A Real-Time Scheduling Method for Lane Control Robots Based on Multi-Dimensional Sensor Data

Abstract

This invention relates to the field of real-time resource scheduling technology, specifically a real-time scheduling method for lane control robots based on multi-dimensional sensor data. The method includes: dividing management areas for arrival and drop-off robots; acquiring multimodal sensor data from the management areas, extracting features, and fusing them using a self-attention mechanism to obtain a fused feature vector; obtaining a drop-off scenario complexity score and a departure intention score through a drop-off scenario evaluation model; simultaneously calculating the real-time resource occupancy index for each lane; calculating differentiated allowed dwell time based on the drop-off scenario complexity score and the overall resource occupancy status of the drop-off platform; and providing early warnings and management based on the drop-off scenario complexity score, departure intention score, real-time resource occupancy index, differentiated allowed dwell time, and empty space detection results. This invention improves flexible decision-making capabilities and resource utilization, and is suitable for real-time resource allocation and decision support in complex traffic scenarios.

Description

Technical Field

[0001] This invention relates to the field of real-time resource scheduling technology, specifically to a real-time scheduling method for lane control robots based on multi-dimensional sensor data. Background Technology

[0002] Passenger service scenarios in large transportation hubs require highly accurate dynamic matching of service resources. With technological advancements, intelligent systems have been widely applied in this field. Especially in real-time information updates and passenger demand forecasting, advanced computing algorithms and big data analytics provide more precise and efficient solutions for resource allocation. Intelligent scheduling systems can not only monitor the availability of various resources in real time but also adjust service modes according to passenger needs.

[0003] In this process, predictive models based on artificial intelligence and machine learning, combined with real-time data from various facilities within the transportation hub, can effectively avoid resource waste while improving service quality. This precise resource matching system can automatically analyze fluctuations in passenger flow, predict peak-hour passenger density, and adjust service frequency and staffing in a timely manner to ensure that various services can efficiently cope with different load demands.

[0004] Furthermore, with the widespread adoption of IoT technology, more and more facilities and equipment are becoming interconnected, providing greater data support and technical assurance for the dynamic matching of service resources in large transportation hubs. Through precise resource management and efficient scheduling systems, passenger services at transportation hubs will achieve a more intelligent and personalized experience, fully meeting passenger travel needs, improving operational efficiency, and reducing operating costs.

[0005] However, existing technologies still fall short in their ability to respond in real time and quantify differentiated needs in complex dynamic scenarios, resulting in limited efficiency in real-time resource scheduling. To address this, a real-time scheduling method for lane control robots based on multi-dimensional sensor data is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time scheduling method for lane control robots based on multi-dimensional sensor data. This method involves: dividing management areas into arrival and drop-off zones for robots; acquiring multimodal sensor data from these zones, extracting features, and fusing them using a self-attention mechanism to obtain a fused feature vector; obtaining drop-off scenario complexity and departure intention scores through a drop-off scenario evaluation model; simultaneously calculating the real-time resource occupancy index for each lane; calculating differentiated allowed dwell time based on the drop-off scenario complexity score and the overall resource occupancy status of the drop-off platform; and providing early warnings and management based on the drop-off scenario complexity score, departure intention score, real-time resource occupancy index, differentiated allowed dwell time, and available lane detection results. This invention improves flexible decision-making capabilities and resource utilization, and is suitable for real-time resource allocation and decision support in complex traffic scenarios.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A real-time scheduling method for lane control robots based on multi-dimensional sensor data includes:

[0009] Deploy arrival and drop-off robots along the passenger drop-off area, and assign a management area to each robot;

[0010] Acquire multimodal sensor data of the drop-off section robot management area, including video data, thermal imaging data, lidar data and sound data, extract features from the multimodal sensor data, and generate a fused feature vector using a self-attention mechanism;

[0011] The fused feature vector is input into the passenger disembarkation scenario assessment model, which outputs a passenger disembarkation scenario complexity score and a departure intention score.

[0012] Based on the traffic monitoring video of the drop-off platform, calculate the real-time resource occupancy index for each lane;

[0013] The differentiated allowed stay time is calculated based on the passenger drop-off scenario complexity score and the overall resource occupancy status of the passenger drop-off platform;

[0014] Based on the passenger drop-off scenario complexity score, the departure intention score, the real-time resource occupancy index, and the differentiated allowed dwell time, the drop-off section robot provides early warnings and lane allocation for vehicles within the management area.

[0015] The fused feature vector is input into the vacancy detection model to obtain the vacancy coordinates and confidence level. The vacancy information is then fed back to the arriving vehicle by the arrival segment robot.

[0016] Preferably, the process of generating the fused feature vector includes: preprocessing the multimodal sensing data, extracting preliminary features from the preprocessed multimodal sensing data using a feature extraction algorithm corresponding to the data type; performing feature optimization and dimensionality reduction on the preliminary features using an autoencoder to obtain a multimodal feature vector; and fusing the multimodal feature vectors using a self-attention mechanism to obtain the fused feature vector.

[0017] Preferably, the vacancy information includes: when the confidence level is less than a threshold, it indicates that there is no vacancy at the vacancy coordinates; when the confidence level is greater than or equal to a first threshold, it indicates that there is a vacancy at the vacancy coordinates; the vacancy coordinates are the management area coordinates of each of the drop-off section robots.

[0018] Preferably, the passenger disembarkation scenario evaluation model includes an input layer, a vehicle detection layer, a scenario complexity analysis layer, and a departure intention analysis layer; the input layer is used to input a fusion feature vector sequence within a continuous time window; the vehicle detection layer is used to detect target vehicles and vehicle categories within the management area based on the fusion feature vector sequence; the scenario complexity analysis layer includes a scenario detection module and a complexity score output module; the scenario detection module is used to detect the number of passengers, passenger categories, luggage quantity, luggage area, and passenger movement fluency within the target vehicle area; the complexity score output module is used to calculate the passenger disembarkation scenario complexity score of the target vehicle based on the output result of the scenario detection module; the departure intention analysis layer includes an intention detection module and an intention score output module; the intention detection module is used to detect the passenger disembarkation completion rate, door closing status, driver positioning status, and vehicle orientation change of the target vehicle; the intention score output module is used to calculate the departure intention score of the target vehicle based on the output result of the intention detection module; the calculation formulas for the passenger disembarkation scenario complexity score and the departure intention score are:

[0019] ;

[0020] in, This represents the complexity score of the passenger disembarkation scenario. Indicates the number of passengers. Indicates the number of luggage items. Indicates the first The area of ​​each piece of luggage. Indicates the first Passenger category identifier for each passenger. Indicates the first Category weights for each passenger. Indicates the smoothness of passenger movements. , , and Indicates the weighting coefficient. Represents a random disturbance term; This indicates the score for the stated intention to leave. This indicates the passenger disembarkation completion rate. This indicates that the car door is closed. This indicates that the driver is in position. This represents the activation function. Indicates a change in vehicle orientation. The temporal fusion features represent the fused feature vector sequence. , , , and This represents the weighting coefficient.

[0021] Preferably, calculating the real-time resource occupancy index for each lane includes: preprocessing the traffic monitoring video, including perspective correction, illumination equalization, and noise filtering; dividing the video frame into detection areas for each lane, using a target detection algorithm to identify vehicles in each lane detection area, and combining a multi-target tracking algorithm to calculate the vehicle density, average speed, and average travel time of each lane detection area in real time; and calculating the real-time resource occupancy index for each lane based on the vehicle density, the average speed, and the average travel time.

[0022] Preferably, calculating the differentiated allowed dwell time includes: obtaining the target vehicle category and setting a basic dwell time based on the target vehicle category; calculating a scenario complexity coefficient based on historical data; calculating a resource occupancy adjustment factor based on the overall resource occupancy status of the drop-off platform; and calculating the differentiated allowed dwell time based on the basic dwell time, the scenario complexity coefficient, the drop-off scenario complexity score, and the resource occupancy adjustment factor, using the following formula: ;in, This indicates the allowable dwell time for the differentiation. Indicates the target vehicle category The basic length of stay, This represents the context complexity coefficient. This represents the complexity score of the passenger disembarkation scenario. This indicates the resource usage adjustment factor.

[0023] Preferably, the warning and lane allocation for vehicles within the management area includes: when the departure intention score of the target vehicle is greater than or equal to the second threshold, prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index; when the departure intention score of the target vehicle is less than the second threshold and the vehicle's dwell time reaches the differentiated allowable dwell time, issuing a warning and prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] 1. This invention, through the collection and fusion of multi-dimensional data, can comprehensively perceive the status of vehicles and passengers, obtain complementary information from heterogeneous data, and reduce the impact of environmental interference. The passenger drop-off scenario assessment model adopts a refined multi-layer architecture, from vehicle detection, scenario detection to intent detection, achieving accurate identification of multiple micro-factors, and quantifying these complex factors into a passenger drop-off scenario complexity score and a departure intent score. Based on deep feature-driven scenario understanding capabilities, it can consider the actual needs of each vehicle, solving the problem of not being able to accurately perceive and quantify the differentiated needs of different vehicles in the high-speed rail station drop-off area scenario, providing a data foundation for differentiated intelligent scheduling, improving the adaptability of lane control on the drop-off area platform, and thus improving scheduling efficiency.

[0026] 2. This invention proposes a dynamic calculation mechanism for differentiated allowed dwell time. First, different basic dwell times are set according to vehicle type, taking into account the different basic needs of different vehicles. Then, a situational complexity coefficient is used to convert the passenger drop-off situation complexity score, ensuring sufficient drop-off time in complex situations. Simultaneously, a resource occupancy adjustment factor is introduced to appropriately shorten the dwell time of individual vehicles when platform congestion is high, balancing individual needs and overall efficiency. This refined time management method avoids the poor experience caused by a simple one-size-fits-all approach, while preventing regional congestion caused by unlimited dwell time, achieving a balance between fairness and efficiency in resource allocation. It solves the problem that fixed-duration management of passenger drop-off areas in high-speed rail stations cannot adapt to different situational needs, improving fairness and adaptability.

[0027] 3. This invention designs a lane collaborative scheduling strategy based on real-time overall resource occupancy status. A mathematical model is constructed using three key indicators: vehicle density, average speed, and average travel time, to accurately quantify the resource occupancy level of each lane. Simultaneously, a vehicle scheduling mechanism is set up based on departure intention scoring and resource occupancy index. When a vehicle shows a departure intention or its dwell time reaches its limit, the system intelligently schedules the vehicle to select the least congested exit lane, avoiding secondary congestion. In particular, the information linkage between robots at the entrance and drop-off sections achieves optimized allocation of arriving vehicles, improving overall operational efficiency. This solves the problems of localized congestion and uneven resource allocation in high-speed rail station drop-off areas, improving overall traffic flow management efficiency. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a real-time scheduling method for lane control robots based on multi-dimensional sensor data according to the present invention.

[0029] Figure 2 This is a schematic diagram of the passenger disembarkation scenario assessment model of the present invention;

[0030] Figure 3 This is a schematic diagram of the data flow for calculating the differentiated allowable dwell time according to the present invention. Detailed Implementation

[0031] 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.

[0032] Please see Figures 1 to 3 This invention provides a real-time scheduling method for lane control robots based on multi-dimensional sensor data, the technical solution of which is as follows:

[0033] Example 1:

[0034] A real-time scheduling method for lane control robots based on multi-dimensional sensor data, see [link / reference]. Figure 1 ,include:

[0035] Deploy arrival and drop-off robots along the passenger drop-off area, and assign a management area to each robot;

[0036] Acquire multimodal sensor data of the drop-off section robot management area, including video data, thermal imaging data, lidar data and sound data, extract features from the multimodal sensor data, and generate a fused feature vector using a self-attention mechanism;

[0037] The fused feature vector is input into the passenger disembarkation scenario assessment model, which outputs a passenger disembarkation scenario complexity score and a departure intention score.

[0038] Based on the traffic monitoring video of the drop-off platform, calculate the real-time resource occupancy index for each lane;

[0039] The differentiated allowed stay time is calculated based on the passenger drop-off scenario complexity score and the overall resource occupancy status of the passenger drop-off platform;

[0040] Based on the passenger drop-off scenario complexity score, the departure intention score, the real-time resource occupancy index, and the differentiated allowed dwell time, the drop-off section robot provides early warnings and lane allocation for vehicles within the management area.

[0041] The fused feature vector is input into the vacancy detection model to obtain the vacancy coordinates and confidence level. The vacancy information is then fed back to the arriving vehicle by the arrival segment robot.

[0042] This invention achieves a refined understanding of passenger drop-off scenarios through multimodal sensor data fusion and deep learning technology; it addresses the inadequacy of fixed-duration control by dynamically calculating differentiated allowable dwell times; and it optimizes overall resource allocation and traffic flow organization through real-time overall resource occupancy status assessment and collaborative scheduling strategies. It considers both the needs of special passengers and maintains overall system efficiency, significantly improving passenger experience while enhancing the efficiency and safety of drop-off areas. This also elevates the intelligence level of large transportation hubs such as high-speed rail stations.

[0043] Furthermore, the process of generating the fused feature vector includes: preprocessing the multimodal sensing data, extracting preliminary features from the preprocessed multimodal sensing data using a feature extraction algorithm corresponding to the data type; performing feature optimization and dimensionality reduction on the preliminary features using an autoencoder to obtain a multimodal feature vector; and fusing the multimodal feature vectors using a self-attention mechanism to obtain the fused feature vector.

[0044] This embodiment achieves real-time data processing through edge computing devices. Specific preprocessing content and feature extraction methods are shown in Table 1. Furthermore, it includes spatiotemporal alignment and normalization: using the timestamp of the video data as a reference, other sensor data are aligned through linear interpolation; spatial alignment maps the LiDAR point cloud to the pixel coordinates of the video frame by calibrating the extrinsic parameter matrix of the multi-sensor system.

[0045] Table 1. Multidimensional data preprocessing and feature extraction

[0046]

[0047] By employing multimodal sensing data and using autoencoders and self-attention mechanisms for feature extraction and fusion, the limitations of single sensors are overcome, the system's ability to perceive complex environments and resist interference is improved, and more comprehensive and accurate basic data is provided for subsequent situation assessment and decision-making.

[0048] Furthermore, the vacancy information includes: when the confidence level is less than a threshold, it indicates that there is no vacancy at the vacancy coordinates; when the confidence level is greater than or equal to a first threshold, it indicates that there is a vacancy at the vacancy coordinates; the vacancy coordinates are the management area coordinates of each of the drop-off section robots.

[0049] In this embodiment, the empty space recognition model uses a YOLOv5 deep learning network, fusing feature vectors as input, with a first threshold set to 0.75. The empty space recognition model also includes a temporal consistency check mechanism, determining the final empty space detection result through voting results from five consecutive frames of data. The drop-off robot transmits the empty space recognition result to the arrival robot via a low-latency wireless network. Based on the received real-time empty space detection result, the arrival robot provides guidance information to arriving vehicles through an LED display screen, improving the resource utilization rate of the entire drop-off area and preventing vehicles from congesting at the entrance of the drop-off area, thus avoiding disruption to normal traffic.

[0050] Further, see Figure 2 The passenger disembarkation scenario assessment model includes an input layer, a vehicle detection layer, a scenario complexity analysis layer, and a departure intention analysis layer. The input layer is used to input a fusion feature vector sequence within a continuous time window. The vehicle detection layer is used to detect target vehicles and vehicle categories within the management area based on the fusion feature vector sequence. The scenario complexity analysis layer includes a scenario detection module and a complexity score output module. The scenario detection module detects the number of passengers, passenger categories, luggage quantity, luggage area, and passenger movement fluency within the target vehicle area. The complexity score output module calculates the passenger disembarkation scenario complexity score of the target vehicle based on the output of the scenario detection module. The departure intention analysis layer includes an intention detection module and an intention score output module. The intention detection module detects the passenger disembarkation completion rate, door closing status, driver's position status, and vehicle orientation change of the target vehicle. The intention score output module calculates the departure intention score of the target vehicle based on the output of the intention detection module. The calculation formulas for the passenger disembarkation scenario complexity score and the departure intention score are as follows:

[0051] ;

[0052] in, This represents the complexity score of the passenger disembarkation scenario. Indicates the number of passengers. Indicates the number of luggage items. Indicates the first The area of ​​each piece of luggage. Indicates the first Passenger category identifier for each passenger. Indicates the first Category weights for each passenger. Indicates the smoothness of passenger movements. , , and Indicates the weighting coefficient. Represents a random disturbance term; This indicates the score for the stated intention to leave. This indicates the passenger disembarkation completion rate. This indicates that the car door is closed. This indicates that the driver is in position. This represents the activation function. Indicates a change in vehicle orientation. The temporal fusion features represent the fused feature vector sequence. , , , and This represents the weighting coefficient.

[0053] The multi-level passenger disembarkation scenario assessment model enables a refined understanding of the passenger disembarkation process. Through a structured mathematical model, complex scenarios are quantified into scoring indicators. The model can identify complex factors such as passengers with special needs and large luggage, and predict vehicle behavior through departure intention scoring. This provides a scientific basis for differentiated management and control, effectively balances passenger experience and system efficiency, and solves the rigidity problem of traditional fixed-duration management.

[0054] Further, calculating the real-time resource occupancy index for each lane includes: preprocessing the traffic monitoring video, including perspective correction, illumination equalization, and noise filtering; dividing the video frame into detection areas for each lane, using a target detection algorithm to identify vehicles in each lane detection area, and combining a multi-target tracking algorithm to calculate the vehicle density, average speed, and average travel time of each lane detection area in real time; and calculating the real-time resource occupancy index for each lane based on the vehicle density, the average speed, and the average travel time.

[0055] The vehicle density is the ratio of the number of vehicles in the lane to the length of the drop-off area lane; the average speed is the average speed of all vehicles in the lane; the average travel time is the average difference between the timestamps of all vehicles entering and leaving the lane detection area. The formula for calculating the real-time resource occupancy index is:

[0056] ;

[0057] in, This represents the real-time resource usage index. and These represent vehicle density and maximum vehicle density, respectively. and These represent average vehicle speed and free-flow speed, respectively. Free-flow speed represents the average speed of a vehicle under ideal conditions where there is no traffic congestion and it is not affected by other vehicles. and These represent the average travel time and the reference travel time, respectively. , and This represents the weighting coefficient, and the optimal value of the weighting coefficient can be determined through a multi-objective optimization algorithm.

[0058] This embodiment uses the YOLOv7 object detection algorithm to identify vehicles in the video and combines it with the DeepSORT multi-object tracking algorithm to achieve continuous vehicle tracking. After preprocessing the traffic monitoring video, the 2D image is converted into a bird's-eye view coordinate system by calibrating the mapping relationship between the actual coordinates of the lane lines and the pixel coordinates of the video image. Each lane detection area is defined as a rectangular area between the lane lines in the bird's-eye view.

[0059] The real-time resource occupancy index calculation method based on vehicle density, average vehicle speed, and average travel time enables the scientific quantification of lane congestion. It not only considers static vehicle distribution but also focuses on dynamic flow characteristics, enabling a more comprehensive assessment of traffic conditions. This provides a reliable basis for vehicle scheduling decisions and reduces congestion chain reactions.

[0060] Further, see Figure 3 The calculation of the differentiated allowed stay time includes: obtaining the target vehicle category and setting a basic stay time based on the target vehicle category; calculating the scenario complexity coefficient based on historical data; calculating the resource occupancy adjustment factor based on the overall resource occupancy status of the drop-off platform; and calculating the differentiated allowed stay time based on the basic stay time, the scenario complexity coefficient, the drop-off scenario complexity score, and the resource occupancy adjustment factor. The calculation formula is as follows: ;in, This indicates the allowable dwell time for the differentiation. Indicates the target vehicle category The basic length of stay, This represents the context complexity coefficient. This represents the complexity score of the passenger disembarkation scenario. This represents the resource usage adjustment factor. The differentiated allowed stay time needs to be subject to upper and lower limit constraints. A minimum allowed stay time and a maximum allowed stay time are set. If the differentiated allowed stay time is less than the minimum allowed stay time, the minimum allowed stay time is used; if the differentiated allowed stay time is greater than the maximum allowed stay time, the maximum allowed stay time is used.

[0061] The situation complexity coefficient The calculation formula is:

[0062] ;

[0063] in, This indicates the number of samples in the historical data. Indicates the first The actual dwell time of each sample Indicates the first Vehicle categories in the sample The basic length of stay, This represents the complexity score of the passenger disembarkation scenario. express The mapping coefficients in this embodiment A value of 5 indicates that each 1 point of situational complexity score corresponds to an additional 5 seconds of time.

[0064] The resource usage adjustment factor The calculation formula is:

[0065] ;

[0066] in, This indicates the overall resource occupancy status of the current drop-off platform, calculated as the ratio of the total number of vehicles in the drop-off area to the maximum capacity.

[0067] The dynamic calculation mechanism for differentiated permitted stay time breaks away from the traditional fixed-duration management model, introducing an adaptive time allocation strategy based on vehicle type, situational complexity, and regional congestion. By learning from historical data, the situational complexity coefficient possesses the ability to accumulate experience, enabling more accurate estimation of reasonable stay time under different situations. This time management approach, balancing individual needs with overall efficiency, improves system fairness and passenger satisfaction.

[0068] Furthermore, the warning and lane allocation for vehicles within the management area includes: when the departure intention score of a target vehicle is greater than or equal to a second threshold, prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index; when the departure intention score of a target vehicle is less than the second threshold and the vehicle's dwell time reaches the differentiated allowable dwell time, issuing a warning and prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index.

[0069] A vehicle dispatching mechanism based on departure intention scoring and resource occupancy index enables intelligent control of the vehicle departure process. It proactively identifies vehicles preparing to leave and prompts them to choose the optimal exit route, reducing secondary congestion. For vehicles exceeding their designated time limit, a combination of early warning and dispatching effectively improves vehicle turnover efficiency.

[0070] Example 2:

[0071] This embodiment applies to the drop-off platform of the south square of a high-speed rail station in a certain city. The platform has four parallel one-way lanes, each 180 meters long and 3.5 meters wide. The lanes closest to the high-speed rail station are designated as the drop-off area, which frequently experiences congestion during morning and evening rush hours, reducing passenger travel experience and causing traffic pressure in front of the station. This embodiment implements a real-time scheduling method for lane control robots based on multi-dimensional sensor data by evenly deploying 16 lane control robots in the drop-off area, including:

[0072] Deploy arrival and drop-off robots along the passenger drop-off area, and assign a management area to each robot;

[0073] Acquire multimodal sensor data of the drop-off section robot management area, including video data, thermal imaging data, lidar data and sound data, extract features from the multimodal sensor data, and generate a fused feature vector using a self-attention mechanism;

[0074] The fused feature vector is input into the passenger disembarkation scenario assessment model, which outputs a passenger disembarkation scenario complexity score and a departure intention score.

[0075] Based on the traffic monitoring video of the drop-off platform, calculate the real-time resource occupancy index for each lane;

[0076] The differentiated allowed stay time is calculated based on the passenger drop-off scenario complexity score and the overall resource occupancy status of the passenger drop-off platform;

[0077] Based on the passenger drop-off scenario complexity score, the departure intention score, the real-time resource occupancy index, and the differentiated allowed dwell time, the drop-off section robot provides early warnings and lane allocation for vehicles within the management area.

[0078] The fused feature vector is input into the vacancy detection model to obtain the vacancy coordinates and confidence level. The vacancy information is then fed back to the arriving vehicle by the arrival segment robot.

[0079] Furthermore, the process of generating the fused feature vector includes: preprocessing the multimodal sensing data, extracting preliminary features from the preprocessed multimodal sensing data using a feature extraction algorithm corresponding to the data type; performing feature optimization and dimensionality reduction on the preliminary features using an autoencoder to obtain a multimodal feature vector; and fusing the multimodal feature vectors using a self-attention mechanism to obtain the fused feature vector.

[0080] Furthermore, the vacancy information includes: when the confidence level is less than a threshold, it indicates that there is no vacancy at the vacancy coordinates; when the confidence level is greater than or equal to a first threshold, it indicates that there is a vacancy at the vacancy coordinates; the vacancy coordinates are the management area coordinates of each of the drop-off section robots.

[0081] Furthermore, the passenger disembarkation scenario assessment model includes an input layer, a vehicle detection layer, a scenario complexity analysis layer, and a departure intention analysis layer. The input layer is used to input a fusion feature vector sequence within a continuous time window. The vehicle detection layer is used to detect target vehicles and vehicle categories within the management area based on the fusion feature vector sequence. The scenario complexity analysis layer includes a scenario detection module and a complexity score output module. The scenario detection module is used to detect the number of passengers, passenger categories, luggage quantity, luggage area, and passenger movement fluency within the target vehicle area. The complexity score output module is used to calculate the passenger disembarkation scenario complexity score of the target vehicle based on the output result of the scenario detection module. The departure intention analysis layer includes an intention detection module and an intention score output module. The intention detection module is used to detect the passenger disembarkation completion rate, door closing status, driver positioning status, and vehicle orientation change of the target vehicle. The intention score output module is used to calculate the departure intention score of the target vehicle based on the output result of the intention detection module.

[0082] Further, calculating the real-time resource occupancy index for each lane includes: preprocessing the traffic monitoring video, including perspective correction, illumination equalization, and noise filtering; dividing the video frame into detection areas for each lane, using a target detection algorithm to identify vehicles in each lane detection area, and combining a multi-target tracking algorithm to calculate the vehicle density, average speed, and average travel time of each lane detection area in real time; and calculating the real-time resource occupancy index for each lane based on the vehicle density, the average speed, and the average travel time.

[0083] Further, calculating the differentiated allowed stay time includes: obtaining the target vehicle category and setting a basic stay time based on the target vehicle category; calculating a context complexity coefficient based on historical data; calculating a resource occupancy adjustment factor based on the overall resource occupancy status of the drop-off platform; and calculating the differentiated allowed stay time based on the basic stay time, the context complexity coefficient, and the resource occupancy adjustment factor.

[0084] Furthermore, the warning and lane allocation for vehicles within the management area includes: when the departure intention score of a target vehicle is greater than or equal to a second threshold, prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index; when the departure intention score of a target vehicle is less than the second threshold and the vehicle's dwell time reaches the differentiated allowable dwell time, issuing a warning and prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index.

[0085] Table 2 shows examples of performance metrics for the passenger drop-off scenario assessment model on some sub-tasks. Trained and validated on 5000 labeled samples, it achieved high accuracy, where the F1 score is the harmonic mean of precision and recall. The overall model's average inference time is 125ms, meeting real-time processing requirements. Table 3 presents the changes in key metrics before and after applying this invention. By comparing the data after applying this invention with historical data from the same month before application, the results show that applying this invention significantly reduced the frequency of congestion and vehicle dwell time, thereby increasing traffic flow.

[0086] Table 2 Examples of performance indicators for passenger drop-off scenario evaluation models

[0087]

[0088] Table 3 Comparison Results of Key Indicators

[0089]

[0090] 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 equivalents.

Claims

1. A lane management robot real-time scheduling method based on multi-dimensional sensing data, characterized in that, include: Deploy arrival and drop-off robots along the passenger drop-off area, and assign a management area to each robot; Acquire multimodal sensor data of the drop-off section robot management area, including video data, thermal imaging data, lidar data and sound data, extract features from the multimodal sensor data, and generate a fused feature vector using a self-attention mechanism; The fused feature vectors are input into the passenger disembarkation scenario evaluation model, which outputs a passenger disembarkation scenario complexity score and a departure intention score. The passenger disembarkation scenario evaluation model includes an input layer, a vehicle detection layer, a scenario complexity analysis layer, and a departure intention analysis layer. The input layer is used to input a sequence of fused feature vectors within a continuous time window. The vehicle detection layer is used to detect target vehicles and vehicle categories within the management area based on the fused feature vector sequence. The scenario complexity analysis layer includes a scenario detection module and a complexity score output module. The scenario detection module is used to detect the number of passengers, passenger categories, luggage quantity, luggage area, and passenger movement fluency within the target vehicle area. The complexity score output module is used to calculate the passenger disembarkation scenario complexity score of the target vehicle based on the output of the scenario detection module. The departure intention analysis layer includes an intention detection module and an intention score output module. The intention detection module is used to detect the passenger disembarkation completion rate, door closing status, driver's position status, and vehicle orientation change of the target vehicle. The intention score output module is used to calculate the departure intention score of the target vehicle based on the output of the intention detection module. Based on the traffic monitoring video of the drop-off platform, calculate the real-time resource occupancy index for each lane; The differentiated allowed stay time is calculated based on the passenger drop-off scenario complexity score and the overall resource occupancy status of the passenger drop-off platform; Based on the passenger drop-off scenario complexity score, the departure intention score, the real-time resource occupancy index, and the differentiated allowed dwell time, the drop-off section robot provides early warnings and lane allocation for vehicles within the management area. The fused feature vector is input into the vacancy detection model to obtain the vacancy coordinates and confidence level. The vacancy information is then fed back to the arriving vehicle by the arrival segment robot.

2. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, The process of generating the fused feature vector includes: preprocessing the multimodal sensing data, extracting preliminary features from the preprocessed multimodal sensing data using a feature extraction algorithm corresponding to the data type; performing feature optimization and dimensionality reduction on the preliminary features using an autoencoder to obtain a multimodal feature vector; and fusing the multimodal feature vectors using a self-attention mechanism to obtain the fused feature vector.

3. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, The vacancy information includes: when the confidence level is less than a first threshold, it means that there is no vacancy at the vacancy coordinates; when the confidence level is greater than or equal to the first threshold, it means that there is a vacancy at the vacancy coordinates; the vacancy coordinates are the management area coordinates of each of the drop-off section robots.

4. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, The formulas for calculating the passenger disembarkation situation complexity score and the departure intention score are as follows: ； in, This represents the complexity score of the passenger disembarkation scenario. Indicates the number of passengers. Indicates the number of luggage items. Indicates the first The area of ​​each piece of luggage. Indicates the first Passenger category identifier for each passenger. Indicates the first Category weights for each passenger. Indicates the smoothness of passenger movements. and Indicates the weighting coefficient. Represents a random disturbance term; This indicates the score for the stated intention to leave. This indicates the passenger disembarkation completion rate. This indicates that the car door is closed. This indicates that the driver is in position. This represents the activation function. Indicates a change in vehicle orientation. The temporal fusion features represent the fused feature vector sequence. and This represents the weighting coefficient.

5. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, Calculating the real-time resource occupancy index for each lane includes: preprocessing the traffic monitoring video, including perspective correction, illumination equalization, and noise filtering; dividing the video frame into detection areas for each lane, using a target detection algorithm to identify vehicles in each lane detection area, and combining this with a multi-target tracking algorithm to calculate the vehicle density, average speed, and average travel time of each lane detection area in real time; and calculating the real-time resource occupancy index for each lane based on the vehicle density, average speed, and average travel time.

6. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, Calculating the differentiated allowed dwell time includes: obtaining the target vehicle category and setting a base dwell time based on the target vehicle category; calculating the scenario complexity coefficient based on historical data; calculating a resource occupancy adjustment factor based on the overall resource occupancy status of the drop-off platform; and calculating the differentiated allowed dwell time based on the base dwell time, the scenario complexity coefficient, the drop-off scenario complexity score, and the resource occupancy adjustment factor. The calculation formula is as follows: ;in, This indicates the allowable dwell time for the differentiation. Let K represent the base dwell time for target vehicle category C, and K represent the situation complexity coefficient. J represents the complexity score of the passenger disembarkation scenario, and J represents the resource usage adjustment factor.

7. The real-time scheduling method for lane control robots based on multi-dimensional sensor data according to claim 1, characterized in that, The warning and lane allocation for vehicles within the management area include: when the departure intention score of a target vehicle is greater than or equal to a second threshold, prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index; when the departure intention score of a target vehicle is less than the second threshold and the vehicle's dwell time reaches the differentiated allowable dwell time, issuing a warning and prompting the vehicle to leave through the lane with the lowest real-time resource occupancy index.

Images