Cloud-edge collaborative AI-based field vehicle safety operation intelligent monitoring and adaptive scheduling management platform

Abstract

The application provides a kind of based on cloud edge collaborative AI's field vehicle safety operation intelligent monitoring and adaptive scheduling management platform, comprising: data analysis and alarm module, based on the AI model deployed on field vehicle intelligent terminal, the behavior of driver and the surrounding environment of vehicle are analyzed, and when there is abnormal driving behavior or safety risk, immediate alarm is carried out;Global analysis index determination module, the behavior recognition result obtained by analysis and the time series data of environmental risk information are uploaded to cloud platform, the time series data of each field vehicle in space area is aggregated and mined based on cloud platform, and the global analysis index of the work efficiency and safety situation of field vehicle is obtained;Scheduling management module, based on global analysis index, dynamically generate task scheduling instruction and path planning instruction of each field vehicle, and issue task scheduling instruction and path planning instruction to field vehicle intelligent terminal for operation behavior collaborative management. Significantly improve operation safety and operational efficiency, reduce human error and resource waste.

Description

Technical Field

[0001] This invention relates to the field of monitoring and data processing technology, and in particular to an intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI. Background Technology

[0002] Currently, with the rapid development of industries such as logistics warehousing and manufacturing, the frequency and density of use of yard vehicles as important on-site operation tools are increasing day by day. However, the operating environment of yard vehicles is usually complex, with people and vehicles mixed together, resulting in a high risk of safety accidents. Traditional safety management models mainly rely on manual supervision, fixed surveillance cameras, and simple overspeed alarm devices. These models have prominent problems such as large blind spots, delayed warnings, and reliance on personnel experience, making it difficult to achieve real-time and accurate perception and intervention of unsafe driver behavior and risks in complex environments. In addition, existing vehicle dispatch management focuses on task allocation and route planning to improve logistics efficiency. It is usually based on static task lists and map information, lacking deep integration with real-time safety situation. The dispatch system cannot perceive changes in driver status, sudden risks in the surrounding area, and fluctuations in the overall safety level of the area. This may cause dispatch instructions to inadvertently lead vehicles to high-risk areas or exacerbate regional conflicts, resulting in the risk of reduced safety and efficiency. Therefore, in order to overcome the above-mentioned technical problems, the present invention provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles. Summary of the Invention

[0003] This invention provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles. Through cloud-edge collaborative AI technology, it enables intelligent management of the entire parking vehicle operation process. At the edge terminal, AI models analyze driver behavior and the surrounding environment in real time. Upon detecting anomalies or risks, alarms are immediately triggered, effectively preventing safety accidents. Simultaneously, the real-time analyzed time-series data is uploaded to the cloud for aggregation and in-depth analysis, generating global indicators reflecting overall operational efficiency and safety status, thus providing decision support. Based on these global indicators, the platform dynamically adjusts the task allocation and driving paths of each vehicle and issues instructions to the terminal for execution, achieving collaborative operation and adaptive scheduling between parking lots and vehicles. This significantly improves operational safety and efficiency, and reduces human error and resource waste.

[0004] This invention provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, comprising: The data analysis and alarm module is used to perform real-time analysis of driver behavior and the surrounding environment of the vehicle based on the AI ​​model deployed on the intelligent terminal of the parking lot, and to issue an immediate alarm when there is abnormal driving behavior or safety risk. The global analysis index determination module is used to upload the time-series data of the behavior recognition results and environmental risk information obtained from the analysis to the cloud platform, and aggregate and mine the time-series data of each vehicle in the spatial area based on the cloud platform to obtain global analysis indexes of the vehicle's operating efficiency and safety status. The scheduling and management module is used to dynamically generate task scheduling instructions and route planning instructions for each vehicle based on global analysis indicators, and to send the task scheduling instructions and route planning instructions to the intelligent terminal of the vehicle for collaborative management of operation behavior.

[0005] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a data analysis and alarm module, comprising: Image acquisition unit, used for: The vehicle uses a camera device to collect real-time images of the driver's face and posture, as well as images of the surrounding environment. Keyframe extraction and face region localization are performed on the acquired facial and pose image sequences, and the facial and pose image sequences and environmental image sequences are standardized and preprocessed based on the data input conditions of the AI ​​model. The data analysis unit is used for: The preprocessed keyframes, face region localization results, and data to be analyzed are input into the AI ​​model deployed on the intelligent terminal of the parking lot. Based on the behavior recognition layer in the AI ​​model, the keyframes and face region localization results are identified and analyzed to determine the driver's behavior recognition results. Meanwhile, the environmental perception layer in the AI ​​model is used to analyze the preprocessed environmental image sequence to obtain environmental risk information around the vehicle. The alarm unit is used for alarm management based on behavior recognition results and environmental risk information.

[0006] Preferably, an intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI includes an alarm unit comprising: The alarm judgment subunit is used for: Based on the management terminal, the alarm thresholds corresponding to driving behavior and environmental safety risks are obtained respectively, and the obtained behavior recognition results and environmental risk information are compared with the corresponding alarm warning thresholds respectively. When the identification result or environmental risk information is greater than or equal to the corresponding alarm threshold, the alarm condition is determined to be met. The alarm execution subunit is used for: Based on the judgment result, a local alarm command is generated, and based on the local alarm command, the audible and visual alarm device on the parking vehicle is activated to provide an immediate warning.

[0007] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a data analysis unit comprising: Behavior determination subunit, used for: The standardized preprocessed facial and pose image frames are input into the feature extraction sublayer in the behavior recognition layer, and the spatial features of the facial and pose image frames are extracted layer by layer based on the convolutional neural network. Based on the key point localization layer in the behavior recognition layer, the coordinates of the driver's facial key points and the Euler angles of the head posture are determined according to the extracted spatial features. Based on the facial and posture image sequence, the coordinates of the facial key points and the Euler angles of the head posture of each image frame are summarized. Furthermore, based on the temporal convolutional network, the summarization results are analyzed for temporal features to determine the dynamic changes of the driver's blinking frequency, eye-closing duration and head offset angle. Based on a preset classification algorithm, the continuous temporal features obtained are mapped into discrete driver state labels according to the dynamic change pattern, and the corresponding behavior recognition results are obtained. The labels include normal, fatigued, looking around, and using a handheld device. The environmental risk determination subunit is used for: The standardized preprocessed environmental image sequence is input into the environment perception layer, and based on depthwise separable convolution operation, an environmental feature map containing multiple scales is obtained; Based on the environmental feature map, pedestrians, fixed obstacles, and the boundaries of drivable and dangerous areas in the image are identified and bounded. Based on the pixel coordinates of the bounding target detection box, known camera intrinsic parameters, and preset prior knowledge of actual size, the relative distance and azimuth angle between pedestrians or obstacles and the field vehicle are evaluated. The relative distance and azimuth are analyzed based on a preset safety threshold function for distance and speed to obtain the instantaneous risk level of each detected target or area; The environmental risk information around the vehicle is obtained by summarizing the target type, relative distance and azimuth angle between the detected target and the vehicle.

[0008] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a global analysis indicator determination module, comprising: Data transmission unit, used for: Obtain the vehicle identification number of each vehicle and extract the timestamps corresponding to the behavioral recognition results of the drivers on each vehicle and the environmental risk information; The vehicle identification, behavior recognition results, environmental risk information, and corresponding timestamps for each vehicle are packaged together and uploaded to the cloud platform via a wireless communication network. The data analysis unit is used for: Based on the standardized rules preset on the cloud platform, the received behavior recognition results and environmental risk information are formatted and normalized. At the same time, a vehicle time series database is built for each vehicle on the cloud platform, and the formatted and normalized behavior recognition results and environmental risk information are stored in the corresponding vehicle time series database based on the timestamp and vehicle identifier. Based on the aggregation computing engine, the vehicle time series database is periodically scanned, and statistical calculations are performed on the same data of all vehicles in the specified time window and specified spatial area based on the periodic scan results to obtain the primary aggregation index of each vehicle. The primary aggregation index includes the total frequency of fatigue driving events, the incidence rate of high-risk behaviors, and the density of regional risk points. The global indicator determination unit is used for: Based on a deep learning model, primary aggregation indicators are integrated with historical operation pattern database and static environment map information to determine the spatiotemporal correlation between safety events and predict the evolution trend of systemic risks. Extract the metadata of the operation tasks of each vehicle in the yard, and perform joint analysis of the spatiotemporal correlation, risk evolution trend and operation task metadata based on a multi-objective optimization algorithm to obtain global analysis indicators of the operation efficiency and safety status of the vehicles in the yard.

[0009] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a global indicator determination unit, comprising: The indicator analysis subunit is used for: Extract the specified time window and specified spatial area corresponding to the primary aggregation index, and retrieve the historical operation load and historical event sequence that match the specified time window and specified spatial area from the historical operation mode library. At the same time, extract the topology, key node location and functional area attributes of each spatial area from the static environment map information to obtain multi-source heterogeneous data. Multi-source heterogeneous data is converted into numerical vectors of a unified dimension, and the numerical vectors are concatenated according to preset rules to obtain a fused feature vector. Based on the deep learning model, the fused feature vectors are converted into a relationship graph of regional nodes and edges according to the topology of each spatial region. Based on the relationship graph of regional nodes and edges, the spatial region dependencies are determined. At the same time, based on the spatial region dependencies, the temporal dependencies between historical event sequences and the current status of the field vehicle are determined. Based on spatial region dependency and temporal dependency, the association weights between different security event characteristics are determined, and the association patterns between security events are determined based on the association weights, resulting in a spatiotemporal association rule table for security events. Based on a deep learning model, iterative analysis is performed on the fused feature vector and the spatiotemporal correlation rule table of security events to obtain the risk evolution probability distribution in each spatial region, and the predicted systemic risk evolution trend is obtained based on the risk evolution probability distribution.

[0010] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a global analysis indicator determination module, comprising: The result acquisition unit is used to acquire global analysis indicators of the operating efficiency and safety status of the field vehicles, and extract the time nodes corresponding to the global analysis indicators. The results recording unit is used to construct the indicator recording table and to sequentially enter different time nodes and their corresponding global analysis indicators into the indicator recording table for storage based on the development order of time nodes.

[0011] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a scheduling management module comprising: The optimization target determination unit is used to obtain the obtained global analysis indicators, and at the same time, obtain the operation status information of each vehicle and the queue of tasks to be executed, and determine the optimization target when scheduling tasks based on the global analysis indicators, the operation status information of each vehicle and the queue of tasks to be executed. The task scheduling instruction generation unit is used to construct a multi-objective decision model based on the optimization objective, and to analyze the global analysis indicators, the operation status information of each vehicle and the queue of tasks to be executed based on the objective decision model to obtain a macro scheduling scheme, and to generate task scheduling instructions based on the macro scheduling scheme. The path planning instruction generation unit is used for: Obtain the static environment map corresponding to the scene's working area, and use the macro scheduling scheme as a constraint to generate a collision-free path from the current location to the target point for each vehicle in the field. Assess the estimated travel time and overall risk value of each route, and obtain the optimal route for each vehicle based on the assessment results; Extract the path point sequence, speed suggestions, and risk warnings corresponding to the optimal path, and generate path planning instructions for each vehicle based on the path point sequence, speed suggestions, and risk warnings.

[0012] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a scheduling management module comprising: The instruction issuing unit is used to encapsulate the task scheduling instructions and route planning instructions of the same vehicle based on the identity tags of each vehicle, and to send the encapsulation results to the corresponding intelligent terminal of the vehicle based on the communication interface of the cloud platform. The collaborative management unit is used to configure the working parameters of each vehicle based on the issued results, and to manage the tasks and routes of each vehicle based on the configuration results. The management results are then fed back to the cloud platform to complete the collaborative management of work behavior.

[0013] Preferably, a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles includes a collaborative management unit comprising: The tracking subunit is used to periodically obtain the operating status parameters of each vehicle after the task and route of each vehicle have been effectively managed, and to re-analyze the operating status parameters to obtain global analysis indicators of the operating efficiency and safety status of the vehicle under the current status. The real-time optimization management subunit is used to compare global analysis indicators with preset thresholds, and to replan the tasks and routes of each vehicle when a global analysis indicator fails to meet the preset threshold.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: By leveraging cloud-edge collaborative AI technology, intelligent management of the entire process of vehicle operations is achieved. At the edge terminal, AI models analyze driver behavior and the surrounding environment in real time. Once an anomaly or risk is detected, an alarm is triggered immediately, effectively preventing safety accidents. At the same time, the real-time analyzed time-series data is uploaded to the cloud for aggregation and in-depth analysis, generating global indicators reflecting overall operational efficiency and safety status, thereby providing decision support. Based on the global indicators of operational efficiency and safety status, the platform dynamically adjusts the task allocation and driving routes of each vehicle and issues instructions to the terminal for execution, realizing collaborative operation and adaptive scheduling between the vehicle and the terminal. This significantly improves operational safety and efficiency, and reduces human error and resource waste.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, as described in an embodiment of the present invention. Figure 2 This is a structural diagram of the data analysis and alarm module in a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, as described in an embodiment of the present invention. Figure 3 This is a structural diagram of the global analysis index determination module in a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, as described in an embodiment of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] Example 1: This example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, such as... Figure 1 As shown, it includes: The data analysis and alarm module is used to perform real-time analysis of driver behavior and the surrounding environment of the vehicle based on the AI ​​model deployed on the intelligent terminal of the parking lot, and to issue an immediate alarm when there is abnormal driving behavior or safety risk. The global analysis index determination module is used to upload the time-series data of the behavior recognition results and environmental risk information obtained from the analysis to the cloud platform, and aggregate and mine the time-series data of each vehicle in the spatial area based on the cloud platform to obtain global analysis indexes of the vehicle's operating efficiency and safety status. The scheduling and management module is used to dynamically generate task scheduling instructions and route planning instructions for each vehicle based on global analysis indicators, and to send the task scheduling instructions and route planning instructions to the intelligent terminal of the vehicle for collaborative management of operation behavior.

[0020] In this embodiment, "site vehicle" refers to a vehicle that operates in a specific location (such as a factory, warehouse, port, etc.), such as a forklift, a pallet truck, or a tractor.

[0021] In this embodiment, the AI ​​model refers to the artificial intelligence algorithm model deployed on the intelligent terminal of the parking lot, which is used to identify driver behavior (such as fatigue driving, illegal operation) and risks in the surrounding environment of the vehicle (such as obstacles, pedestrians) in real time.

[0022] In this embodiment, time-series data refers to a data sequence of behavior identification results and environmental risk information recorded in chronological order, which is used to trace and analyze the dynamic changes during the operation of the parking lot vehicle.

[0023] In this embodiment, the cloud platform refers to a cloud-based backend service platform that is responsible for receiving, storing, and processing time-series data from multiple parking lots, and performing aggregation and mining calculations.

[0024] In this embodiment, the global analysis index refers to the quantitative index obtained by comprehensively analyzing all vehicle data in the spatial area through the cloud, which is used to evaluate the overall operational efficiency (such as task completion rate and vehicle utilization rate) and safety status (such as risk event frequency and alarm distribution).

[0025] In this embodiment, the task scheduling instruction refers to the task allocation command that is dynamically generated based on global analysis indicators and issued to the site vehicles, which is used to coordinate the work sequence and load of each vehicle.

[0026] In this embodiment, the route planning instruction refers to the optimized driving route command that is dynamically generated based on global analysis indicators and issued to the parking vehicle to avoid collisions, congestion, or dangerous areas.

[0027] In this embodiment, the intelligent terminal of the parking lot vehicle refers to an intelligent device installed on the parking lot vehicle, which has AI analysis, data collection and communication functions, and is used to perform local analysis and receive cloud instructions.

[0028] In this embodiment, the spatial area refers to the physical scope of the vehicle operation, such as a specific geographical area within a warehouse, factory area, or logistics park.

[0029] The beneficial effects of the above technical solution are as follows: Through cloud-edge collaborative AI technology, intelligent management of the entire process of vehicle operations is achieved; AI models at the edge terminal analyze driver behavior and the surrounding environment in real time, triggering alarms immediately upon detecting anomalies or risks, effectively preventing safety accidents; simultaneously, the real-time analyzed time-series data is uploaded to the cloud for aggregation and in-depth analysis, generating global indicators reflecting overall operational efficiency and safety status, thereby providing decision support; based on these global indicators of operational efficiency and safety status, the platform dynamically adjusts the task allocation and driving paths of each vehicle and issues instructions to the terminal for execution, realizing collaborative operation and adaptive scheduling between the vehicle and the terminal, significantly improving operational safety and efficiency, and reducing human error and resource waste.

[0030] Example 2: Based on Example 1, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, such as... Figure 2 As shown, the data analysis and alarm module includes: Image acquisition unit, used for: The vehicle uses a camera device to collect real-time images of the driver's face and posture, as well as images of the surrounding environment. Keyframe extraction and face region localization are performed on the acquired facial and pose image sequences, and the facial and pose image sequences and environmental image sequences are standardized and preprocessed based on the data input conditions of the AI ​​model. The data analysis unit is used for: The preprocessed keyframes, face region localization results, and data to be analyzed are input into the AI ​​model deployed on the intelligent terminal of the parking lot. Based on the behavior recognition layer in the AI ​​model, the keyframes and face region localization results are identified and analyzed to determine the driver's behavior recognition results. Meanwhile, the environmental perception layer in the AI ​​model is used to analyze the preprocessed environmental image sequence to obtain environmental risk information around the vehicle. The alarm unit is used for alarm management based on behavior recognition results and environmental risk information.

[0031] In this embodiment, environmental risk information includes the proximity of pedestrians, obstacles, or dangerous areas around the vehicle.

[0032] In this embodiment, the behavior recognition layer refers to the part of the AI ​​model specifically used to analyze the driver's facial and posture images. It determines the driver's behavior state, such as fatigue, distraction, or violation of regulations, by recognizing key frames and facial regions.

[0033] In this embodiment, the environmental perception layer refers to the part of the AI ​​model specifically used to analyze images of the vehicle's surrounding environment, detect pedestrians, obstacles, or dangerous areas in the environment, and assess their proximity to generate environmental risk information.

[0034] In this embodiment, the behavior recognition result refers to the driver status information obtained through behavior recognition layer analysis, including driver fatigue, distracted actions, or violations.

[0035] The beneficial effects of the above technical solution are as follows: by collecting images of the driver's face, posture, and the surrounding environment of the vehicle in real time, and extracting key frames and face regions for standardized processing, the AI ​​model is used to identify driver fatigue, distraction, or violations, while also sensing risks such as pedestrians and obstacles in the surrounding area, thereby achieving immediate alarms; effectively improving the accuracy of driver behavior monitoring and the real-time nature of environmental risk detection, preventing safety accidents, and enhancing the overall safety of the work site.

[0036] Example 3. Based on Example 2, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, including an alarm unit comprising: The alarm judgment subunit is used for: Based on the management terminal, the alarm thresholds corresponding to driving behavior and environmental safety risks are obtained respectively, and the obtained behavior recognition results and environmental risk information are compared with the corresponding alarm warning thresholds respectively. When the identification result or environmental risk information is greater than or equal to the corresponding alarm threshold, the alarm condition is determined to be met. The alarm execution subunit is used for: Based on the judgment result, a local alarm command is generated, and based on the local alarm command, the audible and visual alarm device on the parking vehicle is activated to provide an immediate warning.

[0037] In this embodiment, the alarm threshold refers to the critical value that is preset by the management terminal to trigger an alarm, and is set for driving behavior (such as fatigue level) and environmental safety risks (such as obstacle distance).

[0038] In this embodiment, the management terminal refers to a device or system interface used by administrators to configure and issue various alarm thresholds and other system parameters.

[0039] In this embodiment, the local alarm command refers to the command signal generated by the intelligent terminal of the parking lot when the alarm conditions are met, which is used to directly control the activation of the local alarm device.

[0040] In this embodiment, the audible and visual alarm device refers to a physical device installed on the vehicle that can emit sound (such as a buzzer) and light (such as a warning light) signals simultaneously or separately for on-site immediate warning.

[0041] The beneficial effects of the above technical solution are as follows: by comparing the behavioral and environmental data obtained from real-time analysis with preset, flexibly configurable safety thresholds, the risk assessment is standardized and objective; once the threshold is reached or exceeded, the system will automatically trigger local audible and visual warnings, providing drivers with immediate and clear on-site reminders, effectively avoiding response delays caused by subjective judgment delays or omissions, significantly improving the real-time performance and accuracy of risk warnings, thereby intervening in potential dangers at the first moment, preventing accidents from occurring or escalating, and ensuring the safety of people and vehicles at the work site.

[0042] Example 4: Based on Example 2, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, including a data analysis unit: Behavior determination subunit, used for: The standardized preprocessed facial and pose image frames are input into the feature extraction sublayer in the behavior recognition layer, and the spatial features of the facial and pose image frames are extracted layer by layer based on the convolutional neural network. Based on the key point localization layer in the behavior recognition layer, the coordinates of the driver's facial key points and the Euler angles of the head posture are determined according to the extracted spatial features. Based on the facial and posture image sequence, the coordinates of the facial key points and the Euler angles of the head posture of each image frame are summarized. Furthermore, based on the temporal convolutional network, the summarization results are analyzed for temporal features to determine the dynamic changes of the driver's blinking frequency, eye-closing duration and head offset angle. Based on a preset classification algorithm, the continuous temporal features obtained are mapped into discrete driver state labels according to the dynamic change pattern, and the corresponding behavior recognition results are obtained. The labels include normal, fatigued, looking around, and using a handheld device. The environmental risk determination subunit is used for: The standardized preprocessed environmental image sequence is input into the environment perception layer, and based on depthwise separable convolution operation, an environmental feature map containing multiple scales is obtained; Based on the environmental feature map, pedestrians, fixed obstacles, and the boundaries of drivable and dangerous areas in the image are identified and bounded. Based on the pixel coordinates of the bounding target detection box, known camera intrinsic parameters, and preset prior knowledge of actual size, the relative distance and azimuth angle between pedestrians or obstacles and the field vehicle are evaluated. The relative distance and azimuth are analyzed based on a preset safety threshold function for distance and speed to obtain the instantaneous risk level of each detected target or area; The environmental risk information around the vehicle is obtained by summarizing the target type, relative distance and azimuth angle between the detected target and the vehicle.

[0043] In this embodiment, spatial features refer to multidimensional data representations extracted from single-frame facial and pose images through convolutional neural networks, which can characterize key information in the image (such as eye and mouth contours, head orientation).

[0044] In this embodiment, the key point localization layer refers to a specific network layer in the behavior recognition layer. Its function is to output the pixel coordinates of predefined facial key points (such as the corners of the eyes and mouth) and Euler angles representing the head rotation angle based on the input spatial features.

[0045] In this embodiment, Euler angles refer to a mathematical representation of the attitude of an object (in this case, the driver's head) in three-dimensional space using three angles of rotation about an axis (usually yaw, pitch, and roll).

[0046] In this embodiment, temporal convolutional network refers to a convolutional neural network structure specifically designed for processing temporal data (such as feature sequences of consecutive image frames), which can capture the dynamic changes of data in the time dimension.

[0047] In this embodiment, depthwise separable convolution operation refers to a lightweight convolution calculation method that decomposes standard convolution into two steps: depthwise convolution and pointwise convolution, which can significantly reduce the amount of computation while maintaining certain model performance.

[0048] In this embodiment, the environmental feature map refers to the set of feature matrices obtained by performing convolution operations on the environmental image sequence, which contains semantic information at different levels of abstraction (such as edges, shapes, and target components).

[0049] In this embodiment, the target detection box refers to a rectangular area on the environmental feature map or the original image used to frame and locate the identified target (such as a pedestrian or obstacle), which is usually represented by pixel coordinates.

[0050] In this embodiment, the risk level refers to a level index calculated by a preset safety model based on the relative distance, azimuth angle, and speed of movement between the detected target and the vehicle, which is used to quantify the current degree of danger.

[0051] The beneficial effects of the above technical solution are as follows: Firstly, by continuously extracting spatiotemporal features of the face and posture, and using a time-series model to analyze dynamic patterns, it can accurately identify subtle and continuous state changes from normal to fatigued or distracted states, greatly improving the finesse and reliability of behavior recognition. Secondly, by utilizing efficient convolutional networks to extract multi-scale environmental features and combining geometric principles to convert image coordinates into distance and orientation in real space, it achieves a quantitative assessment of the risk level of targets such as pedestrians and obstacles. This provides a high-confidence data foundation for subsequent accurate alarms and dispatching, enabling safety monitoring to move from qualitative perception to quantitative decision-making, effectively preventing safety accidents caused by subtle changes in state or misjudgments of distance.

[0052] Example 5: Based on Example 1, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, such as... Figure 3 As shown, the global analysis indicator determination module includes: Data transmission unit, used for: Obtain the vehicle identification number of each vehicle and extract the timestamps corresponding to the behavioral recognition results of the drivers on each vehicle and the environmental risk information; The vehicle identification, behavior recognition results, environmental risk information, and corresponding timestamps for each vehicle are packaged together and uploaded to the cloud platform via a wireless communication network. The data analysis unit is used for: Based on the standardized rules preset on the cloud platform, the received behavior recognition results and environmental risk information are formatted and normalized. At the same time, a vehicle time series database is built for each vehicle on the cloud platform, and the formatted and normalized behavior recognition results and environmental risk information are stored in the corresponding vehicle time series database based on the timestamp and vehicle identifier. Based on the aggregation computing engine, the vehicle time series database is periodically scanned, and statistical calculations are performed on the same data of all vehicles in the specified time window and specified spatial area based on the periodic scan results to obtain the primary aggregation index of each vehicle. The primary aggregation index includes the total frequency of fatigue driving events, the incidence rate of high-risk behaviors, and the density of regional risk points. The global indicator determination unit is used for: Based on a deep learning model, primary aggregation indicators are integrated with historical operation pattern database and static environment map information to determine the spatiotemporal correlation between safety events and predict the evolution trend of systemic risks. Extract the metadata of the operation tasks of each vehicle in the yard, and perform joint analysis of the spatiotemporal correlation, risk evolution trend and operation task metadata based on a multi-objective optimization algorithm to obtain global analysis indicators of the operation efficiency and safety status of the vehicles in the yard.

[0053] In this embodiment, the timestamp refers to the precise time data that records the moment when each behavior identification result or environmental risk information is generated.

[0054] In this embodiment, packetization refers to the process of packaging vehicle identification, data content, and their timestamps into an independent data unit according to a predetermined format and protocol.

[0055] In this embodiment, the vehicle time-series database refers to a database created independently for each parking vehicle on a cloud platform, which stores all its behavioral and environmental-related structured data in chronological order.

[0056] In this embodiment, the aggregation computing engine refers to a dedicated software module on the cloud platform used to periodically scan time-series databases and perform statistical operations (such as summation, counting, and density calculation) on massive amounts of data within specified dimensions and time windows.

[0057] In this embodiment, the primary aggregation index refers to the intermediate results obtained by performing preliminary statistical calculations on the original time-series data, such as the total frequency of fatigue driving events, the incidence of high-risk behaviors, and the density of regional risk points.

[0058] In this embodiment, the historical operation pattern library refers to a data set stored in the cloud that describes the typical operation patterns of the site vehicle in a specific area or time period (such as high-frequency driving routes and regular task cycles).

[0059] In this embodiment, static environment map information refers to digital information about the site environment that is pre-entered into the system and does not change frequently, including road layout, location of fixed facilities, functional area division, etc.

[0060] In this embodiment, spatiotemporal correlation refers to the inherent connection or pattern between the time of occurrence and the geographical location of security events, as discovered through analysis.

[0061] In this embodiment, the systemic risk evolution trend refers to the direction and pattern of overall change in the safety risks of the entire work area over a future period of time, predicted based on current and historical data.

[0062] In this embodiment, the task metadata refers to data that describes the basic attributes of the task performed by the vehicle, such as task type, start and end point, planned duration, priority, etc.

[0063] In this embodiment, the multi-objective optimization algorithm refers to a mathematical solution method used to find the optimal balance between multiple conflicting objectives (such as maximizing operational efficiency and minimizing safety risks), and to estimate the overall operational efficiency coefficient.

[0064] The beneficial effects of the above technical solution are as follows: By systematically uploading data with precise time stamps from each vehicle to the cloud and performing standardized processing and storage, a complete vehicle operation time-series database is constructed; through periodic aggregation calculations, scattered real-time information is transformed into primary statistical indicators reflecting the overall situation of the region; combining prior knowledge such as historical operation patterns and static maps, deep learning is used to mine the spatiotemporal correlation of safety events and predict risk trends; finally, by integrating real-time task data and performing multi-objective optimization, a global indicator that balances efficiency and safety is derived. This enables management decisions to transcend the immediate risks of individual vehicles, gain insight into potential hazards and predict situations from a systemic perspective, thereby providing accurate data-driven basis for forward-looking collaborative scheduling and resource optimization, and comprehensively improving the safety level and operational efficiency of regional operations.

[0065] Example 6: Based on Example 5, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, including a global indicator determination unit: The indicator analysis subunit is used for: Extract the specified time window and specified spatial area corresponding to the primary aggregation index, and retrieve the historical operation load and historical event sequence that match the specified time window and specified spatial area from the historical operation mode library. At the same time, extract the topology, key node location and functional area attributes of each spatial area from the static environment map information to obtain multi-source heterogeneous data. Multi-source heterogeneous data is converted into numerical vectors of a unified dimension, and the numerical vectors are concatenated according to preset rules to obtain a fused feature vector. Based on the deep learning model, the fused feature vectors are converted into a relationship graph of regional nodes and edges according to the topology of each spatial region. Based on the relationship graph of regional nodes and edges, the spatial region dependencies are determined. At the same time, based on the spatial region dependencies, the temporal dependencies between historical event sequences and the current status of the field vehicle are determined. Based on spatial region dependency and temporal dependency, the association weights between different security event characteristics are determined, and the association patterns between security events are determined based on the association weights, resulting in a spatiotemporal association rule table for security events. Based on a deep learning model, iterative analysis is performed on the fused feature vector and the spatiotemporal correlation rule table of security events to obtain the risk evolution probability distribution in each spatial region, and the predicted systemic risk evolution trend is obtained based on the risk evolution probability distribution.

[0066] In this embodiment, multi-source heterogeneous data refers to a collection of data from different sources, structures, and formats, including primary aggregation indicators, matched historical workloads and historical event sequences, as well as topological structures, node locations, and functional area attributes extracted from map information.

[0067] In this embodiment, the fused feature vector refers to a feature vector that comprehensively represents the overall state of a specified time window and spatial region after multi-source heterogeneous data is converted into a numerical vector of a unified dimension and spliced ​​together according to preset rules.

[0068] In this embodiment, the relationship graph between regional nodes and edges refers to a graph data structure constructed based on the topological structure of spatial regions, where each sub-region or its key nodes are represented as "nodes" and the connections or influence relationships between regions are represented as "edges," which is used to model spatial dependencies.

[0069] In this embodiment, spatial regional dependency refers to the mutual influence or association between different geographical sub-regions in terms of security situation, as revealed by the relationship graph of regional nodes and edges.

[0070] In this embodiment, the temporal dependency relationship refers to the correlation or evolutionary pattern of security events in the time dimension, which is derived from the analysis of historical event sequences and the current situation.

[0071] In this embodiment, the association weight refers to a numerical value determined through model learning, used to quantify the strength of the association between different security event characteristics (such as a certain behavior and a certain type of environmental risk).

[0072] In this embodiment, the security event spatiotemporal association rule table refers to the association pattern between security events recorded in a structured form (such as a table). Its rules usually include elements such as event type, spatial relationship, temporal relationship and association weight.

[0073] In this embodiment, the risk evolution probability distribution refers to the quantitative probability distribution of different levels or types of risks occurring in various spatial regions within a future period, as predicted by the model.

[0074] The beneficial effects of the above technical solution are as follows: By deeply integrating real-time aggregated indicators, historical operational patterns, and static environmental information, a multi-dimensional data feature that comprehensively describes the operational scenario is constructed; by using a deep learning model, these features are transformed into a graph structure that reflects spatial topological relationships, thereby accurately depicting the mutual influence between different regions and the temporal dependence between historical and current situations; intelligently analyzing and quantifying the intrinsic correlation weights between different safety event features, forming clear spatiotemporal correlation rules; finally, through iterative analysis, the future risk probability distribution of each region is predicted, achieving a forward-looking judgment on the evolution trend of systemic risks; and providing in-depth decision-making insights for implementing precise proactive intervention and preventive scheduling, greatly improving the predictability and scientific nature of safety management.

[0075] Example 7: Based on Example 1, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, including a global analysis indicator determination module, comprising: The result acquisition unit is used to acquire global analysis indicators of the operating efficiency and safety status of the field vehicles, and extract the time nodes corresponding to the global analysis indicators. The results recording unit is used to construct the indicator recording table and to sequentially enter different time nodes and their corresponding global analysis indicators into the indicator recording table for storage based on the development order of time nodes.

[0076] In this embodiment, a time node refers to the specific moment or time period corresponding to the generation or acquisition of a set of global analysis indicators.

[0077] In this embodiment, the indicator record table refers to a database table or structured file used to systematically store historical global analysis indicators in chronological order.

[0078] The beneficial effects of the above technical solution are as follows: by retaining global analysis indicators at each time point, a complete historical situation archive is constructed, enabling managers to trace changes in operational efficiency and safety levels over different time periods, clearly understand the actual effects and long-term trends of management measures, provide reliable quantitative basis for continuous optimization of scheduling strategies and evaluation of improvement plans, and the standardized records also provide complete data support for post-event analysis, accountability tracing and compliance audits, thereby improving the scientific nature and traceability of overall management.

[0079] Example 8: Based on Example 1, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles. The scheduling management module includes: The system obtains global analysis metrics, as well as the operational status information of each vehicle and the queue of tasks to be executed. Based on the global analysis metrics, the operational status information of each vehicle and the queue of tasks to be executed, the system determines the optimization objectives for task scheduling. A multi-objective decision-making model is constructed based on the optimization objective. Based on the objective decision-making model, global analysis indicators, operational status information of vehicles in each yard, and queue of tasks to be executed are analyzed to obtain a macro-scheduling scheme. Based on the macro-scheduling scheme, task scheduling instructions are generated. Obtain the static environment map corresponding to the scene's working area, and use the macro scheduling scheme as a constraint to generate a collision-free path from the current location to the target point for each vehicle in the field. Assess the estimated travel time and overall risk value of each route, and obtain the optimal route for each vehicle based on the assessment results; Extract the path point sequence, speed suggestions, and risk warnings corresponding to the optimal path, and generate path planning instructions for each vehicle based on the path point sequence, speed suggestions, and risk warnings.

[0080] In this embodiment, the optimization objective is to minimize the total system risk and maximize the overall operational efficiency.

[0081] In this embodiment, the macro scheduling scheme refers to the task execution order of each vehicle, the start time of the operation, and the suggested average operation speed.

[0082] In this embodiment, the multi-objective decision model refers to a mathematical model or algorithmic framework used to weigh and solve multiple conflicting optimization objectives (such as safety and efficiency).

[0083] In this embodiment, the static environment map refers to map information that is pre-digitally entered into the system and reflects the fixed layout within the work area, including the locations of roads, passages, and fixed facilities.

[0084] In this embodiment, a collision-free path refers to a feasible driving route planned in a static environment map, from the starting point to the target point, without conflicting with static obstacles or area boundaries.

[0085] In this embodiment, the estimated travel time refers to the estimated time required for a vehicle to complete a certain route, based on the route length, recommended speed, and historical traffic data.

[0086] In this embodiment, the comprehensive risk value refers to the quantitative value obtained after risk assessment of a certain path. Its calculation comprehensively considers factors such as the density of historical risk points around the path, real-time risk information, and road complexity.

[0087] In this embodiment, the optimal path refers to the best driving route selected from multiple collision-free paths based on preset evaluation criteria (such as balancing the shortest time and the lowest risk).

[0088] In this embodiment, the path point sequence refers to a series of ordered coordinate points that constitute a driving path, used to guide the vehicle from the starting point to the destination.

[0089] In this embodiment, the speed recommendation refers to the recommended driving speed for different road segments in the route planning process to ensure safety or to coordinate with the overall scheduling.

[0090] In this embodiment, risk warning refers to the text or label warning in the route planning instruction for potential risks (such as blind spots or high-frequency accident points) in a specific road segment or area.

[0091] The beneficial effects of the above technical solution are as follows: by integrating real-time global indicators, vehicle status, and pending tasks, it achieves the core objective of balancing system safety and operational efficiency, intelligently generating macro-level task allocation and time plans, and then combining high-precision maps to plan collision-free and optimal driving routes for each vehicle in real time, providing specific speed suggestions and risk warnings; it effectively realizes end-to-end optimization from task allocation to single-vehicle-level path guidance, can dynamically respond to changes on-site, effectively avoid congestion and risk areas, and significantly improve the collaborative operation efficiency and task completion speed of vehicles on site while ensuring overall safety.

[0092] Example 9: Based on Example 1, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles. The scheduling management module includes: The instruction issuing unit is used to encapsulate the task scheduling instructions and route planning instructions of the same vehicle based on the identity tags of each vehicle, and to send the encapsulation results to the corresponding intelligent terminal of the vehicle based on the communication interface of the cloud platform. The collaborative management unit is used to configure the working parameters of each vehicle based on the issued results, and to manage the tasks and routes of each vehicle based on the configuration results. The management results are then fed back to the cloud platform to complete the collaborative management of work behavior.

[0093] In this embodiment, the identity tag refers to a code or identifier registered in the cloud platform and used to uniquely identify and distinguish different vehicles. It usually corresponds to the vehicle identifier and is used for the targeted issuance of instructions.

[0094] In this embodiment, the communication interface refers to the specific communication protocol or hardware / software channel used for stable and secure data transmission between the cloud platform and the intelligent terminal of the parking lot.

[0095] In this embodiment, the operating parameters refer to the specific operating parameters that the intelligent terminal of the parking lot needs to configure in order to execute specific task scheduling instructions and path planning instructions, such as target speed, navigation mode, task priority identifier, etc.

[0096] In this embodiment, effective management refers to the process by which the intelligent terminal of the parking lot receives and parses the instruction, applies the instruction to the vehicle control system, enables the new task and path to begin formal execution, and monitors and confirms the status of the new task and path.

[0097] The beneficial effects of the above technical solution are as follows: by accurately binding and encapsulating dispatch instructions with specific vehicle identities, the accurate delivery and parsing of instructions are ensured. The vehicle terminal receiving the instructions will automatically configure relevant operation parameters, so that new tasks and route planning take effect immediately, and the execution status will be fed back to the cloud in real time, thus forming a complete management closed loop from instruction generation, issuance, execution to status feedback; the reliability and execution efficiency of dispatch instructions are improved, precise coordination and unified management of multi-vehicle operations are realized, instruction mismatch or response delay is effectively avoided, and the safe and orderly execution of the overall operation plan is guaranteed.

[0098] Example 10: Based on Example 9, this example provides a cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles. The collaborative management unit includes: The tracking subunit is used to periodically obtain the operating status parameters of each vehicle after the task and route of each vehicle have been effectively managed, and to re-analyze the operating status parameters to obtain global analysis indicators of the operating efficiency and safety status of the vehicle under the current status. The real-time optimization management subunit is used to compare global analysis indicators with preset thresholds, and to replan the tasks and routes of each vehicle when a global analysis indicator fails to meet the preset threshold.

[0099] In this embodiment, the operating status parameters refer to the data that are periodically obtained from the intelligent terminal of the parking lot during the task execution process, reflecting the real-time status of the vehicle, such as actual location, speed, task completion progress, and current alarm status.

[0100] In this embodiment, the preset threshold refers to the minimum or maximum standard value set for various global analysis indicators (such as the lower limit of overall operation efficiency, the highest acceptable risk level of the system, etc.) to determine whether the overall operating status of the system meets the standard.

[0101] The beneficial effects of the above technical solution are: it can flexibly respond to various unexpected changes at the work site, continuously ensure that the work process is carried out in an optimal or safe and controllable state, and effectively avoid the decline in efficiency or the accumulation of risks caused by the disconnect between the initial plan and the actual situation.

[0102] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles, characterized in that, include: The data analysis and alarm module is used to perform real-time analysis of driver behavior and the surrounding environment of the vehicle based on the AI ​​model deployed on the intelligent terminal of the parking lot, and to issue an immediate alarm when there is abnormal driving behavior or safety risk. The global analysis index determination module is used to upload the time-series data of the behavior recognition results and environmental risk information obtained from the analysis to the cloud platform, and aggregate and mine the time-series data of each vehicle in the spatial area based on the cloud platform to obtain global analysis indexes of the vehicle's operating efficiency and safety status. The scheduling and management module is used to dynamically generate task scheduling instructions and route planning instructions for each vehicle based on global analysis indicators, and to send the task scheduling instructions and route planning instructions to the intelligent terminal of the vehicle for collaborative management of operation behavior. The global analysis indicator determination module includes: Data transmission unit, used for: Obtain the vehicle identification number of each vehicle and extract the timestamps corresponding to the behavioral recognition results of the drivers on each vehicle and the environmental risk information; The vehicle identification, behavior recognition results, environmental risk information, and corresponding timestamps for each vehicle are packaged together and uploaded to the cloud platform via a wireless communication network. The first data analysis unit is used for: Based on the standardized rules preset on the cloud platform, the received behavior recognition results and environmental risk information are formatted and normalized. At the same time, a vehicle time series database is built for each vehicle on the cloud platform, and the formatted and normalized behavior recognition results and environmental risk information are stored in the corresponding vehicle time series database based on the timestamp and vehicle identifier. Based on the aggregation computing engine, the vehicle time series database is periodically scanned, and statistical calculations are performed on the same data of all vehicles in the specified time window and specified spatial area based on the periodic scan results to obtain the primary aggregation index of each vehicle. The primary aggregation index includes the total frequency of fatigue driving events, the incidence rate of high-risk behaviors, and the density of regional risk points. The global indicator determination unit is used for: Based on a deep learning model, primary aggregation indicators are integrated with historical operation pattern database and static environment map information to determine the spatiotemporal correlation between safety events and predict the evolution trend of systemic risks. Extract the metadata of the operation tasks of each vehicle in the yard, and perform joint analysis of the spatiotemporal correlation, risk evolution trend and operation task metadata based on a multi-objective optimization algorithm to obtain global analysis indicators of the operation efficiency and safety status of the vehicles in the yard.

2. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 1, characterized in that, The data analysis and alarm module includes: Image acquisition unit, used for: The vehicle uses a camera device to collect real-time images of the driver's face and posture, as well as images of the surrounding environment. Keyframe extraction and face region localization are performed on the acquired facial and pose image sequences, and the facial and pose image sequences and environmental image sequences are standardized and preprocessed based on the data input conditions of the AI ​​model. The second data analysis unit is used for: The preprocessed keyframes, face region localization results, and data to be analyzed are input into the AI ​​model deployed on the intelligent terminal of the parking lot. Based on the behavior recognition layer in the AI ​​model, the keyframes and face region localization results are identified and analyzed to determine the driver's behavior recognition results. Meanwhile, the environmental perception layer in the AI ​​model is used to analyze the preprocessed environmental image sequence to obtain environmental risk information around the vehicle. The alarm unit is used for alarm management based on behavior recognition results and environmental risk information.

3. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 2, characterized in that, Alarm unit, including: The alarm judgment subunit is used for: Based on the management terminal, the alarm thresholds corresponding to driving behavior and environmental safety risks are obtained respectively, and the obtained behavior recognition results and environmental risk information are compared with the corresponding alarm thresholds respectively. When the identification result or environmental risk information is greater than or equal to the corresponding alarm threshold, the alarm condition is determined to be met. The alarm execution subunit is used for: Based on the judgment result, a local alarm command is generated, and based on the local alarm command, the audible and visual alarm device on the parking vehicle is activated to provide an immediate warning.

4. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 2, characterized in that, The second data analysis unit includes: Behavior determination subunit, used for: The standardized preprocessed facial and pose image frames are input into the feature extraction sublayer in the behavior recognition layer, and the spatial features of the facial and pose image frames are extracted layer by layer based on the convolutional neural network. Based on the key point localization layer in the behavior recognition layer, the coordinates of the driver's facial key points and the Euler angles of the head posture are determined according to the extracted spatial features. Based on the facial and posture image sequence, the coordinates of the facial key points and the Euler angles of the head posture of each image frame are summarized. Furthermore, based on the temporal convolutional network, the summarization results are analyzed for temporal features to determine the dynamic changes of the driver's blinking frequency, eye-closing duration and head offset angle. Based on a preset classification algorithm, the continuous temporal features obtained are mapped into discrete driver state labels according to the dynamic change pattern, and the corresponding behavior recognition results are obtained. The labels include normal, fatigued, looking around, and using a handheld device. The environmental risk determination subunit is used for: The standardized preprocessed environmental image sequence is input into the environment perception layer, and based on depthwise separable convolution operation, an environmental feature map containing multiple scales is obtained; Based on the environmental feature map, pedestrians, fixed obstacles, and the boundaries of drivable and dangerous areas in the image are identified and bounded. Based on the pixel coordinates of the bounding target detection box, known camera intrinsic parameters, and preset prior knowledge of actual size, the relative distance and azimuth angle between pedestrians or obstacles and the field vehicle are evaluated. The relative distance and azimuth are analyzed based on a preset safety threshold function for distance and speed to obtain the instantaneous risk level of each detected target or area; The environmental risk information around the vehicle is obtained by summarizing the target type, relative distance and azimuth angle between the detected target and the vehicle.

5. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 1, characterized in that, The global indicator determination unit includes: The indicator analysis subunit is used for: Extract the specified time window and specified spatial area corresponding to the primary aggregation index, and retrieve the historical operation load and historical event sequence that match the specified time window and specified spatial area from the historical operation mode library. At the same time, extract the topology, key node location and functional area attributes of each spatial area from the static environment map information to obtain multi-source heterogeneous data. Multi-source heterogeneous data is converted into numerical vectors of a unified dimension, and the numerical vectors are concatenated according to preset rules to obtain a fused feature vector. Based on the deep learning model, the fused feature vectors are converted into a relationship graph of regional nodes and edges according to the topology of each spatial region. Based on the relationship graph of regional nodes and edges, the spatial region dependencies are determined. At the same time, based on the spatial region dependencies, the temporal dependencies between historical event sequences and the current status of the field vehicle are determined. Based on spatial region dependency and temporal dependency, the association weights between different security event characteristics are determined, and the association patterns between security events are determined based on the association weights, resulting in a spatiotemporal association rule table for security events. Based on a deep learning model, iterative analysis is performed on the fused feature vector and the spatiotemporal correlation rule table of security events to obtain the risk evolution probability distribution in each spatial region, and the predicted systemic risk evolution trend is obtained based on the risk evolution probability distribution.

6. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 1, characterized in that, The global analysis indicator determination module includes: The result acquisition unit is used to acquire global analysis indicators of the operating efficiency and safety status of the field vehicles, and extract the time nodes corresponding to the global analysis indicators. The results recording unit is used to construct the indicator recording table and to sequentially enter different time nodes and their corresponding global analysis indicators into the indicator recording table for storage based on the development order of time nodes.

7. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 1, characterized in that, The scheduling management module includes: The optimization target determination unit is used to obtain the obtained global analysis indicators, and at the same time, obtain the operation status information of each vehicle and the queue of tasks to be executed, and determine the optimization target when scheduling tasks based on the global analysis indicators, the operation status information of each vehicle and the queue of tasks to be executed. The task scheduling instruction generation unit is used to construct a multi-objective decision model based on the optimization objective, and to analyze the global analysis indicators, the operation status information of each vehicle and the queue of tasks to be executed based on the objective decision model to obtain a macro scheduling scheme, and to generate task scheduling instructions based on the macro scheduling scheme. The path planning instruction generation unit is used for: Obtain the static environment map corresponding to the scene's working area, and use the macro scheduling scheme as a constraint to generate a collision-free path from the current location to the target point for each vehicle in the field. Assess the estimated travel time and overall risk value of each route, and obtain the optimal route for each vehicle based on the assessment results; Extract the path point sequence, speed suggestions, and risk warnings corresponding to the optimal path, and generate path planning instructions for each vehicle based on the path point sequence, speed suggestions, and risk warnings.

8. The intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles based on cloud-edge collaborative AI as described in claim 1, characterized in that, The scheduling management module includes: The instruction issuing unit is used to encapsulate the task scheduling instructions and route planning instructions of the same vehicle based on the identity tags of each vehicle, and to send the encapsulation results to the corresponding intelligent terminal of the vehicle based on the communication interface of the cloud platform. The collaborative management unit is used to configure the working parameters of each vehicle based on the issued results, and to manage the tasks and routes of each vehicle based on the configuration results. The management results are then fed back to the cloud platform to complete the collaborative management of work behavior.

9. A cloud-edge collaborative AI-based intelligent monitoring and adaptive scheduling management platform for safe operation of parking vehicles according to claim 8, characterized in that, The collaborative management unit includes: The tracking subunit is used to periodically obtain the operating status parameters of each vehicle after the task and route of each vehicle have been effectively managed, and to re-analyze the operating status parameters to obtain global analysis indicators of the operating efficiency and safety status of the vehicle under the current status. The real-time optimization management subunit is used to compare global analysis indicators with preset thresholds, and to replan the tasks and routes of each vehicle when a global analysis indicator fails to meet the preset threshold.

Images