressure of redundant information in the cloud is reduced. Step 6: After completion and deduplication, the spatiotemporally continuous and accurately positioned garbage segmentation results and precise pixel coordinates are output and uploaded to the cloud control platform to provide a reliable garbage positioning basis for cloud path planning and ensure the accuracy of collaborative cleaning.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A city waste cleaning system that coordinates unmanned aerial vehicles (UAVs) and sweeping vehicles, characterized in that: This includes a drone subsystem, a sweeper vehicle system, and a cloud control platform; The unmanned aerial vehicle (UAV) subsystem is equipped with an image recognition unit based on a spatiotemporal fusion semantic segmentation model, which is used for beyond-line-of-sight (BLOS) cruising of urban roads to achieve accurate identification, positioning, and coordinate transformation of garbage in the driving lanes; The sweeper system is equipped with a vision perception unit based on multi-sensor fusion to detect obstacles and garbage, and switch the operation mode according to the type of garbage; The cloud-based control platform receives the ground coordinates of the garbage uploaded by the drone subsystem, performs coordinated scheduling in conjunction with real-time traffic signal data, and issues the optimal cleaning path instruction to the sweeper subsystem.

2. The urban waste cleaning system combining drones and sweepers according to claim 1, characterized in that: The spatiotemporal fusion semantic segmentation model of the UAV subsystem includes a spatiotemporal feature extraction module, a feature enhancement module, and a semantic segmentation module; The spatiotemporal feature extraction module is used to simultaneously extract visual features from images and spatiotemporal information from UAVs; The feature enhancement module is used to fuse visual features with spatiotemporal information to improve the distinguishability of waste features; The semantic segmentation module is used to output single-frame garbage segmentation results.

3. The urban waste cleaning system combining drones and sweepers according to claim 2, characterized in that: The sweeper system's visual perception unit includes an onboard camera, millimeter-wave radar, and lidar, used to acquire environmental point cloud data and image data in real time. The sweeper system also includes a sweeping operation device, which consists of a water spraying device, an adjustable speed disc brush structure, and a lightweight robotic arm. The sweeping operation device automatically switches sweeping modes based on the perceived type and size of the garbage.

4. The urban waste cleaning system combining drones and sweepers according to claim 3, characterized in that: The cloud control platform interacts with the drone subsystem and the sweeper subsystem via a communication network to ensure the real-time transmission of garbage location information and traffic signal data.

5. The path planning method for an urban waste cleaning system that coordinates unmanned aerial vehicles and sweeping vehicles according to any one of claims 1-4, characterized in that, Includes the following steps: S1. Drone patrol and identification: The drone patrols beyond visual range over the cleaning area, and after preprocessing the collected images, it inputs them into the spatiotemporal fusion semantic segmentation model to obtain the pixel coordinates of the garbage. S2. Coordinate Transformation and Upload: Based on the principle of flight path calculation, the coordinates of garbage pixels are converted into geodetic coordinates in real time using complex function affine transformation, and then uploaded to the cloud control platform. S3, Cloud-based Collaborative Scheduling: The cloud-based control platform receives garbage location information and, in conjunction with real-time green wave signal parameters at each intersection and sweeper operating data, establishes a cost matrix that includes energy consumption and time. S4. Dynamic Path Planning: The sweeper receives instructions from the cloud, uses the green wave decision algorithm to calculate the traffic efficiency at the intersection, and solves the cost matrix through dynamic programming to obtain the optimal movement path with low energy consumption and efficient garbage recycling.

6. The path planning method for the urban waste cleaning system coordinating drones and sweepers according to claim 5, characterized in that: The formula for the affine transformation of complex functions in S2 is: ； in, , where is the complex number representation of the garbage point in the image pixel coordinate system. , The x and y coordinates of the garbage points are the pixel coordinates. The imaginary unit of complex numbers. Let be the complex number representation of the garbage collection point in the geodetic coordinate system. , The horizontal and vertical coordinates of the garbage collection point are: , where are the complex coefficients of the transformation. This represents the scaling ratio from pixel coordinates to geodetic coordinates. The angle between the pixel coordinate system and the geodetic coordinate system. , where is the translational complex constant, corresponding to the UAV's GPS geodetic coordinates at the time of image acquisition. , The horizontal and vertical coordinates of the UAV at the moment of image acquisition.

7. The path planning method for the urban waste cleaning system coordinating drones and sweepers according to claim 6, characterized in that: The real-time green wave signal parameters in S3 include the green light duration. Signal period and remaining green light time The sweeper's operating data includes the current remaining battery power, driving speed, and location; The rule for dynamically adjusting the priority of path nodes in the cost matrix is: if the remaining green light time at the intersection corresponding to a certain path node is... Exceeding the threshold If the remaining battery power is sufficient, then the overall cost of that node will be reduced; if the intersection corresponding to a path node is red, that is... The overall cost of this node will be increased.

8. The path planning method for the urban waste cleaning system coordinating drones and sweepers according to claim 7, characterized in that: The formula for intersection traffic efficiency in S4 is: ； in, For traffic efficiency, For the observation time The number of vehicles passing through, This represents the number of vehicles that can pass through during the green light period.

9. The path planning method for an urban waste cleaning system that combines unmanned aerial vehicles (UAVs) and sweepers according to claim 8, characterized in that: When performing spatiotemporal fusion on the multi-frame continuous image segmentation results output by the semantic segmentation module, the spatial transformation relationship between adjacent frames is calculated using the UAV's GPS track, attitude angle, and flight speed. This completes the missing garbage regions in a single frame and removes duplicate recognition results of the same garbage target in multiple frames. Assuming the UAV maintains a constant altitude, the process includes the following steps: Step 1: Input the preprocessed standardized image into the spatiotemporal fusion semantic segmentation model, and output the garbage confidence map for each frame. threshold At that time, it was determined to be the initial garbage area; Step 2: Obtain the UAV GPS coordinates corresponding to two adjacent frames. ), ( ) and attitude angles, calculate the spatial transformation matrix Its formula is: ； in, , which is the difference in the horizontal coordinate. , is the difference in the ordinate. Yaw angle UAV stands for Unmanned Aerial Vehicle. For the number of frames; Step 3: Based on the spatial transformation matrix , will the Pixel coordinates of the initial garbage region of the frame ( Mapped to the first The corresponding pixel coordinates of the frame ( ); Step 4: Compare the first The initial garbage region of the frame and the mapped first garbage region Frame garbage region, if the first No garbage identification result at the corresponding frame location, i.e., confidence level. threshold But the first Frame corresponding region confidence threshold This area is added as the first Garbage areas of frames; Step 5: Calculate the... Frame mapping region and the first The overlapping area of ​​the initial region of the frame, if the overlap degree If the same garbage target is identified, the area with higher confidence level will be retained; Step 6: After completion and deduplication, output the spatiotemporally consistent garbage segmentation results and accurate pixel coordinates, and upload them to the cloud control platform.

10. The path planning method for an urban waste cleaning system coordinating drones and sweepers according to claim 9, characterized in that: In the image preprocessing stage, the UAV subsystem sequentially performs grayscale correction, Gaussian denoising based on the MMSE criterion, and size normalization operations to generate a standardized image input model.

Images