Guardrail obstacle avoidance maintenance method and system based on smart city

Abstract

The invention relates to the field of smart city maintenance, and discloses a smart city-based guardrail obstacle avoidance maintenance method and system, and the method comprises the steps: carrying out the dynamic recognition of a guardrail maintenance robot, a road cleaning robot and a guardrail target structure in a shared road shoulder operation region through a multi-source heterogeneous sensing unit; acquiring a guardrail state data stream sequence subjected to time synchronization calibration, constructing a multi-dimensional time sequence feature vector of an equipment cooperative detection state, and detecting a high-risk precision attenuation candidate event by identifying a deviation degree between track distribution abnormality and a predicted detection time sequence; dividing a guardrail detection section and an influence range corresponding to the high-risk precision attenuation candidate event as candidate intervention areas, constructing a local collaborative detection map, performing intervention judgment on a collaborative detection evolution path, and generating a dynamic obstacle avoidance adjustment and detection precision enhancement result; the method has the advantage of improving the detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of smart city maintenance, specifically to a method and system for maintaining guardrails and preventing obstacles in smart cities. Background Technology

[0002] With the advancement of smart city construction, low-speed autonomous mobile inspection and maintenance robots are increasingly being used for the operation and maintenance of road infrastructure. These robots autonomously travel along one side of guardrails using lidar and camera sensors to complete defect detection, deformation identification, and simple maintenance tasks. Simultaneously, road cleaning robots are also being deployed along urban shoulders or non-motorized vehicle lanes to automatically clean up garbage, fallen leaves, and stagnant water. However, in existing technologies, the path planning and perception system design for both types of equipment are implemented independently, lacking a cross-device collaboration mechanism. When both types of robots operate simultaneously on the same shoulder, the following problems often arise: when the road cleaning robot is running in front of the guardrail robot, it obstructs the sensor's line of sight, leading to missed or false detections of guardrail defects. Especially in busy urban arterial road environments with limited operating time and narrow operating spaces, these problems significantly reduce the real-time detection capability of guardrail inspections. Therefore, it is essential to design a guardrail obstacle avoidance maintenance method and system based on smart cities to improve detection accuracy. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for guardrail obstacle avoidance maintenance based on smart cities, which has the advantage of improving detection accuracy and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving detection accuracy, this invention provides the following technical solution: a guardrail obstacle avoidance maintenance method based on smart cities, comprising the following steps: The multi-source heterogeneous sensing unit dynamically identifies the guardrail maintenance robot, road cleaning robot and guardrail target structure in the shared shoulder operation area, and extracts the robot's shape boundary, occlusion overlap area and guardrail visible area based on the occlusion semantic segmentation algorithm, and obtains the guardrail status data stream sequence after time synchronization calibration. Multi-device cross-frame correlation tracking is performed on the guardrail status data stream sequence, and a multi-dimensional temporal feature vector of the device collaborative detection status is constructed by combining the motion trajectory characteristics of the road cleaning robot, the priority parameters of the guardrail maintenance task and the distribution of occlusion mode. Based on multidimensional temporal feature vectors, a dynamic occlusion recovery clustering algorithm is used to extract the state evolution trajectory in the collaborative detection process. By identifying the deviation between the trajectory distribution anomaly and the expected detection time series, high-risk accuracy decay candidate events are detected. The guardrail detection section and its impact range corresponding to the high-risk accuracy attenuation candidate event are defined as candidate intervention areas. The passage priority nodes, field reconstruction nodes and signal interference suppression nodes in the candidate area are integrated to construct a local collaborative detection map. Based on the local collaborative detection map, combined with the fluctuation trend of global occlusion probability in the potential detection imbalance nodes and the guardrail status data stream sequence, the collaborative detection evolution path is intervened and judged, and dynamic obstacle avoidance adjustment and detection accuracy enhancement results are generated.

[0005] Preferably, the process of extracting the robot's shape boundary, occlusion overlap area, and guardrail visible area based on the occlusion semantic segmentation algorithm is as follows: In the multi-source heterogeneous sensing unit, lidar point cloud, RGB-D camera and millimeter-wave radar are used to collect three-dimensional environmental data of the road shoulder area respectively; A fusion-based semantic segmentation network is applied to the collected environmental data, combining convolutional neural networks and conditional random field models to achieve pixel-level segmentation and regional occlusion reconstruction of the robot and guardrail boundaries. Based on temporal optical flow analysis, dynamic obstructions and static guardrail structures are distinguished, and the confidence level of overlapping obstruction areas is marked to extract the continuity and integrity indicators of the guardrail's visible area. The segmentation results are combined with multi-sensor fusion localization information to output the robot's external boundary contour, occluded area polygons, and complete field-of-view mask of the guardrail.

[0006] Preferably, the process of obtaining the time-synchronized guardrail status data stream sequence is as follows: A unified timestamp synchronization mechanism is used to correct time discrepancies in data collected from different sensors. The corrected multimodal data is then transformed to spatial coordinates and mapped to a unified three-dimensional geographic coordinate system. By combining dynamic filtering algorithms, the influence of sensor noise and environmental interference on the data is filtered out, forming a sequence of guardrail status data streams.

[0007] Preferably, the process of combining the motion trajectory characteristics of the road cleaning robot, the priority parameters of the guardrail maintenance task, and the distribution of occlusion patterns is as follows: Based on the trajectory aggregation algorithm, multi-scale analysis is performed on the historical and real-time trajectory data of road cleaning robots and guardrail maintenance robots to extract trajectory density, speed fluctuation and turning characteristics. Using a priority model, the robot's path priority is dynamically adjusted based on the urgency of the guardrail maintenance task and the risk level of the detection area; Construct an occlusion pattern probability distribution map and combine trajectory features to predict potential occlusion periods and areas; Simultaneous analysis of motion trajectory, task priority, and occlusion pattern forms a fusion scheduling strategy input that covers multiple parameters.

[0008] Preferably, the process of constructing a multi-dimensional temporal feature vector of the device collaborative detection status is as follows: The trajectory information, occlusion semantic segmentation results, and task priority indicators of road cleaning robots and guardrail maintenance robots in the spatiotemporal domain are mapped into a unified feature encoding format. By employing time-series modeling methods and using recurrent neural networks or Transformer architectures, we can capture the collaborative relationships and temporal dependencies between devices and extract multi-dimensional time-series feature vectors. By integrating signal strength, error distribution, and environmental interference indicators from multiple sensors, it reflects the dynamic state evolution of multi-device collaborative detection in real time.

[0009] Preferably, the process of extracting the state evolution trajectory during the collaborative detection process using the dynamic occlusion recovery clustering algorithm is as follows: Based on multidimensional temporal feature vectors, a spatiotemporal density peak clustering algorithm is used to identify typical state patterns and abnormal state distributions during the guardrail detection process. By combining the occlusion recovery model, interpolation and compensation are performed on the detection data affected by occlusion to restore the complete detection trajectory; By employing trajectory segmentation and dynamic threshold adjustment mechanisms, the system accurately captures state transition nodes and persistent abnormal intervals, outputting a collaboratively detected state evolution trajectory that covers the entire time series.

[0010] Preferably, the process for detecting high-risk accuracy decay candidate events is as follows: By comparing and analyzing the evolution trajectory of the collaborative detection state with the preset detection accuracy baseline model, spatiotemporal segments where accuracy significantly decreases are identified. A multi-factor weighted scoring mechanism is introduced to comprehensively consider the degree of occlusion, sensor signal interference, and path complexity, and to assign a risk level to candidate events. Anomaly detection algorithms are used to determine the occurrence time and duration of high-risk precision decay candidate events, which are then integrated into the subsequent intervention decision-making module.

[0011] Preferably, the process of constructing a local collaborative detection map is as follows: Based on the detection zones defined by high-risk candidate events, extract the guardrail structure nodes, operation robot nodes, and environmental perception nodes within the zones; Based on the spatial adjacency relationships between nodes, task collaboration dependencies, and signal interference propagation paths, a multimodal node connection edge set is constructed to form a graph structure that reflects the collaborative logic of guardrail detection. By combining historical operation records and real-time status data, weights are assigned to nodes and edges to express collaborative efficiency and interference intensity; By applying graph embedding and graph convolutional network algorithms, key influencing factors and potential imbalance nodes are identified, forming a local collaborative detection map.

[0012] Preferably, the process of generating dynamic obstacle avoidance adjustment and detection accuracy enhancement results is as follows: Based on the key imbalance nodes and impact paths identified in the local collaborative detection map, and combined with the global occlusion probability fluctuation trend of the current guardrail status data stream, a dynamic obstacle avoidance adjustment strategy is deduced. Dynamically optimize robot path planning and task allocation by using reinforcement learning or model predictive control algorithms; By combining signal interference suppression technology and adjusting the operating frequency and timing of multiple sensors, the sensing error and false trigger rate can be reduced. Taking into account both operational efficiency and detection accuracy, the collaborative detection parameters are adjusted in real time to form a closed-loop feedback control, ultimately outputting a dynamic result report that includes obstacle avoidance path updates and improved detection accuracy.

[0013] A smart city-based guardrail obstacle avoidance maintenance system includes: Perception and Recognition Module: Utilizing multi-source heterogeneous perception units combined with occlusion semantic segmentation algorithms, it enables dynamic and accurate recognition and time-synchronized calibration of guardrail and robot shapes and occlusion areas; The trajectory tracking module performs multi-device cross-frame correlation on the guardrail status data stream, integrates robot motion features and task priorities, and constructs a multi-dimensional temporal feature vector for device collaboration. Anomaly clustering module: Based on multi-dimensional feature vectors, a dynamic occlusion recovery clustering algorithm is used to extract the collaborative detection state evolution trajectory and detect high-risk accuracy decay candidate events; Map construction module: Delineates the impact area of ​​high-risk events, integrates information on traffic priority, field reconstruction and signal suppression nodes, and constructs a local collaborative detection map; Dynamic Response Module: Combining local map data and global occlusion fluctuation trends, it implements collaborative detection path intervention to generate real-time obstacle avoidance adjustment and detection accuracy enhancement schemes.

[0014] Compared with existing technologies, the present invention provides a method and system for guardrail obstacle avoidance maintenance based on smart cities, which has the following beneficial effects: This invention introduces multi-source heterogeneous sensing units to achieve accurate dynamic identification of guardrail maintenance robots, road cleaning robots, and guardrail target structures within a shared shoulder work area. Combined with an occlusion semantic segmentation algorithm, it effectively extracts robot outline boundaries and overlapping occlusion areas, significantly improving the accuracy and timeliness of guardrail status data acquisition. Through multi-device cross-frame correlation tracking and the construction of multi-dimensional temporal feature vectors, the method comprehensively integrates multiple key factors such as motion trajectory, maintenance task priority, and occlusion distribution, achieving efficient collaborative detection among devices. The application of a dynamic occlusion recovery clustering algorithm further enhances the deep analysis capability of state evolution trajectories, effectively identifying high-risk candidate events for detection accuracy degradation, ensuring the safety and reliability of the maintenance process. For candidate events of detection anomalies, the system can accurately delineate intervention areas, combining passage priority nodes, view reconstruction nodes, and signal interference suppression nodes to construct a detailed and dynamic local collaborative detection map, achieving multi-dimensional intervention in space and time. Ultimately, based on the analysis of global occlusion probability fluctuation trends and potential detection imbalance nodes, the system can intelligently intervene in the collaborative detection evolution path in a timely manner, dynamically adjust obstacle avoidance strategies and improve detection accuracy, significantly improve the operating efficiency and safety protection level of the guardrail maintenance robot, reduce detection errors and risks caused by occlusion and environmental complexity, and promote the intelligent, automated and refined development of smart city road maintenance systems. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] 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 embodiments of the present invention, and not all embodiments. 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.

[0017] Example 1: A guardrail obstacle avoidance maintenance method based on smart cities according to this embodiment of the invention includes the following steps: S1: Dynamically identify the guardrail maintenance robot, road cleaning robot and guardrail target structure in the shared shoulder operation area through multi-source heterogeneous sensing unit, and extract the robot's shape boundary, occlusion overlap area and guardrail visible area based on occlusion semantic segmentation algorithm to obtain the guardrail status data stream sequence after time synchronization calibration. The process of extracting the robot's external boundary, occlusion overlap region, and guardrail visible area based on the occlusion semantic segmentation algorithm in S1 is as follows: In the multi-source heterogeneous sensing unit, lidar point cloud, RGB-D camera and millimeter-wave radar are used to collect three-dimensional environmental data of the road shoulder area respectively; The system simultaneously acquires 3D environmental data of the road shoulder area from LiDAR point clouds, RGB-D cameras, and millimeter-wave radar in a multi-source heterogeneous sensing unit. LiDAR provides high-precision spatial point cloud information, RGB-D cameras capture color and depth images, and millimeter-wave radar supplements information on dynamic objects in the environment. By complementing each other through multiple sensors, the system enhances the comprehensiveness and accuracy of environmental perception, providing a rich data foundation for subsequent semantic segmentation.

[0018] A fusion-based semantic segmentation network is applied to the collected environmental data, combining convolutional neural networks and conditional random field models to achieve pixel-level segmentation and regional occlusion reconstruction of the robot and guardrail boundaries. The collected multimodal environmental data is input into a fusion-type semantic segmentation network. This network combines a deep convolutional neural network to automatically extract spatial features and a conditional random field model to optimize pixel classification. It can accurately achieve pixel-level boundary segmentation between robots and guardrails. At the same time, it restores the spatial morphology of the occluded area based on the occlusion reconstruction algorithm, effectively solving the problem of incomplete boundaries caused by partial occlusion and improving the accuracy and robustness of recognition.

[0019] Based on temporal optical flow analysis, dynamic obstructions and static guardrail structures are distinguished, and the confidence level of overlapping obstruction areas is marked to extract the continuity and integrity indicators of the guardrail's visible area. The system utilizes a temporal optical flow algorithm for consecutive frames to analyze pixel motion features at different time points, effectively distinguishing between dynamic occlusions and static guardrail structures. By combining the motion consistency of the occlusion area, the system labels the confidence level of the overlapping occlusion area, thereby accurately extracting the continuity and integrity indicators of the guardrail's visible area, providing a reliable basis for target recognition and status assessment in dynamic environments.

[0020] The segmentation results are combined with multi-sensor fusion localization information to output the robot's external boundary contour, occluded area polygons, and complete field-of-view mask of the guardrail. By combining semantic segmentation and occlusion reconstruction results with multi-sensor fusion localization information, the system accurately locates the robot's outline through spatial coordinate mapping, generates a polygonal representation of the occluded area, and constructs a complete field-of-view mask for the guardrail. This is used for subsequent dynamic obstacle avoidance and maintenance decisions, ensuring the system's real-time and accurate perception of the robot's posture and the guardrail's field-of-view status.

[0021] The process of obtaining the time-synchronized guardrail status data stream sequence in S1 is as follows: A unified timestamp synchronization mechanism is used to correct time discrepancies in data collected from different sensors. The system first collects raw data from different types of sensors and extracts the corresponding timestamp information. By constructing a timestamp sequence consistency model, it calculates the differences and drift between timestamps of different sensors. Combining interpolation and difference algorithms, it analyzes whether there are any non-monotonic increases, repetitions, or delay anomalies in the time series. A high-precision clock synchronization protocol is used as the time reference, and a clock deviation compensation algorithm is used to automatically adjust the timestamps of each sensor to achieve cross-device time synchronization. This ensures the consistency of multi-source heterogeneous data in the time dimension and lays a solid foundation for subsequent data fusion.

[0022] The corrected multimodal data is then transformed to spatial coordinates and mapped to a unified three-dimensional geographic coordinate system. After time synchronization, the multimodal sensor data obtains the spatial position and attitude parameters of each sensor through intrinsic and extrinsic parameter calibration. Based on these calibration parameters, the system transforms the point cloud, image, and radar data collected by each sensor into a unified three-dimensional geographic coordinate system through rigid body transformation and coordinate mapping algorithms. This coordinate system typically adopts the global reference coordinate system in the geographic information system to ensure that the spatial information from different data sources is accurately superimposed under a unified framework. This step effectively solves the spatial error problem between multi-source sensors, realizes the spatial fusion of multimodal information, and improves the spatial consistency and accuracy of subsequent analysis.

[0023] By combining dynamic filtering algorithms, the influence of sensor noise and environmental interference on the data is filtered out, forming a sequence of guardrail status data streams; The fused spatiotemporal synchronized data further enters the dynamic filtering process. The system combines various filtering algorithms such as Kalman filtering, particle filtering, or adaptive filtering to effectively suppress sensor noise, environmental interference, and occasional abnormal data. The filter adjusts the filtering parameters in real time according to the sensor characteristics and the dynamic model of the guardrail to achieve smooth data and accurate prediction. The filtering results undergo multi-dimensional quality evaluation to remove abnormal fluctuation points, ensuring that the generated guardrail status data stream sequence has high stability and robustness, providing reliable data support for subsequent status identification and maintenance decisions.

[0024] S2: Perform multi-device cross-frame correlation tracking on the guardrail status data stream sequence, and combine the motion trajectory features of the road cleaning robot, the priority parameters of the guardrail maintenance task and the distribution of occlusion modes to construct a multi-dimensional temporal feature vector of the collaborative detection status of the devices; The process in S2 that combines the motion trajectory characteristics of the road cleaning robot, the priority parameters of the guardrail maintenance task, and the distribution of occlusion patterns is as follows: Based on the trajectory aggregation algorithm, multi-scale analysis is performed on the historical and real-time trajectory data of road cleaning robots and guardrail maintenance robots to extract trajectory density, speed fluctuation and turning characteristics. Based on a trajectory aggregation algorithm, the system collects and integrates historical and real-time trajectory data of road cleaning robots and guardrail maintenance robots. Through multi-scale analysis, it calculates the spatial density distribution, speed fluctuation range, and turning change characteristics of the trajectory, quantifying the dynamic patterns of robot motion behavior. These rich trajectory features provide an important spatiotemporal information foundation for subsequent path optimization and occlusion prediction.

[0025] Using a priority model, the robot's path priority is dynamically adjusted based on the urgency of the guardrail maintenance task and the risk level of the detection area; By combining the priority parameters of guardrail maintenance tasks, the system constructs a dynamic priority model based on the urgency of the maintenance tasks, the risk level of the detection area, and the current status of the robot. It adjusts the priority allocation of the robot path in real time. Through the reasonable allocation of priority weights, the system achieves efficient scheduling of maintenance tasks and key coverage of risk areas, ensuring priority maintenance and safety detection of critical guardrail areas.

[0026] Construct an occlusion pattern probability distribution map and combine trajectory features to predict potential occlusion periods and areas; The system constructs a probability distribution map of occlusion patterns in the shoulder area by using historical occlusion data and real-time environmental perception. Combined with trajectory features, it predicts the potential occlusion time periods and occlusion spatial areas of the robot, accurately identifies high-risk occlusion areas, and effectively prevents detection blind spots and data accuracy degradation caused by occlusion. This provides a basis for occlusion risk assessment for path planning and collaborative scheduling.

[0027] Simultaneous analysis of motion trajectory, task priority, and occlusion pattern forms a fusion scheduling strategy input covering multi-dimensional parameters; By synchronously analyzing and fusing motion trajectory features, task priority weights, and occlusion pattern probability distributions, a multi-dimensional parameter vector covering time, space, and task urgency is formed. This vector serves as the input for the collaborative robot scheduling strategy, enabling dynamic coordination of path priority, task allocation, and obstacle avoidance behavior, thereby improving the overall intelligent scheduling capability and detection efficiency of the system.

[0028] The process of constructing the multidimensional temporal feature vector of the device collaborative detection status in S2 is as follows: The trajectory information, occlusion semantic segmentation results, and task priority indicators of road cleaning robots and guardrail maintenance robots in the spatiotemporal domain are mapped into a unified feature encoding format. The system first standardizes the trajectory information, occlusion semantic segmentation results, and task priority indicators of the road cleaning robot and the guardrail maintenance robot in the spatiotemporal domain. It then uses a unified encoding format to map multi-source heterogeneous data into structured feature vectors. This process includes the quantitative expression of trajectory coordinates, occlusion area boundaries, and priority weights, providing a consistent and compatible input basis for subsequent multi-dimensional temporal modeling.

[0029] By employing time-series modeling methods and using recurrent neural networks or Transformer architectures, we can capture the collaborative relationships and temporal dependencies between devices and extract multi-dimensional time-series feature vectors. Based on the unified encoding of feature inputs, the system adopts advanced temporal modeling architectures such as recurrent neural networks, long short-term memory networks, or Transformers to deeply explore the collaborative interaction relationships between devices and their temporal dependencies. Through multi-layer network structures, the system encodes and decodes temporal data to extract multi-dimensional temporal feature vectors, reflecting the dynamic evolution and state change trends in the collaborative detection process of devices.

[0030] By integrating signal strength, error distribution, and environmental interference indicators from multiple sensors, the dynamic state evolution of multi-device collaborative detection can be reflected in real time. Based on the extracted multidimensional time-series features, the system further integrates signal strength, error distribution, and environmental interference indicators from multiple source sensors. These real-time dynamic parameters reflect the sensor data quality and the influence of the external environment, comprehensively showcasing the real-time state evolution of multi-device collaborative detection, and providing comprehensive and accurate dynamic feature support for subsequent anomaly detection and system optimization.

[0031] S3: Based on multi-dimensional temporal feature vectors, the dynamic occlusion recovery clustering algorithm is used to extract the state evolution trajectory in the collaborative detection process. By identifying the deviation between the trajectory distribution anomaly and the expected detection time series, high-risk accuracy decay candidate events are detected. The process of extracting the state evolution trajectory in the collaborative detection process using the dynamic occlusion recovery clustering algorithm in S3 is as follows: Based on multidimensional temporal feature vectors, a spatiotemporal density peak clustering algorithm is used to identify typical state patterns and abnormal state distributions during the guardrail detection process. Based on multi-dimensional temporal feature vectors, the system uses a spatiotemporal density peak clustering algorithm to analyze the distribution density of data points in the spatiotemporal domain. By automatically identifying the density peak cluster centers, it can accurately classify typical state patterns and abnormal states in the guardrail detection process. This algorithm can effectively distinguish between normal detection states and abnormal fluctuations, and improve the detection sensitivity and robustness of abnormal patterns.

[0032] By combining the occlusion recovery model, interpolation and compensation are performed on the detection data affected by occlusion to restore the complete detection trajectory; By combining the principle of spatiotemporal continuity, interpolation algorithms are used to reasonably infer missing data, while a compensation mechanism is used to correct detection errors affected by occlusion, thereby restoring a complete and continuous detection trajectory, effectively compensating for information loss caused by occlusion, and ensuring the integrity and accuracy of detection results.

[0033] By using trajectory segmentation and dynamic threshold adjustment mechanisms, the system accurately captures state transition nodes and continuous abnormal intervals, and outputs collaborative detection state evolution trajectories covering the entire time series. By employing trajectory segmentation technology combined with a dynamic threshold adjustment mechanism, the rate of trajectory change and fluctuations in abnormal indicators are monitored in real time. This allows for precise location of state switching nodes and persistent abnormal intervals, thereby generating a collaborative detection state evolution trajectory covering the entire time series. This technology enhances the ability to characterize the state evolution process with fine granularity and provides strong support for the early identification of abnormal events.

[0034] The process of detecting high-risk accuracy decay candidate events in S3 is as follows: By comparing and analyzing the evolution trajectory of the collaborative detection state with the preset detection accuracy baseline model, spatiotemporal segments where accuracy significantly decreases are identified. By comparing the evolution trajectory of the collaborative detection state with the pre-established detection accuracy baseline model in the spatiotemporal dimension, and by calculating the deviation between the actual detection accuracy and the baseline accuracy, the spatiotemporal segments where the accuracy has dropped significantly are automatically identified. This process relies on the dynamic update mechanism of the accuracy baseline, which can adapt to environmental changes and reflect the fluctuation of detection performance in real time, ensuring the accurate location and timely detection of abnormal segments.

[0035] A multi-factor weighted scoring mechanism is introduced to comprehensively consider the degree of occlusion, sensor signal interference, and path complexity, and to assign a risk level to candidate events. For the identified potential areas of accuracy degradation, the system introduces a multi-factor weighted scoring mechanism. It comprehensively analyzes multiple dimensions of factors, including the degree of occlusion, the intensity of sensor signal interference, path complexity, and dynamic changes in the environment. The risk score is calculated through a weighted fusion algorithm, thereby assigning a fine-grained risk level to candidate events. This mechanism enhances the comprehensiveness and objectivity of risk assessment and provides a quantitative basis for precise intervention.

[0036] Anomaly detection algorithms are used to determine the occurrence time and duration of high-risk accuracy decay candidate events, which are then integrated into the subsequent intervention decision-making module. The system employs an anomaly detection algorithm based on statistical learning, combined with time series analysis technology, to accurately determine the occurrence time and duration of high-risk precision decay candidate events. It monitors abnormal trends in real time and integrates the detection results into the subsequent intervention decision-making module. This technical feature ensures early warning and dynamic tracking of potential risk events, improving the overall system's response speed and decision-making efficiency.

[0037] S4: Define the guardrail detection section and its impact range corresponding to the high-risk accuracy attenuation candidate event as the candidate intervention area, and integrate the passage priority node, field reconstruction node and signal interference suppression node in the candidate area to construct a local collaborative detection map; The process of constructing the local collaborative detection map in S4 is as follows: Based on the detection zones defined by high-risk candidate events, extract the guardrail structure nodes, operation robot nodes, and environmental perception nodes within the zones; The corresponding detection area is precisely delineated, and guardrail structure nodes, operation robot nodes, and environmental perception nodes are extracted within the area. The extraction process comprehensively utilizes multi-source sensor data and task scheduling information to ensure that all key entities participating in collaborative detection are covered, forming a comprehensive node set, which provides accurate and complete basic data support for subsequent map construction.

[0038] Based on the spatial adjacency relationships between nodes, task collaboration dependencies, and signal interference propagation paths, a multimodal node connection edge set is constructed to form a graph structure that reflects the collaborative logic of guardrail detection. By analyzing the spatial adjacency relationships, task collaboration dependencies, and signal interference propagation paths between nodes, a multimodal node connection edge set is constructed. This edge set not only describes the physical proximity between nodes but also integrates task flow and signal propagation characteristics to form a complex graph structure that reflects the multidimensional collaborative logic in guardrail detection, thereby achieving accurate modeling of each collaborative factor.

[0039] By combining historical operation records and real-time status data, weights are assigned to nodes and edges to express collaborative efficiency and interference intensity; The system dynamically assigns weights to nodes and edges in the graph structure. Node weights quantify the efficiency and state stability of each entity in the collaboration process, while edge weights reflect the collaboration strength and signal interference level on the connection path. This weight assignment mechanism comprehensively reflects the operational efficiency and potential risks of the collaborative network, providing a quantitative basis for graph analysis and optimization.

[0040] By applying graph embedding and graph convolutional network algorithms, key influencing factors and potential imbalance nodes are identified, forming a local collaborative detection map. The system employs graph embedding technology and graph convolutional network algorithm to perform deep learning processing on the constructed local collaborative detection map, thereby uncovering key factors and potential imbalance nodes that affect the collaborative efficiency of guardrail detection. Through high-dimensional feature extraction and pattern recognition of the graph structure, the system accurately identifies collaborative bottlenecks and abnormal nodes, ultimately forming a local collaborative detection map with diagnostic and optimization guidance value.

[0041] S5: Based on the local collaborative detection map, combined with the potential detection imbalance nodes and the global occlusion probability fluctuation trend in the guardrail status data stream sequence, the collaborative detection evolution path is intervened and judged, and dynamic obstacle avoidance adjustment and detection accuracy enhancement results are generated.

[0042] The process of generating dynamic obstacle avoidance adjustment and detection accuracy enhancement results in S5 is as follows: Based on the key imbalance nodes and impact paths identified in the local collaborative detection map, and combined with the global occlusion probability fluctuation trend of the current guardrail status data stream, a dynamic obstacle avoidance adjustment strategy is deduced. Based on the key imbalance nodes and their associated impact paths identified in the local collaborative detection map, the system combines the global occlusion probability fluctuation trend in the real-time guardrail status data stream to construct a dynamic obstacle avoidance adjustment strategy inference model. This model accurately captures the relationship between occlusion changes and node collaboration through multi-dimensional spatiotemporal analysis, infers potential risk areas and dynamic obstacle avoidance needs, and thus formulates targeted path adjustment and operation priority strategies to improve the stability and efficiency of the overall detection system.

[0043] Dynamically optimize robot path planning and task allocation by using reinforcement learning or model predictive control algorithms; By utilizing reinforcement learning or model predictive control algorithms, the system adjusts its strategies for dynamic obstacle avoidance, optimizing path planning and task allocation for road cleaning robots and guardrail maintenance robots in real time. Reinforcement learning continuously updates strategies through environmental feedback, maximizing intelligent obstacle avoidance and resource scheduling. MPC optimizes control inputs based on future state predictions, ensuring the safety and smoothness of the robot's path in complex environments, thereby improving the dynamic response capability and execution efficiency of collaborative operations.

[0044] By combining signal interference suppression technology and adjusting the operating frequency and timing of multiple sensors, the sensing error and false trigger rate can be reduced. By combining multi-sensor signal interference suppression technology, the system intelligently adjusts the operating frequency and timing of different sensors. By analyzing the interference patterns and error distribution of signals between sensors, it dynamically adjusts the sampling frequency and time synchronization mechanism, effectively reducing signal overlap and cross-interference, reducing perception errors and false triggering rates, ensuring high-precision input of multi-modal sensor data, and further improving the perception quality and accuracy of guardrail detection.

[0045] Taking into account both operational efficiency and detection accuracy, the collaborative detection parameters are adjusted in real time to form a closed-loop feedback control, and finally outputs a dynamic result report that includes obstacle avoidance path updates and improved detection accuracy. The system comprehensively considers operational efficiency and detection accuracy. Based on real-time feedback and performance indicators, it dynamically adjusts key parameters of collaborative detection, including robot movement speed, sensor sampling frequency, and task priority, and constructs a closed-loop feedback control mechanism. This mechanism enables continuous optimization and adaptive adjustment of obstacle avoidance paths and detection strategies, ensuring that the system responds efficiently to environmental changes. Finally, it outputs a dynamic result report that includes obstacle avoidance path updates and improved detection accuracy, providing solid data support and technical assurance for subsequent maintenance scheduling and decision-making.

[0046] Example 2: Please see Figure 2As shown in the embodiment of the present invention, a guardrail obstacle avoidance maintenance system based on a smart city includes: Perception and Recognition Module: Utilizing multi-source heterogeneous perception units combined with occlusion semantic segmentation algorithms, it enables dynamic and accurate recognition and time-synchronized calibration of guardrail and robot shapes and occlusion areas; The trajectory tracking module performs multi-device cross-frame correlation on the guardrail status data stream, integrates robot motion features and task priorities, and constructs a multi-dimensional temporal feature vector for device collaboration. Anomaly clustering module: Based on multi-dimensional feature vectors, a dynamic occlusion recovery clustering algorithm is used to extract the collaborative detection state evolution trajectory and detect high-risk accuracy decay candidate events; Map construction module: Delineates the impact area of ​​high-risk events, integrates information on traffic priority, field reconstruction and signal suppression nodes, and constructs a local collaborative detection map; Dynamic Response Module: Combining local map data and global occlusion fluctuation trends, it implements collaborative detection path intervention to generate real-time obstacle avoidance adjustment and detection accuracy enhancement schemes.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. A method for maintaining guardrails with obstacle avoidance based on smart cities, characterized in that, Includes the following steps: The multi-source heterogeneous sensing unit dynamically identifies the guardrail maintenance robot, road cleaning robot and guardrail target structure in the shared shoulder operation area, and extracts the robot's shape boundary, occlusion overlap area and guardrail visible area based on the occlusion semantic segmentation algorithm, and obtains the guardrail status data stream sequence after time synchronization calibration. Multi-device cross-frame correlation tracking is performed on the guardrail status data stream sequence, and a multi-dimensional temporal feature vector of the device collaborative detection status is constructed by combining the motion trajectory characteristics of the road cleaning robot, the priority parameters of the guardrail maintenance task and the distribution of occlusion mode. Based on multidimensional temporal feature vectors, a dynamic occlusion recovery clustering algorithm is used to extract the state evolution trajectory in the collaborative detection process. By identifying the deviation between the trajectory distribution anomaly and the expected detection time series, high-risk accuracy decay candidate events are detected. The guardrail detection section and its impact range corresponding to the high-risk accuracy attenuation candidate event are defined as candidate intervention areas. The passage priority nodes, field reconstruction nodes and signal interference suppression nodes in the candidate area are integrated to construct a local collaborative detection map. Based on the local collaborative detection map, combined with the fluctuation trend of global occlusion probability in the potential detection imbalance nodes and the guardrail status data stream sequence, the collaborative detection evolution path is intervened and judged, and dynamic obstacle avoidance adjustment and detection accuracy enhancement results are generated.

2. The method for maintaining guardrails based on smart city principles according to claim 1, characterized in that, The process of extracting the robot's shape boundary, occlusion overlap region, and guardrail visible area based on the occlusion semantic segmentation algorithm is as follows: In the multi-source heterogeneous sensing unit, lidar point cloud, RGB-D camera and millimeter-wave radar are used to collect three-dimensional environmental data of the road shoulder area respectively; A fusion-based semantic segmentation network is applied to the collected environmental data, combining convolutional neural networks and conditional random field models to achieve pixel-level segmentation and regional occlusion reconstruction of the robot and guardrail boundaries. Based on temporal optical flow analysis, dynamic obstructions and static guardrail structures are distinguished, and the confidence level of overlapping obstruction areas is marked to extract the continuity and integrity indicators of the guardrail's visible area. The segmentation results are combined with multi-sensor fusion localization information to output the robot's external boundary contour, occluded area polygons, and complete field-of-view mask of the guardrail.

3. The guardrail obstacle avoidance maintenance method based on smart cities according to claim 2, characterized in that, The process of obtaining the time-synchronized guardrail status data stream sequence is as follows: A unified timestamp synchronization mechanism is used to correct time discrepancies in data collected from different sensors. The corrected multimodal data is then transformed to spatial coordinates and mapped to a unified three-dimensional geographic coordinate system. By combining dynamic filtering algorithms, the influence of sensor noise and environmental interference on the data is filtered out, forming a sequence of guardrail status data streams.

4. The guardrail obstacle avoidance maintenance method based on smart cities according to claim 3, characterized in that, Combining the motion trajectory characteristics of the road cleaning robot, the priority parameters of the guardrail maintenance task, and the distribution process of occlusion patterns, the process is as follows: Based on the trajectory aggregation algorithm, multi-scale analysis is performed on the historical and real-time trajectory data of road cleaning robots and guardrail maintenance robots to extract trajectory density, speed fluctuation and turning characteristics. Using a priority model, the robot's path priority is dynamically adjusted based on the urgency of the guardrail maintenance task and the risk level of the detection area; Construct an occlusion pattern probability distribution map and combine trajectory features to predict potential occlusion periods and areas; Simultaneous analysis of motion trajectory, task priority, and occlusion pattern forms a fusion scheduling strategy input that covers multiple parameters.

5. A method for maintaining guardrails based on smart city principles according to claim 4, characterized in that, The process of constructing a multidimensional temporal feature vector of the device collaborative detection status is as follows: The trajectory information, occlusion semantic segmentation results, and task priority indicators of road cleaning robots and guardrail maintenance robots in the spatiotemporal domain are mapped into a unified feature encoding format. By employing time-series modeling methods and using recurrent neural networks or Transformer architectures, we can capture the collaborative relationships and temporal dependencies between devices and extract multi-dimensional time-series feature vectors. By integrating signal strength, error distribution, and environmental interference indicators from multiple sensors, it reflects the dynamic state evolution of multi-device collaborative detection in real time.

6. A method for maintaining guardrails based on smart city obstacle avoidance according to claim 5, characterized in that, The process of extracting the state evolution trajectory in the collaborative detection process using the dynamic occlusion recovery clustering algorithm is as follows: Based on multidimensional temporal feature vectors, a spatiotemporal density peak clustering algorithm is used to identify typical state patterns and abnormal state distributions during the guardrail detection process. By combining the occlusion recovery model, interpolation and compensation are performed on the detection data affected by occlusion to restore the complete detection trajectory; By employing trajectory segmentation and dynamic threshold adjustment mechanisms, the system accurately captures state transition nodes and persistent abnormal intervals, outputting a collaboratively detected state evolution trajectory that covers the entire time series.

7. A method for maintaining guardrails based on smart city obstacle avoidance according to claim 6, characterized in that, The process of detecting high-risk accuracy decay candidate events is as follows: By comparing and analyzing the evolution trajectory of the collaborative detection state with the preset detection accuracy baseline model, spatiotemporal segments where accuracy significantly decreases are identified. A multi-factor weighted scoring mechanism is introduced to comprehensively consider the degree of occlusion, sensor signal interference, and path complexity, and to assign a risk level to candidate events. Anomaly detection algorithms are used to determine the occurrence time and duration of high-risk precision decay candidate events, which are then integrated into the subsequent intervention decision-making module.

8. A method for maintaining guardrails based on smart city obstacle avoidance according to claim 7, characterized in that, The process of constructing a local collaborative detection map is as follows: Based on the detection zones defined by high-risk candidate events, extract the guardrail structure nodes, operation robot nodes, and environmental perception nodes within the zones; Based on the spatial adjacency relationships between nodes, task collaboration dependencies, and signal interference propagation paths, a multimodal node connection edge set is constructed to form a graph structure that reflects the collaborative logic of guardrail detection. By combining historical operation records and real-time status data, weights are assigned to nodes and edges to express collaborative efficiency and interference intensity; By applying graph embedding and graph convolutional network algorithms, key influencing factors and potential imbalance nodes are identified, forming a local collaborative detection map.

9. A method for maintaining guardrails based on smart city principles according to claim 8, characterized in that, The process of generating dynamic obstacle avoidance adjustment and detection accuracy enhancement results is as follows: Based on the key imbalance nodes and impact paths identified in the local collaborative detection map, and combined with the global occlusion probability fluctuation trend of the current guardrail status data stream, a dynamic obstacle avoidance adjustment strategy is deduced. Dynamically optimize robot path planning and task allocation by using reinforcement learning or model predictive control algorithms; By combining signal interference suppression technology and adjusting the operating frequency and timing of multiple sensors, the sensing error and false trigger rate can be reduced. Taking into account both operational efficiency and detection accuracy, the collaborative detection parameters are adjusted in real time to form a closed-loop feedback control, ultimately outputting a dynamic result report that includes obstacle avoidance path updates and improved detection accuracy.

10. A guardrail obstacle avoidance maintenance system based on smart cities, applied to the method described in any one of claims 1-9, characterized in that, include: Perception and Recognition Module: Utilizing multi-source heterogeneous perception units combined with occlusion semantic segmentation algorithms, it enables dynamic and accurate recognition and time-synchronized calibration of guardrail and robot shapes and occlusion areas; The trajectory tracking module performs multi-device cross-frame correlation on the guardrail status data stream, integrates robot motion features and task priorities, and constructs a multi-dimensional temporal feature vector for device collaboration. Anomaly clustering module: Based on multi-dimensional feature vectors, a dynamic occlusion recovery clustering algorithm is used to extract collaborative detection state evolution trajectories and detect high-risk accuracy decay candidate events; Map construction module: Delineates the impact area of ​​high-risk events, integrates information on traffic priority, field reconstruction and signal suppression nodes, and constructs a local collaborative detection map; Dynamic Response Module: Combining local map data and global occlusion fluctuation trends, it implements collaborative detection path intervention to generate real-time obstacle avoidance adjustment and detection accuracy enhancement schemes.

Images