Method and system for rapid and continuous detection of missing street trees based on dynamic edge computing

Abstract

The application discloses a street tree missing plant rapid continuous detection method and system based on dynamic edge calculation. The method comprises the following steps: acquiring images, GPS coordinates and time stamps, and constructing a space-time trajectory; pre-processing the images, identifying and classifying vegetation, municipal facilities and moving shelter targets, and outputting feature information and confidence score to obtain image recognition results; based on the image recognition results, filtering street trees close to lane lines and predicting corresponding positions, constructing motion trajectories through continuous multi-frame data association, calibrating in a coordinate system to generate a predicted attention area, and forming a street tree position prediction result; and comprehensively judging and classifying the missing plant situation in the green belt based on the street tree position prediction result, and generating relevant event information. Through the implementation of the method, the limitations of the prior art can be overcome, more efficient and accurate technical support can be provided for the daily maintenance of urban greening, and the quality and efficiency of urban management can be improved.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically to a method and system for rapid and continuous detection of missing trees in roadside trees based on dynamic edge computing. Background Technology

[0002] With the acceleration of urbanization, urban greening management is becoming increasingly refined. Street trees, as an important component of urban greening, play a vital role in beautifying the environment and improving air quality. However, traditional manual inspection methods suffer from inefficiency, limited coverage, and insufficient accuracy in detecting missing street trees. Although various automation technologies have attempted to address these issues, most solutions primarily rely on single-frame image recognition and static position comparison methods.

[0003] These methods face several challenges in practical applications. Specifically, existing automatic detection technologies typically require the pre-construction of a detailed tree distribution map, which not only increases the system's initialization cost but also makes it difficult for the system to adapt to changing or unrecorded situations. Due to the use of single-frame analysis, existing technologies cannot effectively handle information changes over continuous time, resulting in weak capabilities in handling dynamic scenarios such as vehicle movement and pedestrian crossings. Current detection algorithms often determine the presence of missing trees solely through a binary judgment of tree presence or absence, neglecting complex real-world factors such as occlusion, tree stump remnants, and unplantable areas, leading to a high false alarm rate.

[0004] Therefore, it is necessary to design a new method that can not only overcome the limitations of existing technologies, but also provide more efficient and accurate technical support for the daily maintenance of urban greening, thereby helping to improve the quality and efficiency of urban management. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for rapid and continuous detection of missing trees in roadside trees based on dynamic edge computing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a rapid and continuous detection method for missing trees in roadside trees based on dynamic edge computing, comprising:

[0007] It acquires images, GPS coordinates, and timestamps collected by edge computing devices installed on vehicles, and constructs spatiotemporal trajectories based on the collected data;

[0008] The image is preprocessed to identify and classify vegetation, municipal facilities and moving obstructions, and output feature information and confidence scores to obtain image recognition results;

[0009] Based on the image recognition results, roadside trees close to the lane lines are selected and their corresponding positions are predicted. Motion trajectories are constructed by associating data from multiple consecutive frames and calibrated in the coordinate system to generate a predicted region of interest, thus forming a roadside tree position prediction result.

[0010] Based on the predicted location results of the roadside trees, the missing trees in the green belt are determined and classified, and relevant event information is generated; the events include shading and missing trees; among them, missing trees include areas where planting is not allowed, and missing trees in tree pits or tree stumps.

[0011] The further technical solution is as follows: Based on the image recognition result, roadside trees close to the lane line are selected and their corresponding positions are predicted. A motion trajectory is constructed by associating multiple consecutive frames of data, and calibration is performed in the coordinate system to generate a predicted region of interest, thus forming a roadside tree position prediction result, including:

[0012] Based on the image recognition results, the green belt and the roadside trees inside the lane line are identified and selected, and the selected green belt and roadside trees are associated to obtain the associated target roadside trees and green belt.

[0013] The associated target street trees in continuous frame data are analyzed, a motion trajectory is established for each target street tree, and the motion trajectory is adjusted to the coordinate system to obtain the transformed motion trajectory data.

[0014] Based on the transformed motion trajectory data, the target roadside tree distribution baseline and expected spacing are fitted to obtain the distribution model;

[0015] Within the selected green belt, the roadside tree with the smallest ordinate is selected as the prediction starting point according to the distribution model, and a region of interest is generated based on the prediction starting point. The region of interest is then converted back to the original image coordinate system to obtain the prediction result of the roadside tree position.

[0016] The further technical solution is as follows: The image is preprocessed to identify and classify vegetation, municipal facilities, and moving obstructions, and feature information and confidence scores are output to obtain image recognition results, including:

[0017] The image is preprocessed to obtain a preprocessed image;

[0018] The vertical lines, bark features, and color and texture differences of ground depressions in the preprocessed image are analyzed to identify trees and their tree pits. The trees are then classified as street trees and tree stumps based on their height to obtain vegetation detection results.

[0019] The preprocessed image identifies fixed facilities in the green belt, identifies non-plantable areas, and distinguishes between the green belt and the driveway to obtain the bounding boxes and confidence scores for fixed facilities, non-plantable areas, green belts, and driveways.

[0020] Vehicles are identified and labeled based on their contours, wheels, and window features to obtain vehicle bounding boxes and confidence scores.

[0021] The image recognition result is formed by combining the detection results of vegetation, the corresponding bounding boxes and confidence scores of fixed facilities, the corresponding bounding boxes and confidence scores of non-plantable areas, the bounding boxes and confidence scores of green belts, the bounding boxes and confidence scores of lanes, and the bounding boxes and confidence scores of vehicles.

[0022] The further technical solution is as follows: The process of identifying fixed structures in the green belt of the preprocessed image, identifying non-plantable areas, and distinguishing between the green belt and the driveway, to obtain the bounding boxes and confidence scores for the fixed structures, the bounding boxes and confidence scores for the non-plantable areas, the bounding boxes and confidence scores for the green belts, and the bounding boxes and confidence scores for the driveway, includes:

[0023] The system uses contour and color features to identify fixed facilities in green belts, and uses text signs to identify non-planting areas and distinguish between green belts and driveways, so as to obtain the corresponding bounding boxes and confidence levels of fixed facilities, non-planting areas, green belts, and driveways.

[0024] The further technical solution is as follows: Based on the image recognition result, the green belt and roadside trees near the lane line are identified and selected, and the selected green belt and roadside trees are associated to obtain the associated target roadside trees and green belt, including:

[0025] The image recognition results are analyzed, and the target green belt is determined with the green belt closest to the lane line as the center.

[0026] In the target green belt, the target trees are identified as roadside trees by comparing the positions of the center points of the tree roots;

[0027] Associate the target green belt and the marked roadside trees to obtain the associated target roadside trees and green belt.

[0028] The further technical solution is as follows: Analyzing the associated target street trees in continuous frame data and establishing a motion trajectory for each target street tree includes:

[0029] Based on the associated target roadside trees and green belts, a multi-target tracking algorithm is applied to capture the movement trajectory of each target roadside tree within the corresponding green belt;

[0030] Based on spatial proximity and appearance similarity, a correlation is established between consecutive frames to update or create motion trajectory records for each target roadside tree, wherein the motion trajectory records include timestamps, coordinates, and status identifiers.

[0031] Adjust the motion trajectory to the coordinate system to obtain the transformed motion trajectory data.

[0032] The further technical solution is as follows: the target roadside tree distribution baseline and expected spacing are fitted based on the transformed motion trajectory data to obtain a distribution model, including:

[0033] Analyze the converted motion trajectory data of multiple target roadside trees within the same green belt, calculate the motion vector of the overall forward direction, orthogonally project the position coordinates of the target roadside trees onto the straight line defined by the motion vector, and fit a roadside tree distribution baseline based on the projection points to obtain the distribution baseline of the target roadside trees;

[0034] Select a distribution baseline that meets the requirements of lane distance, calculate the arc length distance between adjacent projection points, and obtain the statistical average value of the arc length distance to obtain the desired spacing;

[0035] The distribution model includes the target street tree distribution baseline and the desired spacing.

[0036] The further technical solution is as follows: Within the selected green belt, the roadside tree with the smallest ordinate is selected as the prediction starting point according to the distribution model, and a region of interest is generated based on the prediction starting point. The region of interest is then converted back to the original image coordinate system to obtain the roadside tree location prediction result, including:

[0037] Within the selected green belt, the roadside tree with the smallest ordinate is selected as the prediction starting point according to the distribution model. The expected spacing is determined by moving average along the distribution baseline of the target roadside tree based on the prediction starting point.

[0038] In the coordinate system, a prediction interest area is created with the predicted point as the center, and the vertical axis is a number of times the standard deviation of the spacing. If there is only one distribution baseline, the width of the green belt is set as the dimension in the horizontal axis. If there are multiple distribution baselines, the intersection of the boundary line of the target green belt closest to the lane line is used as one of the vertices, and the center line of the target roadside tree distribution baseline and the distribution baseline closest to the target roadside tree distribution baseline is calculated as another vertex to define the prediction interest area in the case of multiple baselines.

[0039] The predicted region of interest is inversely transformed from the coordinate system to the original image coordinate system, and the predicted region of interest and its confidence score are output.

[0040] The further technical solution is as follows: Based on the comprehensive prediction results of the roadside tree locations, the missing tree situation in the green belt is determined and classified, and relevant event information is generated, including:

[0041] Based on the predicted location results of the roadside trees, if an "unplantable" marker is detected but no tree characteristics are found, and its continued existence is confirmed through tracking prediction, it is marked as an "unplantable" area and details are recorded. If a moving obstruction is identified without tree characteristics, and if a stable tracking sequence exceeds the required number of frames, it is considered an obstruction and an event record is generated. If a tree pit is detected but not marked as an "unplantable" area, and its existence is confirmed in the required number of tracking frames, it is defined as a missing tree pit and information is recorded. If a tree stump is detected but not in an "unplantable" area, and its existence is confirmed in tracking frames exceeding a set threshold, it is defined as a missing tree stump and corresponding data is recorded.

[0042] This invention also provides a rapid and continuous detection system for missing trees in roadside trees based on dynamic edge computing, comprising:

[0043] The acquisition unit is used to acquire images, GPS coordinates, and timestamps collected by the edge computing device installed on the vehicle, and to construct a spatiotemporal trajectory based on the collected data;

[0044] The recognition unit is used to preprocess the image, identify and classify vegetation, municipal facilities and moving obstruction targets, and output feature information and confidence score to obtain image recognition results.

[0045] The prediction unit is used to filter roadside trees close to the lane line and predict their corresponding positions based on the image recognition results. It constructs a motion trajectory by associating data from multiple consecutive frames and calibrates it in the coordinate system to generate a prediction interest area, thus forming a roadside tree position prediction result.

[0046] The comprehensive classification unit is used to determine and classify the missing trees in the green belt based on the predicted location of the roadside trees, and generate relevant event information; the events include shading and missing trees; among them, missing trees include areas where planting is not allowed, and missing trees in tree pits or tree stumps.

[0047] The advantages of this invention compared to existing technologies are as follows: This invention uses an edge computing device installed on a vehicle to collect images, GPS coordinates, and timestamp data in real time and constructs a spatiotemporal trajectory. The images are preprocessed to identify and classify vegetation, municipal facilities, and moving obstructions, outputting feature information and confidence scores to obtain accurate image recognition results. Based on these results, roadside trees near the lane lines are selected, their locations are predicted, and a predicted area of ​​interest is generated through correlation and calibration of continuous multi-frame data, forming an accurate prediction of roadside tree locations. This method further combines the roadside tree location prediction results to intelligently determine and classify missing trees in green belts, including areas where planting is not possible, obstructions, and missing tree pits or stumps, generating relevant event information. This method not only overcomes the limitations of traditional technologies caused by low efficiency and large errors due to manual inspection, but also provides more efficient and accurate technical support for the daily maintenance of urban greening, greatly improving the quality and efficiency of urban management and realizing the intelligent upgrade of urban greening maintenance work.

[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the rapid and continuous detection method for missing trees in roadside trees based on dynamic edge computing, provided in an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of an image acquired by an edge computing device according to an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of image recognition results provided in an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of a tracking frame provided in an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of the target object in the tree pit provided in an embodiment of the present invention;

[0055] Figure 6 A schematic diagram of the converted motion trajectory data provided in an embodiment of the present invention;

[0056] Figure 7A schematic block diagram of a rapid and continuous detection system for missing roadside trees based on dynamic edge computing provided in an embodiment of the present invention;

[0057] Figure 8 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0060] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0061] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0062] Please see Figure 1 , Figure 1This is a schematic flowchart illustrating a rapid and continuous detection method for missing street trees based on dynamic edge computing, provided in an embodiment of the present invention. This method is applied in a server. The server interacts with edge computing devices, integrating image, GPS coordinates, and timestamp data collected by edge computing devices on vehicles. Advanced image processing techniques are used to identify and classify vegetation, municipal facilities, and moving obstructions in green belts, constructing a motion trajectory and distribution model of the target street trees. A predicted area of ​​interest is then generated and calibrated in the coordinate system to assess the situation of missing trees in the green belt. This method not only overcomes the limitations of existing technologies in detecting missing street trees in complex urban environments, such as inaccuracy and low efficiency, but also updates and records missing tree event information in real time. It provides efficient and accurate technical support for the daily maintenance of urban greening, greatly improving the quality and efficiency of urban management and ensuring the health and aesthetics of the urban green environment. Furthermore, by dynamically adjusting and optimizing the detection process to adapt to the needs of different scenarios, the system's flexibility and practicality are further enhanced.

[0063] Figure 2 This is a flowchart illustrating the rapid and continuous detection method for missing trees in roadside trees based on dynamic edge computing, provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S140.

[0064] S110. Acquire images, GPS coordinates, and timestamps collected by the edge computing device installed on the vehicle, and construct a spatiotemporal trajectory based on the collected data.

[0065] In this embodiment, spatiotemporal trajectory refers to a sequence constructed chronologically by integrating the vehicle's high-precision GPS coordinates, timestamps, and corresponding image data, reflecting the vehicle's driving path and changes in its surrounding environment. This sentence summarizes the core concept of spatiotemporal trajectory: it is a data sequence combining temporal and spatial information, used to track and analyze the dynamic changes in the surrounding environment during vehicle movement.

[0066] High-resolution cameras (e.g., 3840×2160 pixels) mounted on vehicles capture images of the green belts along the roads in real time. These images not only contain information about the roadside trees, but may also include other targets such as municipal facilities and moving obstructions.

[0067] By using a high-precision GPS module in conjunction with an onboard inertial measurement unit (IMU), the vehicle's position (latitude and longitude coordinates), speed, and heading angle are accurately recorded. This allows each image frame to correspond to a specific geographical location.

[0068] Each image and each set of GPS data is accompanied by a precise timestamp to ensure that all data can be synchronized and correlated over time.

[0069] The collected data is packaged in real time, forming data packets containing images, location information, and timestamps. A spatial index is also created for these data packets to facilitate rapid retrieval and association later.

[0070] Based on the aforementioned data packets, the system can construct one or more spatiotemporal trajectory sequences. Specifically, this involves connecting a series of timestamped geographical locations in chronological order to form a spatiotemporal representation of the vehicle's travel path. This trajectory not only reflects the vehicle's movement but can also provide information on the dynamic changes in the roadside environment by associating images at corresponding time points.

[0071] This process is crucial for subsequent functions such as image recognition, target classification, and missing tree detection. By accurately locating the spatial position of each frame of the image, the system can more accurately identify and track roadside trees and other targets in the green belt, thereby improving the accuracy and reliability of missing tree detection.

[0072] In summary, step S110 is not only a data collection process, but also a crucial step in transforming physical world information into processable digital information, providing a solid foundation for subsequent image processing, target recognition, behavior analysis, and other tasks.

[0073] S120. The image is preprocessed to identify and classify vegetation, municipal facilities and moving obstructions, and output feature information and confidence scores to obtain image recognition results.

[0074] In this embodiment, the image recognition result refers to the classification information, feature description, and confidence score of vegetation, municipal facilities, and moving obstruction targets obtained after a series of processing and analysis steps on the acquired high-resolution image.

[0075] In one embodiment, step S120 described above may include steps S121 to S125.

[0076] S121. The image is preprocessed to obtain a preprocessed image.

[0077] In this embodiment, the high-resolution image of 3840×2160 pixels is first normalized to reduce the influence of external factors such as lighting and contrast, and to provide standardized input for subsequent models.

[0078] S122. Analyze the vertical lines, bark features, and color and texture differences of ground depressions in the preprocessed image to identify trees and their tree pits, and classify trees as street trees and tree stumps according to height to obtain vegetation detection results.

[0079] In this embodiment, the preprocessed image is further analyzed using a trained vegetation detection model. This model can identify vertical bar structures, bark color, and texture features in the image, and thereby identify trees and their planting holes. Based on tree height, trees can also be classified as street trees or tree stumps. This step yields vegetation detection results including trunk bounding boxes, center point lines, root center points, height, and confidence scores.

[0080] S123. Identify the fixed facilities in the green belt of the preprocessed image, identify the non-plantable area and distinguish between the green belt and the driveway, so as to obtain the bounding box and confidence level of the fixed facilities, the bounding box and confidence level of the non-plantable area, the bounding box and confidence level of the green belt, and the bounding box and confidence level of the driveway.

[0081] In this embodiment, the fixed facilities in the green belt are identified by contour and color features, and the non-plantable areas are identified by text signs and the green belt and the driveway are distinguished, so as to obtain the bounding boxes and confidence levels of the fixed facilities, the bounding boxes and confidence levels of the non-plantable areas, the bounding boxes and confidence levels of the green belts, and the bounding boxes and confidence levels of the driveways.

[0082] In this phase, a municipal infrastructure detection model is used to identify fixed facilities within green belts (such as manhole covers, electrical boxes, and fire hydrants) and mark areas where planting is prohibited. Furthermore, it can distinguish between green belts and driveways. Through contour, color feature, and semantic matching technology using text signs, the location of various objects is determined, and corresponding bounding boxes and confidence scores are output.

[0083] S124. Identify and label vehicles based on their contours, wheels, and window features to obtain vehicle bounding boxes and confidence levels.

[0084] In this embodiment, a moving occlusion detection model is used to identify vehicles based on their rectangular outline, wheel and window features, and to provide vehicle bounding boxes and confidence scores.

[0085] S125. Combine the detection results of vegetation, the corresponding bounding boxes and confidence scores of fixed facilities, the corresponding bounding boxes and confidence scores of non-plantable areas, the bounding boxes and confidence scores of green belts, the bounding boxes and confidence scores of lanes, and the bounding boxes and confidence scores of vehicles to form image recognition results.

[0086] Finally, all the information obtained in the above steps—including vegetation detection results, bounding boxes and their confidence scores for fixed facilities, bounding boxes and their confidence scores for non-plantable areas, bounding boxes and their confidence scores for green belts, bounding boxes and their confidence scores for lanes, and bounding boxes and their confidence scores for vehicles—is combined to form the final image recognition result. This information collectively constitutes a comprehensive and detailed description of vegetation, municipal facilities, and moving obstructions within a specific area, providing a solid data foundation for subsequent business logic such as missing plant event determination and obstruction determination.

[0087] In this embodiment, the acquired high-resolution images are classified and identified to extract information about vegetation, municipal facilities, and moving obstructions. First, all input images undergo a common preprocessing stage. This stage receives the raw images from the data acquisition module and normalizes them to eliminate the influence of external factors such as changes in light intensity and contrast differences on subsequent analysis, thereby providing standardized input for the vegetation detection model and the municipal facility detection model.

[0088] In the model recognition phase, three specially designed deep learning models were employed to perform different tasks. The first is a vegetation detection model, which focuses on the identification and classification of tree features. This model determines the presence and specific attributes of trees by analyzing the approximate vertical line structure in the image, combined with bark color and texture features. The model can output the bounding box of the trunk, the line connecting the center points, the root center point, the height, and a confidence score; based on the tree's height, it further classifies it as a street tree or a tree stump. In addition, this model can also identify irregular depressions in the ground area—i.e., tree pits—and performs identification based on the significant differences between the soil's color and texture features and the surrounding paved surface, ultimately outputting the bounding box and confidence score of the tree pit.

[0089] Secondly, there's the municipal facility detection model, primarily used to identify the types of fixed facilities within green belts, such as manhole covers, electrical boxes, and fire hydrants. By analyzing the common contour and color features of these objects, the model can accurately locate and output the corresponding bounding boxes and confidence scores. In addition, the model can also identify signs such as "Buried Cables," "Gas Pipelines," and "No Planting," performing semantic matching for these specific types of non-planting areas and providing corresponding bounding boxes and confidence scores. Simultaneously, the model can distinguish between green belts and driveways, accurately separating them from the image using color, texture, and line information, and outputting their respective bounding boxes and confidence scores.

[0090] Finally, there's the moving occlusion detection model, used for vehicle identification. While not focusing on specific vehicle categories, the model can determine a vehicle's presence by analyzing features such as its rectangular outline, wheels, and windows, outputting the vehicle's bounding box and confidence score. This function is crucial for determining whether trees are missing due to vehicle occlusion. Through the collaborative work of these three models, the system can comprehensively and accurately analyze the acquired images, providing solid data support for subsequent business logic processing.

[0091] S130. Based on the image recognition results, select roadside trees close to the lane lines and predict their corresponding positions. Construct motion trajectories by associating data from multiple consecutive frames and calibrate them in the coordinate system to generate a predicted region of interest, thus forming a roadside tree position prediction result.

[0092] In this embodiment, the predicted location of street trees refers to the precise location of each street tree in the world coordinate system, determined by analyzing image recognition results, establishing and calibrating the movement trajectory of the target street trees. Furthermore, it includes regions of interest (ROIs) generated based on a distribution model, which identify locations where street trees may be planted, and typically includes a confidence score to indicate the accuracy of the prediction.

[0093] This step aims to use image recognition technology to select the green belt near the lane lines and the roadside trees within it. Next, by analyzing data from consecutive frames, a trajectory is established for each target roadside tree, and these trajectories are transformed into a world coordinate system, ultimately generating a predicted region of interest (ROI), thereby achieving accurate prediction of the roadside tree's location.

[0094] In one embodiment, step S130 described above may include steps S131 to S134.

[0095] S131. Based on the image recognition results, identify and select the green belt and roadside trees near the lane line, and associate the selected green belt and roadside trees to obtain the associated target roadside trees and green belt.

[0096] In one embodiment, step S131 described above may include steps S1311 to S1313.

[0097] S1311. Analyze the image recognition results and determine the target green belt as the center of the green belt near the lane line.

[0098] In this embodiment, the target green belt refers to a specific green area near the lane lines that may be planted with roadside trees. During image processing, the boundaries of the green belts on both sides of the road are first identified. Then, based on certain criteria (such as distance from the lane lines, vegetation type, etc.), green belts that meet the criteria are selected as objects for further analysis. Therefore, the "target green belt" is the green area that receives focused attention after initial screening.

[0099] S1312. In the target green belt, by comparing the position of the center point of the tree root, the target trees as roadside trees are identified, so as to obtain the target roadside trees.

[0100] In this embodiment, target street trees refer to trees located within the target green belt that meet specific conditions (e.g., height greater than 50cm). These trees are considered key targets of the system, and their location, status, and other information will be recorded and analyzed in detail. Through in-depth processing of image data, the specific location and characteristics of each target street tree can be accurately identified.

[0101] S1313, associate the target green belt and the marked roadside trees to obtain the associated target roadside trees and green belt.

[0102] First, the image recognition results need to be carefully analyzed to determine which green belts are close to the lane lines and define them as target green belts. Target green belts refer to specific green areas located on both sides of the road that may be planted with roadside trees. This process typically involves image processing techniques such as edge detection and color segmentation to accurately define the boundary between the green belt and the lane lines.

[0103] Next, within the identified target greenbelt, street trees are further identified by comparing the positions of the center points of the tree roots. Target street trees are those that meet certain conditions (such as being taller than 50cm) and are located within the target greenbelt. This step may involve using machine learning models or feature matching algorithms to accurately identify the specific location of each tree.

[0104] Finally, the identified target green belts are associated with their internal roadside trees to ensure a clear one-to-one correspondence between each green belt and its internal roadside trees. This association not only helps with subsequent trajectory tracking but also improves the robustness and accuracy of the overall system.

[0105] S132. Analyze the associated target street trees in the continuous frame data, establish motion trajectories for each target street tree, and adjust the motion trajectories to the coordinate system to obtain the transformed motion trajectory data.

[0106] In this embodiment, the transformed motion trajectory data refers to the motion trajectory of the roadside trees, originally existing in the image coordinate system, which has been mapped to the world coordinate system after coordinate transformation. This process eliminates errors caused by factors such as camera perspective and distortion, allowing positional changes between different frames to be compared and analyzed on a real-world scale. The transformed data includes information such as timestamps, coordinate positions, and status identifiers.

[0107] In one embodiment, step S132 described above may include steps S1321 to S1323.

[0108] S1321. Based on the associated target roadside trees and green belts, a multi-target tracking algorithm is applied to capture the movement trajectory of each target roadside tree within the corresponding green belt.

[0109] In this embodiment, the motion trajectory refers to the record of the change in the spatial position of each target street tree within its corresponding green belt over time. These changes are captured by a multi-target tracking algorithm to form a continuous path. The motion trajectory not only describes the movement of the trees but may also include dynamic characteristics such as velocity and acceleration. It is the foundation for constructing the distribution model and is crucial for understanding the layout patterns of street trees.

[0110] S1322. Based on spatial proximity and appearance similarity, establish the association between consecutive frames and update or create the motion trajectory record of each target road tree, wherein the motion trajectory record includes timestamp, coordinates and status identifier.

[0111] In this embodiment, a timestamp refers to the precise point in time when a specific event occurred, typically expressed as a date and time (e.g., 2025-09-12T14:46:00Z). When tracking the movement of target roadside trees, timestamps are used to mark the time when each location data point was collected. It provides the chronological order of events, allowing for the analysis of the target's behavioral patterns or trends in chronological order. While roadside trees themselves do not move actively, data collection at different time periods helps understand whether the trees' positions have changed due to external factors (such as construction, natural growth, etc.).

[0112] Coordinates refer to the specific location of the target street tree in a coordinate system. This could be an image coordinate system (pixel coordinates), a camera coordinate system, or a world coordinate system. Transformed motion trajectory data typically uses the world coordinate system to more accurately reflect positional relationships in the real world. Coordinates allow for precise determination of the target street tree's spatial location. This is crucial for analyzing relative distances between trees, their layout patterns, and predicting future planting or removal plans.

[0113] A status indicator describes a target street tree's attribute or condition at a specific point in time. It can be a simple flag (e.g., present / absent) or contain more detailed information (e.g., health status, maintenance required). Status indicators help understand the specific condition of the target street tree at different times. For example, if a tree's status indicator shows "needs pruning," relevant departments can take appropriate measures based on this information. Furthermore, status indicators can also be used to mark trees for any abnormalities (e.g., pests, diseases, risk of falling), allowing for timely intervention.

[0114] S1323. Adjust the motion trajectory to the coordinate system to obtain the converted motion trajectory data.

[0115] In this embodiment, a multi-target tracking algorithm is applied to capture the motion trajectory of each target roadside tree within its corresponding green belt. The motion trajectory refers to the record of the spatial position change of each tree at different points in time. This typically involves advanced algorithms such as Kalman filters and particle filters, used to stably track the positional changes of each tree between consecutive frames.

[0116] Based on spatial proximity and visual similarity, a correlation is established between consecutive frames to update or create motion trajectory records for each target roadside tree. These records include timestamps, coordinates, and status identifiers. For example, if a tree does not move significantly between adjacent frames, it can be considered to have maintained the same state; otherwise, its trajectory record needs to be updated.

[0117] The obtained motion trajectory is then adjusted to the world coordinate system to obtain the transformed motion trajectory data. This means that data originally existing in the image coordinate system needs to be mapped to the real-world coordinate system using information such as camera parameters to eliminate the influence of factors such as perspective distortion.

[0118] S133. Based on the transformed motion trajectory data, fit the target roadside tree distribution baseline and expected spacing to obtain the distribution model.

[0119] In one embodiment, step S133 described above may include steps S1331 to S1332.

[0120] S1331. Analyze the converted motion trajectory data of multiple target roadside trees in the same green belt, calculate the motion vector of the overall forward direction, orthogonally project the position coordinates of the target roadside trees onto the straight line defined by the motion vector, and fit the roadside tree distribution baseline based on the projection points to obtain the distribution baseline of the target roadside trees.

[0121] S1332. Select a distribution baseline that meets the requirements of the lane distance, calculate the arc length distance between adjacent projection points, and obtain the statistical average value of the arc length distance to obtain the desired spacing.

[0122] The distribution model includes the target street tree distribution baseline and the desired spacing.

[0123] In urban greening management or similar applications, distribution models are important tools for guiding tree planting, maintenance, and monitoring. The distribution model mainly includes two core elements: the baseline distribution of target street trees and the desired spacing.

[0124] A distribution baseline is a virtual straight line that follows the arrangement of street trees along a road or other designated area. This line is usually determined based on the road edge, sidewalk boundary, or pre-planned green belt layout, and serves as a basic reference line for tree arrangement.

[0125] The desired spacing refers to the ideal distance between two adjacent trees. This parameter is set based on a variety of factors, including but not limited to tree species characteristics, growth rate, environmental conditions, and aesthetic considerations.

[0126] In practice, combining distribution baselines and desired spacing allows for effective management and optimized allocation of street trees. For example, during the planning phase of new tree planting projects, detailed planting plans are developed based on these two indicators; and in subsequent maintenance, this information guides daily inspections, problem diagnosis, and necessary adjustments.

[0127] The movement trajectory data of multiple target roadside trees within the same green belt are analyzed to calculate the overall movement vector of the direction of movement. A baseline for the distribution of roadside trees is then fitted based on points orthogonally projected onto the line defining this vector. This process involves complex geometric transformations and statistical methods, aiming to find a straight line that best represents the arrangement patterns of all roadside trees.

[0128] After selecting a suitable distribution baseline, the arc distance between adjacent projection points is calculated, and the average value is taken to obtain the desired spacing. The distribution model includes not only the distribution baseline of the target street trees but also the average spacing between them. This step is crucial for understanding the overall layout of the street trees.

[0129] S134. Within the selected green belt, the roadside tree with the smallest ordinate is selected as the prediction starting point according to the distribution model, and a region of interest is generated based on the prediction starting point. The region of interest is then converted back to the original image coordinate system to obtain the prediction result of the roadside tree position.

[0130] In one embodiment, step S134 described above may include steps S1341 to S1343.

[0131] S1341. Within the selected green belt, the roadside tree with the smallest ordinate is selected as the prediction starting point according to the distribution model. The expected spacing is determined by moving average along the distribution baseline of the target roadside tree based on the prediction starting point.

[0132] S1342. In the coordinate system, with the predicted point as the center, create a predicted interest area with a vertical axis that is several times the standard deviation of the spacing. If there is only one distribution baseline, the width of the green belt is set as the dimension in the horizontal axis direction. If there are multiple distribution baselines, the intersection of the boundary line of the target green belt closest to the lane line is used as one of the vertices, and the center line of the target roadside tree distribution baseline and the distribution baseline closest to the target roadside tree distribution baseline is calculated as another vertex to define the predicted interest area in the case of multiple baselines.

[0133] In this embodiment, the nearest principle means that the distance is less than a set threshold.

[0134] S1343. Transform the predicted region of interest from the coordinate system to the original image coordinate system, and output the predicted region of interest and its confidence score.

[0135] The tree with the smallest ordinate in the distribution model is selected as the starting point for prediction, and the world coordinates of the prediction points are determined along the distribution baseline according to the expected spacing. This step helps the system find a reasonable starting point for subsequent prediction work.

[0136] In the coordinate system, a prediction interest area of ​​a certain size is created with the prediction point as the center. The specific size depends on the actual situation. For example, if there is only one distribution baseline, the width of the green belt may be considered; if there are multiple distribution baselines, a more complex method is required to define the prediction interest area.

[0137] Finally, the predicted region of interest is inversely transformed from the original image coordinate system to the original image coordinate system, and the predicted region of interest and its confidence score are output. This not only directly marks the potential locations of street trees on the original image, providing an intuitive visual verification basis, but also evaluates the reliability of the prediction results.

[0138] In this embodiment, step S130 first determines which trees belong to the target green belt and are close to the lane lines. By analyzing each frame of the image, the system first identifies all possible green belts and lanes, and then determines whether these green belts are located on both sides of the lane. Next, the system compares the position of the trees within the green belt (i.e., the center point of the tree roots) with the position of the green belt. If a tree is found to be located within the green belt, it is marked as a target roadside tree.

[0139] After identifying the target trees, the next step is to track them to capture their positional changes at different points in time. This typically involves the application of multi-target tracking algorithms, which use spatial proximity and appearance similarity to match the same tree in consecutive frames. The motion trajectory of each tree is recorded, including information such as timestamps, spatial coordinates, and status indicators.

[0140] To accurately represent the location of trees, the system needs to convert the pixel coordinates obtained from the image to coordinates in the world coordinate system. This step uses an inverse perspective transformation model to eliminate perspective distortion in the image, thereby obtaining more accurate tree location information. The origin of the coordinate system is set as the horizontal projection point of the vehicle at the corresponding position in the current frame image, and the X-axis and Y-axis represent lateral and longitudinal displacements, respectively.

[0141] Based on the data obtained in the previous step, the system attempts to abstract the distribution patterns of the roadside trees, such as baselines and desired spacing. This includes calculating the average displacement vectors of multiple trees to fit a motion vector V representing the overall direction of movement, and using this vector to determine the arrangement of the roadside trees. Furthermore, the system also calculates the average distance D between adjacent trees. avg As the desired physical distance.

[0142] Next, the most suitable prediction starting point is selected. Specifically, the tree with the smallest coordinate value (i.e., closest to the vehicle) in the target green belt is chosen as the prediction reference point. This selection is to ensure the reliability of the prediction starting point.

[0143] Finally, the system predicts its location based on the information obtained from the above steps. It first moves a fixed distance D along the distribution baseline L. avg Find the next expected location P where the tree will appear. predbev Then, a region of interest (ROI) is defined in the world coordinate system, the size of which is based on the standard deviation σ of the spacing between trees and the width of the green belt. Finally, this ROI is transformed back to the original image coordinate system for subsequent validation.

[0144] The entire process combines computer vision technology, multi-object tracking algorithms, and mathematical modeling, aiming to effectively identify and predict the location of roadside trees, which is of great significance for urban planning and road maintenance.

[0145] First, determine P predbev That is, the predicted center point in the coordinate system.

[0146] The standard deviation σ is calculated based on the data distribution, and twice this deviation, 2σ, is used as the dimension of the ROI along the Y-axis. This ensures that the ROI encompasses most of the data distribution, improving prediction accuracy.

[0147] If there is only one distribution baseline L, then the width of the green belt is used to determine the width of the X-axis. This is because green belts typically provide relatively fixed boundaries, making it easier to define the lateral extent of the ROI.

[0148] At this point, the two vertices are determined by the boundaries on both sides of the green belt, thus forming a rectangular area.

[0149] If there are multiple distribution baselines L, then the intersection of the boundary line of the target green belt that is closest to the lane line needs to be found as one of the vertices.

[0150] The other vertex is determined by calculating the centerline position of the target street tree distribution baseline and the distribution baseline closest to the target street tree distribution baseline.

[0151] In this way, even in complex environments, the shape and position of the ROI can be flexibly adjusted according to the actual road structure.

[0152] After completing the above steps in the coordinate system, a specific ROI region is obtained.

[0153] Using inverse perspective mapping, this ROI region is transformed from the BEV coordinate system back to the original image coordinate system. This step is crucial because it maps the analysis results based on the simplified model back to the actual scene, facilitating subsequent processing and understanding.

[0154] The final output includes the specific location and shape of the ROI in the original image coordinate system. This information can be used for further object detection, tracking, or other computer vision tasks.

[0155] This approach not only allows for precise definition of the region of interest but also effectively adapts to varying environmental conditions and layout changes, significantly enhancing the system's robustness and usability. This method is particularly suitable for perception systems in autonomous vehicles, helping to accurately identify and respond to the surrounding environment.

[0156] S140. Based on the predicted location results of the roadside trees, determine and classify the missing trees in the green belt, and generate relevant event information; the events include shading and missing trees; among them, missing trees include missing trees in unplantable areas, tree pits or tree stumps.

[0157] In this embodiment, based on the predicted location results of the roadside trees, if an unplantable marker is detected but no tree characteristics are found, and the continued existence is confirmed through tracking prediction, it is marked as an unplantable area and details are recorded. If a moving obstruction is identified without tree characteristics, and if the number of stable tracking frames exceeds the required number, it is considered an obstruction and an event record is generated. If a tree pit is detected but not marked as an unplantable area, and its existence is confirmed in the required number of tracking frames, it is defined as a missing tree pit and information is recorded. If a tree stump is detected but not in an unplantable area, and its existence is confirmed in tracking frames exceeding a set threshold, it is defined as a missing tree stump and corresponding data is recorded.

[0158] In this embodiment, this step can identify areas where trees cannot be planted due to obstacles such as manhole covers, electrical boxes, fire hydrants, or "no planting" signs.

[0159] When one of the above targets is detected within the ROI region and its average detection confidence is higher than 0.5, if no tree features are detected at the corresponding location, the non-plantable area missing tree determination is initiated.

[0160] Based on the green belt distribution baseline L and the current target's location, the system will predict the target's location in subsequent frames and incorporate these predicted locations into the tracking sequence.

[0161] If more than 60% of the tracking frames confirm the presence of the target at the same location, it is marked as a non-plantable area. The system will generate a corresponding event log, including the target ID, geographic coordinates, evidence, timestamp, and confidence assessment.

[0162] It can also identify occlusion caused by moving objects, which may affect the normal growth of trees or monitoring.

[0163] When a moving obstruction (such as a vehicle, temporary building, etc.) is detected and its detection confidence exceeds 0.5, but no tree features are detected at the corresponding location, an occlusion determination is triggered.

[0164] Using a similar method, the system tracks the state changes at that location in subsequent frames, based on the distribution baseline L and the target predicted location.

[0165] If an obstruction is consistently detected in more than 70% of the tracking frames, and there are no tree features at the corresponding location, it is considered an obstruction. The system will record the obstruction event, including the obstruction ID, geographic coordinates, predicted tree location, evidence, timestamp, and confidence assessment.

[0166] In addition, it can also identify missing trees due to empty tree pits or remaining tree stumps.

[0167] Within the ROI area, if a tree pit or stump is detected and its average detection confidence is higher than 0.5, and it is confirmed that the area is not an unplantable area, then a missing tree pit or stump determination is triggered.

[0168] If more than 60% of the tracking frames confirm the existence of the "tree hole" target, a tree hole missing tree determination is made and a corresponding event record is generated.

[0169] If more than 70% of the tracking frames verify the existence of the "tree stump" target, a tree stump missing determination is made, and a detailed event log is also generated.

[0170] Each judgment method employs precise algorithms to ensure the accuracy and reliability of the results. By combining real-time data processing, machine learning models, and meticulous post-processing steps, it can effectively identify various types of missing plants, providing strong support for urban greening management.

[0171] For example, firstly, Figure 2 The raw images captured by the edge computing device are presented, forming the basis for subsequent analysis. Next, Figure 3 The image recognition results extracted after processing are shown, including the main lane and its confidence score, the target green belts on both sides of the lane and their confidence scores, the corresponding trees in the target green belts and their confidence scores, and the tree pits and their confidence scores. It is worth noting that... Figure 3 In the diagram, the green belt on the left is excluded from the evaluation due to the lack of tree features; while the green belt on the right contains the target tree and is therefore included in the evaluation process. Based on this, the system generates a red dashed rectangle as the predicted Region of Interest (ROI) for the next expected tree. If a tree pit appears within this ROI, the logic for determining missing tree pits / tree stumps will be triggered.

[0172] Next, based on the distribution baseline L of the tree arrangement and the current tree pit location, the system predicts possible locations in subsequent frames and includes subsequent frames containing their predicted locations in the tracking frame sequence. In this example, the tracking frame sequence contains only one image, namely... Figure 4 A "tree pit" target was detected at the predicted location in the tracking frame. Since tree pit targets (such as...) were identified in 100% of the images throughout the entire tracking frame sequence... Figure 5 As shown in the figure, this proportion exceeded the set value of 60%, so it was finally determined that the tree pit was missing a tree.

[0173] also, Figure 3 The detection results, which are incorporated into the calculations from the image perspective, are shown. Figure 6These results are then mapped to a world coordinate system centered on the vehicle using inverse perspective transformation (IPM). In this view, the system first calculates the distribution baseline (L) of the tree arrangement, and then calculates the average physical spacing (D). avg Then, the tree whose coordinates lie on the distribution baseline and whose Y-coordinate value is the smallest is selected as the most suitable prediction baseline. Based on this prediction baseline, the tree is moved along the distribution baseline L by a desired distance D. avg The distance is used to obtain the world coordinates P of the predicted point. predbev Based on this, a rectangular region is created in the world coordinate system as the region of interest (ROI) for key validation, with its center point being P. predbev The magnitude of the Y-axis is twice the standard deviation of the spacing σ, and the magnitude of the X-axis is the width of the green belt. Finally, this ROI region is transformed back to the original image coordinate system for further analysis and verification.

[0174] Through the above steps, the system can effectively predict and verify the dynamic status of street trees, ensuring the health and integrity of urban greening. This technology not only improves maintenance efficiency but also reduces errors from manual inspections, providing strong support for urban management.

[0175] The aforementioned rapid and continuous detection method for missing roadside trees based on dynamic edge computing acquires images, GPS coordinates, and timestamp data in real time using edge computing devices installed on vehicles, and constructs spatiotemporal trajectories. The images are preprocessed to identify and classify vegetation, municipal facilities, and moving obstructions, outputting feature information and confidence scores to obtain accurate image recognition results. Based on these results, roadside trees near the lane lines are selected, their locations are predicted, and a predicted region of interest is generated through correlation and calibration of continuous multi-frame data, forming an accurate prediction of roadside tree locations. This method further combines the roadside tree location prediction results to intelligently determine and classify missing trees in green belts, including non-planting areas, obstructions, and missing trees in tree pits or stumps, generating relevant event information. This method not only overcomes the limitations of traditional technologies caused by low efficiency and large errors due to manual inspection, but also provides more efficient and accurate technical support for the daily maintenance of urban greening, greatly improving the quality and efficiency of urban management and realizing the intelligent upgrade of urban greening maintenance work.

[0176] Figure 7 This is a schematic block diagram of a rapid and continuous detection system 300 for missing roadside trees based on dynamic edge computing, provided in an embodiment of the present invention. Figure 7As shown, corresponding to the above-described method for rapid and continuous detection of missing street trees based on dynamic edge computing, this invention also provides a system 300 for rapid and continuous detection of missing street trees based on dynamic edge computing. This system 300 includes a unit for executing the above-described method for rapid and continuous detection of missing street trees based on dynamic edge computing, and the system can be configured in a server. Specifically, please refer to... Figure 7 The rapid and continuous detection system for missing trees in roadside trees based on dynamic edge computing 300 includes an acquisition unit 301, an identification unit 302, a prediction unit 303, and a comprehensive classification unit 304.

[0177] The acquisition unit 301 is used to acquire images, GPS coordinates, and timestamps collected by the edge computing device installed on the vehicle, and construct a spatiotemporal trajectory based on the acquired data; the recognition unit 302 is used to preprocess the images, identify and classify vegetation, municipal facilities, and moving obstructions, and output feature information and confidence scores to obtain image recognition results; the prediction unit 303 is used to select roadside trees close to the lane lines and predict their corresponding positions based on the image recognition results, construct motion trajectories by associating continuous multi-frame data, and calibrate them in the coordinate system to generate a predicted area of ​​interest, forming a roadside tree position prediction result; the comprehensive classification unit 304 is used to comprehensively determine and classify the missing trees in the green belt based on the roadside tree position prediction results, and generate relevant event information; the events include obstruction and missing trees; among which, missing trees include areas where planting is not possible, tree pits, or tree stumps.

[0178] In one embodiment, the prediction unit 303 includes:

[0179] The system comprises the following sub-units: an identification and association sub-unit, used to identify and select the green belt and roadside trees near the lane line based on the image recognition results, and associate the selected green belt and roadside trees to obtain the associated target roadside trees and green belt; an analysis and transformation sub-unit, used to analyze the associated target roadside trees in continuous frame data, establish a motion trajectory for each target roadside tree, and adjust the motion trajectory to a coordinate system to obtain transformed motion trajectory data; a fitting sub-unit, used to fit the target roadside tree distribution baseline and expected spacing based on the transformed motion trajectory data to obtain a distribution model; and a prediction sub-unit, used to select the roadside tree with the smallest ordinate within the selected green belt according to the distribution model as the prediction starting point, generate a region of interest based on the prediction starting point, and transform the region of interest back to the original image coordinate system to obtain the predicted roadside tree position.

[0180] In one embodiment, the identification unit 302 includes:

[0181] The preprocessing subunit is used to preprocess the image to obtain a preprocessed image; the vegetation detection subunit is used to analyze the vertical lines, bark features, and color and texture differences of ground depressions in the preprocessed image to identify trees and their tree pits, and classify trees into street trees and tree stumps according to height to obtain vegetation detection results; the facility detection subunit is used to identify fixed facilities in the green belt of the preprocessed image, identify non-plantable areas, and distinguish between green belts and driveways to obtain the corresponding bounding boxes of fixed facilities. The system includes a confidence level, a bounding box and confidence level for non-plantable areas, a bounding box and confidence level for green belts, and a bounding box and confidence level for lanes; a vehicle detection subunit, used to identify and label vehicles based on their contours, wheels, and window features to obtain vehicle bounding boxes and confidence levels; and a combination subunit, used to combine vegetation detection results, bounding boxes and confidence levels for fixed facilities, bounding boxes and confidence levels for non-plantable areas, bounding boxes and confidence levels for green belts, bounding boxes and confidence levels for lanes, and bounding boxes and confidence levels for vehicles to form image recognition results.

[0182] In one embodiment, the facility detection unit is used to identify fixed facilities in the green belt using contour and color features, and to identify non-plantable areas and distinguish between the green belt and the driveway using text signs, so as to obtain the corresponding bounding boxes and confidence levels of the fixed facilities, the corresponding bounding boxes and confidence levels of the non-plantable areas, the bounding boxes and confidence levels of the green belt, and the bounding boxes and confidence levels of the driveway.

[0183] In one embodiment, identifying associated subunits includes:

[0184] The analysis module is used to analyze the image recognition results and determine the target green belt as the center of the green belt near the lane line; the tree recognition module is used to identify the target trees as roadside trees in the target green belt by comparing the position of the center point of the tree root, so as to obtain the target roadside trees; the association module is used to associate the target green belt and the target roadside trees to obtain the associated target roadside trees and green belt.

[0185] In one embodiment, the analysis and transformation subunit includes:

[0186] The motion trajectory capture module is used to capture the motion trajectory of each target roadside tree within the corresponding green belt based on the associated target roadside trees and green belts using a multi-target tracking algorithm; the update module is used to establish the association between consecutive frames based on spatial proximity and appearance similarity, and update or create motion trajectory records for each target roadside tree, wherein the motion trajectory records include timestamps, coordinates, and status identifiers; the adjustment module is used to adjust the motion trajectory to a coordinate system to obtain the transformed motion trajectory data.

[0187] In one embodiment, the fitting unit includes:

[0188] The baseline determination module is used to analyze the converted motion trajectory data of multiple target roadside trees within the same green belt, calculate the motion vector of the overall forward direction, orthogonally project the position coordinates of the target roadside trees onto the straight line defined by the motion vector, and fit a roadside tree distribution baseline based on the projection points to obtain the distribution baseline of the target roadside trees; the desired spacing determination module is used to select a distribution baseline that meets the requirements of the lane distance, calculate the arc distance between adjacent projection points, and obtain the statistical average value of the arc distance to obtain the desired spacing; wherein, the distribution model includes the target roadside tree distribution baseline and the desired spacing.

[0189] In one embodiment, the prediction unit 303 includes:

[0190] The averaging subunit is used to select the roadside tree with the smallest ordinate within the selected green belt as the prediction starting point according to the distribution model, and to determine the world coordinates of the prediction point by averaging the expected spacing along the distribution baseline of the target roadside trees based on the prediction starting point. The region creation subunit is used to create a prediction interest region in the coordinate system with the prediction point as the center, with the vertical axis being several times the standard deviation of the spacing. Wherein, if there is only one distribution baseline, the width of the green belt is set as the dimension in the horizontal axis direction; if there are multiple distribution baselines, the intersection of the boundary line of the target green belt closest to the lane line is used as one vertex, and the center line of the target roadside tree distribution baseline and the distribution baseline closest to the target roadside tree distribution baseline is calculated as another vertex to define the prediction interest region in the case of multiple baselines. The transformation subunit is used to inversely transform the prediction interest region from the coordinate system to the original image coordinate system, and output the prediction interest region and its confidence score.

[0191] In one embodiment, the comprehensive classification unit 304 is used to integrate the street tree location prediction results. If an unplantable marker is detected but no tree characteristics are found, and the presence is confirmed through tracking prediction, it is marked as an unplantable area and details are recorded. If a moving obstruction is identified without tree characteristics, and if the number of stable tracking frames exceeds the required number, it is considered an obstruction and an event record is generated. If a tree pit is detected but not marked as an unplantable area, and its presence is confirmed in the required number of tracking frames, it is defined as a missing tree pit and information is recorded. If a tree stump is detected but not in an unplantable area, and its presence is confirmed in the tracking frames exceeding a set threshold, it is defined as a missing tree stump and corresponding data is recorded.

[0192] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned rapid and continuous detection system for missing roadside trees based on dynamic edge computing 300 and each unit can be referred to the corresponding description in the aforementioned method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0193] The aforementioned rapid and continuous detection system 300 for missing roadside trees based on dynamic edge computing can be implemented as a computer program, which can be used in various ways, such as... Figure 8 It runs on the computer device shown.

[0194] Please see Figure 8 , Figure 8 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 is a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0195] See Figure 8 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0196] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a fast and continuous detection method for missing trees in roadside trees based on dynamic edge computing.

[0197] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0198] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a method for rapid and continuous detection of missing trees in roadside trees based on dynamic edge computing.

[0199] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0200] The processor 502 is used to run the computer program 5032 stored in the memory to implement all the steps of the fast and continuous detection method for missing trees in roadside trees based on dynamic edge computing.

[0201] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0202] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0203] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform all the steps of the described method for rapid and continuous detection of missing trees in roadside trees based on dynamic edge computing.

[0204] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0205] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0206] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0207] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the system of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0208] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0209] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for fast and continuous detection of street tree missing poles based on dynamic edge computing, characterized in that, The method comprises the following steps: acquiring images, GPS coordinates, and timestamps collected by an edge computing device installed on a vehicle, and constructing a spatiotemporal trajectory based on the collected data; preprocessing the images, identifying and classifying vegetation, municipal facilities, and moving shelter targets, outputting feature information and confidence scores, and obtaining image recognition results; based on the image recognition results, screening street trees close to lane lines and predicting corresponding positions, constructing motion trajectories through continuous multi-frame data association, and calibrating in a coordinate system to generate a predicted attention area, forming street tree position prediction results; the motion trajectory refers to the spatial position change record of each tree at different time points; comprehensively determining and classifying the lack of trees in the green belt based on the street tree position prediction results, and generating relevant event information; including: based on the street tree position prediction results, if no tree signs are found after detecting unplantable signs, the existence is confirmed through tracking prediction, and then the area is marked as unplantable and the details are recorded; if no tree signs are found after identifying moving shelters, if stable tracking exceeds the required number of frames, the area is determined as a shelter and an event record is generated; if a tree hole is detected and the area is not marked as unplantable, the existence is confirmed in the tracking frames that meet the requirements, and the area is determined as a tree hole and the information is recorded; if a tree stump is detected and the area is not unplantable, the existence is confirmed in the tracking frames that exceed the set threshold, and the area is defined as a tree stump and the corresponding data is recorded; the event includes shelter and lack of trees; wherein the lack of trees includes unplantable area, tree hole, or tree stump. 2.The method of claim 1, wherein, The method of screening street trees close to lane lines and predicting corresponding positions based on the image recognition results, constructing motion trajectories through continuous multi-frame data association, and calibrating in a coordinate system to generate a predicted attention area, forming street tree position prediction results, comprises: based on the image recognition results, identifying and selecting green belts close to lane lines and street trees inside, and associating the selected green belts and street trees to obtain associated target street trees and green belts; analyzing the associated target street trees in continuous frame data, establishing a motion trajectory for each target street tree, and adjusting the motion trajectory to the coordinate system to obtain converted motion trajectory data; fitting the target street tree distribution reference line and the expected distance based on the converted motion trajectory data to obtain a distribution model; in the selected green belt, selecting the street tree with the smallest ordinate as the predicted starting point according to the distribution model, and generating an attention area based on the predicted starting point, and converting the attention area back to the original image coordinate system to obtain the street tree position prediction results. 3.The method of claim 1, wherein, The method of preprocessing the images, identifying and classifying vegetation, municipal facilities, and moving shelter targets, outputting feature information and confidence scores, and obtaining image recognition results comprises: preprocessing the images to obtain preprocessed images; analyzing the color and texture differences of vertical lines, bark features, and ground depressions in the preprocessed images to identify trees and their tree holes, and classifying trees into street trees and tree stumps according to height to obtain vegetation detection results; identify fixed facilities in the green belt of the pre-processed image, identify non-plantable areas, and distinguish the green belt from the lane, to obtain fixed facility corresponding bounding boxes and confidence, non-plantable area corresponding bounding boxes and confidence, green belt bounding boxes and confidence, and lane bounding boxes and confidence; identify and mark vehicles based on vehicle outlines, wheels, and window features, to obtain vehicle bounding boxes and confidence; combine the vegetation detection results, fixed facility corresponding bounding boxes and confidence, non-plantable area corresponding bounding boxes and confidence, green belt bounding boxes and confidence, lane bounding boxes and confidence, and vehicle bounding boxes and confidence to form image recognition results.

4. The method of claim 3, wherein the method further comprises: The identification of fixed facilities in the green belt of the pre-processed image, the identification of non-plantable areas, and the distinction between the green belt and the lane to obtain fixed facility corresponding bounding boxes and confidence, non-plantable area corresponding bounding boxes and confidence, green belt bounding boxes and confidence, and lane bounding boxes and confidence includes: identify fixed facilities in the green belt using contour and color features, and identify non-plantable areas and distinguish the green belt from the lane through text signs to obtain fixed facility corresponding bounding boxes and confidence, non-plantable area corresponding bounding boxes and confidence, green belt bounding boxes and confidence, and lane bounding boxes and confidence.

5. The method of claim 2, wherein the method further comprises: The identification and selection of green belts near lane lines and internal street trees based on the image recognition results, and the association of the selected green belts and street trees to obtain associated target street trees and green belts includes: Based on the image recognition results, determine the target green belt as the center of the green belt near the lane line. In the target green belt, identify the target tree as a street tree by comparing the positions of tree root center points to obtain the target street tree. Associate the target green belt and the target street tree to obtain the associated target street tree and green belt.

6. The method of claim 2, wherein the method further comprises: The analysis of the associated target street trees in the continuous frame data to establish a motion trajectory for each target street tree includes: Based on the associated target street trees and green belts, apply a multi-target tracking algorithm to capture the motion trajectory of each target street tree within the corresponding green belt. Based on spatial proximity and appearance similarity, establish the association between consecutive frames, update or create the motion trajectory record of each target street tree, wherein the motion trajectory record includes a timestamp, coordinates, and a state identifier. Adjust the motion trajectory to the coordinate system to obtain converted motion trajectory data.

7. The method of claim 2, wherein the method further comprises: The fitting of the target street tree distribution reference line and the expected distance based on the converted motion trajectory data to obtain a distribution model includes: Analyze the converted motion trajectory data of multiple target street trees within the green belt, calculate the motion vector of the overall forward direction, orthogonally project the position coordinates of the target street tree onto the straight line defined by the motion vector, and fit the street tree distribution reference line based on the projection point to obtain the distribution reference line of the target street tree. Select a distribution reference line meeting the distance requirement, calculate the arc length distance between adjacent projection points, and obtain the statistical average of the arc length distance to obtain the expected interval; The distribution model includes the target street tree distribution reference line and the expected interval. 8.The method of claim 5, wherein, The selected green belt is selected as the prediction starting point according to the distribution model, and a region of interest is generated based on the prediction starting point. The region of interest is converted back to the original image coordinate system to obtain a street tree position prediction result, including: In the selected green belt, the street tree with the minimum ordinate is selected as the prediction starting point according to the distribution model, and the world coordinates of the prediction point are determined by moving the prediction starting point along the distribution reference line of the target street tree by an average of the expected interval; In the coordinate system, a prediction region of interest is created with the prediction point as the center, and the longitudinal axis direction is several times the standard deviation of the interval. If there is only one distribution reference line, the width of the green belt is set as the size of the horizontal axis direction. If there are multiple distribution reference lines, the intersection of the side boundary line closest to the lane line of the target green belt is used as a vertex, and the center line of the target street tree distribution reference line and the distribution reference line closest to the target street tree distribution reference line is calculated as another vertex to define the prediction region of interest in the case of multiple reference lines. The prediction region of interest is inversely transformed from the coordinate system to the original image coordinate system, and the prediction region of interest and its confidence score are output.

9. A street tree missing pole rapid and continuous detection system based on dynamic edge computing, characterized in that, Including: An acquisition unit is configured to acquire images, GPS coordinates, and timestamps collected by an edge computing device installed on a vehicle, and construct a spatiotemporal trajectory based on the collected data; An identification unit is configured to preprocess the images, identify and classify vegetation, municipal facilities, and moving occlusion objects, and output feature information and confidence scores to obtain image recognition results; A prediction unit is configured to filter street trees close to lane lines and predict corresponding positions based on the image recognition results, construct a motion trajectory by correlating consecutive multiple frames of data, and calibrate in a coordinate system to generate a prediction region of interest, forming a street tree position prediction result. The motion trajectory refers to the spatial position change record of each tree at different time points. A comprehensive classification unit is configured to determine and classify the lack of trees in the green belt based on the street tree position prediction result, and generate relevant event information. Including: after detecting an unplantable mark and finding no tree signs, the continuous existence is confirmed by tracking prediction, and the unplantable area is marked and the details are recorded. When a moving occlusion object is identified without tree signs, if the stable tracking exceeds the required number of frames, it is determined as an occlusion and an event record is generated. After detecting a tree hole and marking it as an unplantable area, the existence is confirmed in the tracking frames that meet the requirements, and the tree hole is defined as a lack of trees and the information is recorded. After detecting a tree stump and not being an unplantable area, the existence is confirmed in the tracking frames that exceed the set threshold, and the tree stump is defined as a lack of trees and the corresponding data is recorded. The event includes occlusion and lack of trees. The lack of trees includes unplantable area, tree hole, or tree stump.

Images