A method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints

Abstract

This invention discloses a method for assessing and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints, relating to the field of urban traffic management technology. The specific steps of this method are as follows: First, typical intersections are selected, and video footage and spatiotemporal parameters of electric bicycles are collected at multiple time periods. Next, trajectories are reconstructed based on the data, and feature parameters are extracted for quality control. Then, a three-dimensional indicator system is constructed to quantify conflict intensity, calculate scores, classify levels, and identify high-risk combinations. Based on this, differentiated optimization measures are formulated. Finally, the same intersection is retested, and relevant data are compared to complete the closed-loop verification of the optimization effect. This invention, by constructing a three-dimensional indicator system, characterizes the left-turn status of electric bicycles from multiple dimensions, scientifically classifies applicability levels, formulates differentiated optimization measures, and performs closed-loop verification. This avoids traditional problems, provides an operable method, promotes the transformation of traffic management towards data-driven approaches, improves the safety, orderliness, and efficiency of intersection operations, and contributes to the continuous improvement of the traffic environment.

Description

Technical Field

[0001] This invention relates to the field of urban traffic management technology, specifically to a method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints. Background Technology

[0002] In urban transportation systems, electric bicycles have become an important choice for short-distance travel due to their flexibility, economy, and environmental friendliness. Their ownership and usage frequency continue to rise. While alleviating traffic pressure, they also place higher demands on traffic order and safety management. Intersections, as core nodes for traffic flow convergence and transformation, bring together various traffic participants such as electric bicycles, pedestrians, and motor vehicles, resulting in complex and diverse traffic behaviors. Left turns, as a critical operational link for electric bicycles at intersections, involve multiple aspects such as route changes and traffic flow conflict avoidance. Affected by multiple spatiotemporal factors such as time of day, traffic flow, intersection layout, and signal control, they are prone to becoming high-incidence points for traffic conflicts and accidents. With the increasing demand for refined urban traffic management, how to scientifically assess the operational status of electric bicycles turning left at intersections, identify potential risks, and optimize traffic organization has become an important issue for ensuring traffic safety, improving traffic efficiency, and regulating traffic behavior. The research and application of related technologies have urgent practical significance.

[0003] Traditional assessment and optimization technologies for electric bicycles turning left at intersections have many limitations in practical applications. In terms of assessment dimensions, they often focus on a single aspect, failing to comprehensively integrate key elements such as safety, efficiency, and compliance. This results in unsystematic assessments that fail to reflect the overall situation of left-turn operations. Data collection lacks full-time, multi-scenario coverage, insufficiently considering traffic flow changes at different times and differences in intersection types, making it difficult to guarantee the accuracy and reliability of collected data and affecting the accuracy of subsequent analysis. In terms of optimization measures, a uniform approach is often adopted, ignoring the differences in functional characteristics of different intersections, traffic flow distribution at different times, and high-risk scenarios. This leads to weak targeting of optimization measures and limited implementation effects. Furthermore, traditional technologies lack a robust closed-loop verification mechanism, making it impossible to accurately measure the improvement effects on key indicators such as conflict intensity, traffic efficiency, and compliance rate after the optimization measures are implemented. This hinders the formation of a virtuous cycle of continuous optimization and fails to meet the actual needs of refined and scientific urban traffic management. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for assessing and optimizing the suitability of electric bicycles turning left at intersections under spatiotemporal constraints. This method acquires accurate operational data through multi-scenario data collection covering all time periods, combined with target detection, trajectory reconstruction, and rigorous quality control technologies. It constructs a three-dimensional indicator system of safety, efficiency, and compliance to quantify the intensity of spatiotemporal coupling conflicts, classify suitability levels, and identify high-risk spatiotemporal combinations. Based on actual scenario differences, it formulates differentiated optimization measures that coordinate geometric design, signal control, and management strategies. The optimization effectiveness is verified through closed-loop retesting. The solution systematically integrates multiple technologies to achieve accurate assessment, differentiated optimization, and quantifiable results, effectively mitigating left-turn safety hazards, improving traffic efficiency, and standardizing traffic behavior, providing reliable technical support for the refined management of urban intersection traffic.

[0005] To solve the above-mentioned technical problems, this invention provides the following technical solution: a method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints, the specific steps of which are as follows: S1, Multi-time period data acquisition: Screen typical intersections covering different types and functional scenarios, deploy high-definition cameras, and collect video of electric bicycles, pedestrians and motor vehicles during morning peak, evening peak, off-peak and nighttime, as well as spatiotemporal parameters such as signal cycle, intersection geometry and time period information. S2, Trajectory Reconstruction and Quality Control: Based on the collected data, target recognition and association are completed through target detection and multi-target tracking. High-precision trajectory is reconstructed through coordinate transformation and smoothing. Feature parameters such as turning radius, instantaneous speed, path offset, collision time, intrusion time, and minimum spacing are extracted. Data quality control is completed by setting trajectory continuity thresholds and outlier filtering rules. S3, 3D Applicability Modeling: Based on the extracted feature parameters, a 3D indicator system of safety, efficiency and compliance is constructed. The comprehensive conflict intensity is calculated through the spatiotemporal coupling conflict intensity quantification formula. The comprehensive score is obtained by combining the 3D coupling applicability scoring formula, applicability levels are divided and high-risk spatiotemporal condition combinations are identified. S4, Adaptive Solution Development: Based on the applicability level and high-risk combination, and combined with the differences in intersection type, time period characteristics and functional scenarios, develop differentiated optimization measures for the coordination of geometric design, signal control, and management strategies. S5, Closed-loop effect verification: Video retesting is conducted at the same intersection using the same equipment and processing methods to obtain optimized feature parameters and indicators. The overall conflict intensity, overall score, left turn delay, and violation rate are compared to complete the optimization effect verification.

[0006] Furthermore, the selection criteria for typical intersections include main road-to-main road crossroads, main road-to-secondary road T-junctions, secondary road-to-secondary road crossroads, and compact crossroads in commercial areas, with at least one intersection of each type selected. The selected intersections must meet the following requirements: non-motorized vehicle traffic must account for 20% to 30% of the total traffic flow; the proportion of left-turning electric bicycles in the total number of non-motorized vehicles must be no less than 25% on average; and the frequency of left-turning accidents involving electric bicycles must be higher than the average accident frequency of urban intersections in the past year. The installation of fixed-point high-definition cameras must also meet the following requirements: the horizontal viewing angle of the lens must be no less than 90 degrees, the vertical viewing angle must be no less than 60 degrees, the ratio of the distance between the lens and the center of the intersection to the radius of the maximum circumscribed circle of the intersection must be 1:5 to 1:8, the installation location must be free from high-voltage power lines and trees, and the vertical distance from ground obstacles must be no less than 2 meters.

[0007] Furthermore, the target detection uses the YOLOv8 algorithm, and the multi-target tracking uses the BoT-SORT algorithm. The confidence threshold of the YOLOv8 algorithm is set to no less than 0.8, the IOU threshold is set to no less than 0.5, the input image size is set to 640×640 pixels, the batch processing size is set to 8, the number of training iterations is set to 1000, the initial learning rate is set to 0.01, and a cosine annealing learning rate decay strategy is adopted. In the BoT-SORT algorithm, the process noise covariance matrix Q of the Kalman filter is set to diag([1e-2, 1e-2, 1e-2, 1e-2, 1e-4]), and the observation noise covariance matrix R is set to diag([1e-1, 1e-1, 1e-1, 1e-1]). When the target occlusion duration exceeds 3 frames, the trajectory prediction interpolation method is used to supplement the trajectory points.

[0008] Furthermore, the coordinate transformation employs a homography mapping method. Specifically, a square calibration board with a side length of 1 meter is set up at the data acquisition site. The calibration board is covered with a 10×10 alternating black and white grid, with each grid having a side length of 10 centimeters. During setup, it is ensured that the horizontal distance between the center of the calibration board and the center of the intersection does not exceed 5 meters, and that the plane of the calibration board is parallel to the ground. At least four ground control points are selected to establish a mapping relationship between pixel coordinates and actual ground coordinates. The coordinates of the ground control points are measured using a GPS positioning device, with a positioning accuracy not exceeding ±2 centimeters. The smoothing process employs Savitzky-Golay smoothing technology. The window size of this technology is set to 7 frames, and the polynomial order is set to 3rd order. By performing weighted fitting processing on abnormal fluctuation points in the original trajectory data, noise interference caused by shooting shake and slight target offset is eliminated, outputting continuous, smooth, and high-precision trajectory data. The positional deviation of the trajectory data is controlled within 0.05 meters.

[0009] Furthermore, the specific rules for data quality control are as follows: the number of valid frames for a single trajectory is not less than 30 frames, and the time interval between adjacent valid trajectory points does not exceed 0.1 seconds; the 3σ principle is used to filter out abnormal values ​​of feature parameters, the normal range of instantaneous speed is set to 0.5 km / h to 25 km / h, and the normal range of acceleration is set to -3 m / s² to 3 m / s²; when the distance between trajectory points does not exceed 0.1 meters or the speed of 3 consecutive frames is not less than 25 km / h, it is judged as an abnormal point and is removed.

[0010] Furthermore, the specific content of the three-dimensional indicator system of safety, efficiency, and compliance is as follows: Safety dimension indicators include speed fluctuation for speed stability, path deviation for path compliance, conflict density for conflict frequency, collision time for collision urgency, intrusion time for intrusion severity, and minimum spacing for space margin; Efficiency dimension indicators include left-turn delay for time loss, travel time for total journey time, number of stops and starts for smooth driving, green light utilization rate for green light time utilization efficiency, and throughput for traffic capacity; Compliance dimension indicators include guideline deviation rate for path compliance and violation rate for traffic behavior violations.

[0011] Furthermore, the mathematical expression of the spatiotemporal coupling conflict intensity quantification formula is as follows: ,in To assess the overall intensity of the conflict, Indicates the conflict type identifier. 1, 2, and 3 correspond to horizontal conflict, vertical conflict, and intersectional conflict, respectively. For conflict type weights, This is the basic risk value for horizontal conflict. This is the basic risk value for vertical conflict. The basic risk value for convergent conflicts, Weights are based on the time dimension. This is a time correction factor. For spatial dimension weights, This is the spatial correction factor. For dynamic adjustment of flow rate, For real-time left-turning electric bicycle traffic flow, This is the baseline flow rate.

[0012] Furthermore, the mathematical expression of the three-dimensional coupling applicability scoring formula is as follows: ,in For comprehensive scoring, For security dimension weighting, As the weight for the efficiency dimension, As a weighting factor for compliance dimensions, To score for the security dimension, To score on the efficiency dimension, To score in the compliance dimension, For three-dimensional coupling coefficients, To assess the overall intensity of the conflict, The threshold for maximum conflict intensity. To optimize the potential correction coefficient.

[0013] Furthermore, the specific implementation method for classifying applicability levels and identifying high-risk spatiotemporal condition combinations is as follows: Applicability levels are classified using FCM fuzzy clustering technology, with a cluster size of 5. Weighted fuzzy clustering is employed, and the clustering iteration termination condition is that the distance between the cluster centers of two iterations does not exceed 0.001. The maximum number of iterations is set to 100. The classification criteria are determined based on a comprehensive score: a comprehensive score between 0.8 and 1.0 indicates high applicability; between 0.6 and 0.8 indicates relatively high applicability; between 0.4 and 0.6 indicates moderate applicability; and between 0.2 and 0.4 indicates low applicability. Applicability is defined as low applicability, with scores between 0 and 0.2. Apriori technology is used to mine high-risk spatiotemporal condition combinations, setting a minimum support of 0.2, a minimum confidence of 0.7, and a maximum itemset length of 4. The selection criteria are a comprehensive score of low applicability and a comprehensive conflict intensity higher than 3.5. The mined spatiotemporal conditions include intersection type, time period, left-turning electric bicycle flow range, left-turning motor vehicle flow range, signal cycle duration, and approach lane width. Redundant rules are removed using a pruning strategy, retaining the top 20% of rules by confidence level to form a set of high-risk spatiotemporal condition combinations.

[0014] Furthermore, the specific content of the differentiated optimization measures includes: geometric design optimization measures, including setting up a dedicated left-turn lane for non-motorized vehicles, with a lane width of 1.5 to 2.0 meters, a guide line curvature of 15 to 20 meters, and using solid white lines to mark the guide lines with a line width of 15 centimeters; setting up a pre-waiting area, with dimensions of 5 to 8 meters long and 2 to 3 meters wide, paved with red anti-slip pavement, and surrounded by yellow warning lines with a warning line width of 10 centimeters; signal control optimization measures include... The measures include setting up dedicated left-turn phases for non-motorized vehicles, with a dedicated left-turn phase duration of 30 to 40 seconds; during peak hours, the green light duration for electric bicycles turning left should not be less than 30 seconds; and during off-peak hours, the green light ratio should be adjusted according to real-time traffic flow, with an adjustment range of 10% to 15%. The optimized control strategy includes implementing peak-hour left-turn restrictions, with the restricted hours set from 7:00 to 9:00 AM and from 5:00 PM to 7:00 PM. No-left-turn warning signs will be installed 50 meters from the intersection entrance, with a sign interval of 10 meters and a sign size of 1.2 meters × 0.8 meters.

[0015] Beneficial effects Compared with existing technologies, this method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints has the following advantages: I. This invention, by screening intersections covering different types and functional scenarios and conducting all-time data collection, combines target detection, trajectory reconstruction, and strict data quality control to construct a three-dimensional indicator system of safety, efficiency, and compliance. This enables a multi-dimensional and comprehensive characterization of the left-turn operation status of electric bicycles. Through the quantification of spatiotemporal coupling conflict intensity and three-dimensional coupling applicability scoring, the system integrates various influencing factors, scientifically classifies applicability levels, and accurately identifies high-risk spatiotemporal condition combinations. This breaks the limitations of single-dimensional assessment, making the assessment results more consistent with actual traffic operation characteristics. Relying on multi-technology integrated data analysis methods, it effectively avoids decision-making biases caused by one-sided data and extensive analysis in traditional assessments, providing a comprehensive and reliable basis for the formulation of subsequent optimization measures. This improves the systematicness and accuracy of the assessment of left turns by electric bicycles at intersections and provides data support for traffic management.

[0016] Second, this invention, by combining the differences in intersection type, time period characteristics, and functional scenarios, formulates differentiated optimization measures for the coordinated use of geometric design, signal control, and management strategies. This achieves precise adaptation to different applicable scenarios, avoiding the problems of single-mode and insufficient targeting in traditional optimization. Through a closed-loop effect verification mechanism, retesting is conducted at the same intersection using consistent standards, comparing changes in key indicators before and after optimization to ensure the actual effectiveness of the optimization measures. With the implementation path of multi-dimensional collaborative optimization and full-process verification, it effectively resolves problems such as prominent safety hazards, low traffic efficiency, and frequent violations during left turns of electric bicycles. It provides an operable technical method for intersection traffic organization optimization, promotes the transformation of traffic management from experience-driven to data-driven, improves the overall safety, orderliness, and efficiency of intersection operation, and contributes to the continuous improvement of the traffic environment.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart of a method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints; Figure 2 This is a schematic diagram illustrating the data transmission between steps in a method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1: Example of a T-shaped intersection between a main road and a secondary road.

[0022] S1, multi-time period data collection: A T-junction of a main road and a secondary road was selected. Non-motorized vehicles accounted for 22% of the total traffic flow at this intersection, with electric bicycles making up 28% of the total non-motorized traffic. The frequency of accidents involving electric bicycles turning left was higher than the average for urban intersections over the past year, meeting the criteria for a typical intersection. High-definition cameras were installed at this intersection. The cameras have a horizontal viewing angle of 95 degrees and a vertical viewing angle of 65 degrees. The ratio of the lens distance to the intersection center is 1:6. The installation location is free from high-voltage power lines and trees, and the vertical distance from ground obstacles is 2.2 meters. These installation parameters comprehensively capture the operational status of traffic participants in all areas of the intersection. Video recordings of electric bicycles, pedestrians, and motor vehicles were collected during all time periods: morning rush hour (7:00-9:00), evening rush hour (17:00-19:00), off-peak hour (10:00-16:00), and nighttime (20:00-22:00). Simultaneously, spatiotemporal parameters such as signal cycles, intersection geometry, and time period markers were recorded. This full-time data collection comprehensively presents left-turn traffic patterns under different traffic flow characteristics, providing comprehensive and accurate foundational data for subsequent analysis. Figure 1 As shown.

[0023] S2, Trajectory Reconstruction and Quality Control: Based on the collected data, the YOLOv8 algorithm was used for target detection. The confidence threshold was set to 0.82, the IOU threshold to 0.53, the input image size to 640×640 pixels, the batch size to 8, the number of training iterations to 1000, and the initial learning rate to 0.01. A cosine annealing learning rate decay strategy was adopted. These parameter settings can accurately identify targets such as electric bicycles, pedestrians, and motor vehicles. For multi-target tracking, the BoT-SORT algorithm was used. The process noise covariance matrix Q and observation noise covariance matrix R of the Kalman filter were set according to standard. When the target occlusion time exceeded 3 frames, trajectory prediction interpolation was used to supplement trajectory points to ensure the continuity of target tracking. After target identification and association are completed using the two algorithms described above, a homography mapping method is used for coordinate transformation. A square calibration board with a side length of 1 meter is set up at the data collection site. The calibration board is divided into 10×10 alternating black and white squares with a side length of 10 centimeters. During the setup, it is ensured that the horizontal distance between the center of the calibration board and the center of the intersection is 4.5 meters, and that the plane of the calibration board is parallel to the ground. Five ground control points are selected to establish a mapping relationship between pixel coordinates and real ground coordinates. The coordinates of the ground control points are measured using a GPS positioning instrument with a positioning accuracy of ±1.8 centimeters, achieving accurate conversion from pixel coordinates to real geographic coordinates. Smoothing is performed using Savitzky-Golay smoothing technology. The window size of this technology is set to 7 frames, and the polynomial order is set to 3rd order. Abnormal fluctuation points in the original trajectory data are weighted and fitted to remove noise interference caused by shooting shake and slight target offset, outputting continuous, smooth, and high-precision trajectory data. The positional deviation of the trajectory data is controlled within 0.05 meters, making the trajectory data more closely match the actual running trajectory. Feature parameters such as turning radius, instantaneous speed, path offset, collision time, intrusion time, and minimum spacing are extracted and processed according to data quality control rules. The number of valid frames for a single trajectory is no less than 30 frames, and the time interval between adjacent valid trajectory points does not exceed 0.1 seconds to ensure the integrity of the trajectory data. The 3σ principle is used to filter out abnormal values ​​of feature parameters. The normal range of instantaneous speed is set to 0.5 km / h to 25 km / h, and the normal range of acceleration is set to -3 m / s² to 3 m / s². Trajectory points with a spacing of no more than 0.1 meters or a speed of no less than 25 km / h for 3 consecutive frames are identified as abnormal points and removed to ensure the authenticity and validity of feature parameters and provide high-quality data support for subsequent modeling.

[0024] S3, 3D Applicability Modeling: Based on the extracted feature parameters, a three-dimensional indicator system of safety, efficiency, and compliance is constructed. Safety indicators include speed fluctuation, path deviation, conflict density, collision time, intrusion time, and minimum distance. Efficiency indicators include left-turn delay, passage time, number of stops and starts, green light utilization rate, and throughput. Compliance indicators include guideline deviation rate and violation rate, comprehensively covering the key evaluation dimensions of left-turn operation for electric bicycles. The comprehensive conflict intensity is calculated using a spatiotemporal coupling conflict intensity quantification formula. The mathematical expression of the spatiotemporal coupling conflict intensity quantification formula is as follows: ,in To assess the overall intensity of the conflict, Indicates the conflict type identifier. 1, 2, and 3 correspond to horizontal conflict, vertical conflict, and intersectional conflict, respectively. For conflict type weights, This is the basic risk value for horizontal conflict. This is the basic risk value for vertical conflict. The basic risk value for convergent conflicts, Weights are based on the time dimension. This is a time correction factor. For spatial dimension weights, This is the spatial correction factor. For dynamic adjustment of flow rate, For real-time left-turning electric bicycle traffic flow, Using the baseline traffic flow, the cumulative impact of different types of conflicts is objectively reflected; combined with the three-dimensional coupling applicability scoring formula, a comprehensive score is obtained, achieving a comprehensive quantitative assessment of the applicability of left turns at intersections. The mathematical expression of the three-dimensional coupling applicability scoring formula is as follows: ,in For comprehensive scoring, For security dimension weighting, As the weight for the efficiency dimension, As a weighting factor for compliance dimensions, To score for the security dimension, To score on the efficiency dimension, To score in the compliance dimension, For three-dimensional coupling coefficients, To assess the overall intensity of the conflict, The threshold for maximum conflict intensity. To optimize the potential correction coefficient.

[0025] The applicability level was determined using FCM fuzzy clustering technology, with a cluster size of 5. Weighted fuzzy clustering was employed, and the iteration termination condition was that the distance between the cluster centers of two iterations did not exceed 0.001. The maximum number of iterations was set to 100. Based on the comprehensive score, the applicability level of the intersection was determined to be low applicability, clearly defining the current operational status. Apriori technology was used to identify high-risk spatiotemporal condition combinations, with a minimum support of 0.2, a minimum confidence of 0.7, and a maximum itemset length of 4. A comprehensive score of low applicability and a comprehensive conflict intensity higher than 3.5 were used as screening criteria. The identified spatiotemporal conditions included the T-junction type, morning rush hour, high left-turn electric bicycle traffic volume, medium left-turn motor vehicle traffic volume, medium signal cycle duration, and narrow approach lane width. Redundant rules were removed using a pruning strategy, retaining the top 20% of rules by confidence, forming a high-risk spatiotemporal condition combination set to accurately locate key risk factors affecting left-turn operation.

[0026] S4, Adaptive Solution Development: Based on the lower applicable grade of the intersection and the combination of high-risk temporal and spatial conditions discovered, and considering the differences in type, time characteristics, and functional scenarios of the T-shaped intersection between the main road and the secondary road, differentiated optimization measures were formulated. In terms of geometric design optimization, a dedicated left-turn lane for non-motorized vehicles was set up, with a width of 1.8 meters and a guide line curvature of 18 meters. The guide lines were marked with solid white lines, 15 centimeters wide, to regulate the left-turn path of electric bicycles and reduce intersections with other road users. A pre-waiting area was also set up, with dimensions of 6 meters long and 2.5 meters wide, paved with red anti-slip pavement, and surrounded by yellow warning lines, 10 centimeters wide, to clearly define the waiting area for electric bicycles and improve safety during waiting periods. In terms of signal control optimization, a dedicated left-turn phase for non-motorized vehicles is set, with a phase duration of 35 seconds. During peak hours, the green light duration for electric bicycles turning left is set to 32 seconds to ensure the left-turn passage needs of electric bicycles during peak hours. During off-peak hours, the green light ratio is adjusted based on real-time traffic flow, with an adjustment range of 12%, achieving dynamic matching between signal timing and traffic flow. Regarding management strategy optimization, a peak-hour left-turn restriction is implemented, with restricted hours set from 7:00 AM to 9:00 AM and from 5:00 PM to 7:00 PM. No-left-turn warning signs are installed 50 meters from the intersection entrance, with a 10-meter interval and a sign size of 1.2 meters × 0.8 meters, to remind drivers in advance and reduce traffic conflicts during peak hours. These three measures work together to improve the safety, efficiency, and compliance of left-turn operations at intersections.

[0027] S5, Closed-loop performance verification: At the T-junction of the main road and secondary road, video retesting was conducted using the same equipment and processing methods to ensure the consistency and comparability of the retested data with the initial collected data, and to obtain various characteristic parameters and indicators after optimization. The overall conflict intensity, overall score, left-turn delay, and violation rate before and after optimization were compared to verify the optimization effect. The results showed that the overall conflict intensity was significantly reduced after optimization, the overall score improved to a moderately applicable level, the left-turn delay was shortened, and the violation rate decreased. This clearly demonstrates the actual effect of the optimization measures on improving the left-turn operation status of the intersection, ensuring the pertinence and effectiveness of the optimization plan.

[0028] In summary, this embodiment focuses on T-shaped intersections between main roads and secondary roads, strictly adhering to the entire process of the assessment and optimization method for the suitability of electric bicycles turning left at intersections under spatiotemporal constraints. Comprehensive basic data is acquired through multi-time-period data collection conforming to standards. High-precision trajectory reconstruction and quality control are achieved using the YOLOv8 algorithm, BoT-SORT algorithm, and related technologies. Suitability assessment and high-risk combination mining are completed based on a three-dimensional index system and dedicated quantitative formulas. Targeted optimization measures are formulated, coordinating geometric design, signal control, and management strategies. Closed-loop verification demonstrates a reduction in overall conflict intensity, an improvement in the overall score to a moderate suitability level, and a decrease in left-turn delays and violation rates. This fully reflects the method's relevance and effectiveness at this type of intersection, providing a reliable reference for improving the left-turn operation of electric bicycles at similar intersections.

[0029] Example 2: Example of a compact crossroads in a commercial area.

[0030] S1, multi-time period data collection: A compact intersection in a commercial area was selected as the implementation target. Non-motorized vehicles account for 27% of the total traffic flow at this intersection, with electric bicycles making up 30% of the total non-motorized vehicle traffic. Over the past year, the frequency of accidents involving electric bicycles turning left has been higher than the average accident frequency at urban intersections, meeting the selection criteria for a typical intersection. High-definition cameras were deployed at this intersection. The cameras have a horizontal viewing angle of 92 degrees and a vertical viewing angle of 63 degrees. The ratio of the lens distance to the intersection center is 1:7. The installation location is free from high-voltage power lines and tree obstructions, and the vertical distance from ground obstacles is 2.3 meters, suitable for the compact space of a commercial area intersection, ensuring clear capture of the operational details of all traffic participants. Video footage of electric bicycles, pedestrians, and motor vehicles is collected throughout the day, including morning peak hours (7:30-9:30), evening peak hours (17:30-19:30), off-peak hours (10:30-16:30), and nighttime hours (20:30-22:30). Simultaneously, spatiotemporal parameters such as signal cycles, intersection geometry, and time-of-day markings are recorded. This full-time coverage effectively reflects the left-turn operation characteristics under varying traffic flow patterns in the commercial area, providing comprehensive and realistic data support for subsequent assessments. Figure 2As shown.

[0031] S2, Trajectory Reconstruction and Quality Control: Based on the collected data, the YOLOv8 algorithm was used for target detection, with a confidence threshold of 0.85 and an IOU threshold of 0.55. The input image size was 640×640 pixels, the batch size was 8, the training iterations were 1000, and the initial learning rate was 0.01. A cosine annealing learning rate decay strategy was adopted to adapt to the scenario of dense traffic participants and complex targets in commercial areas, thereby improving the accuracy of target recognition. For multi-target tracking, the BoT-SORT algorithm was used. The process noise covariance matrix Q and observation noise covariance matrix R of the Kalman filter were set according to the standard. When the target occlusion time exceeded 3 frames, the trajectory prediction interpolation method was used to supplement the trajectory points to ensure the continuity of target tracking in complex traffic environments. After target identification and association, a homography mapping method is used for coordinate transformation. A square calibration board with a side length of 1 meter is set up on site. The calibration board has 10×10 alternating black and white squares with a side length of 10 centimeters. The horizontal distance between the center of the calibration board and the center of the intersection is 4.8 meters, and it is parallel to the ground. Six ground control points are selected to establish a mapping relationship between pixel coordinates and real ground coordinates. The GPS positioning instrument measures the coordinates of the ground control points with a positioning accuracy of ±1.5 centimeters, achieving accurate coordinate transformation. Savitzky-Golay smoothing technology is used for smoothing. The window size is 7 frames, the polynomial order is 3, and after removing noise interference, high-precision trajectory data is output. The position deviation is controlled within 0.05 meters, so that the trajectory data more accurately reflects the actual operation of traffic participants. After extracting relevant feature parameters, the data is filtered according to the data quality control rules. The number of valid frames for a single trajectory is no less than 30 frames, and the time interval between adjacent valid trajectory points is no more than 0.1 seconds to ensure the integrity of the trajectory. The 3σ principle is used to filter out outliers. If the instantaneous velocity and acceleration are within the specified normal range, outliers with a trajectory point spacing of no more than 0.1 meters or a velocity of no less than 25 km / h for 3 consecutive frames are removed to ensure that the feature parameters are true and reliable, laying a solid foundation for subsequent modeling and analysis.

[0032] S3, 3D Applicability Modeling: A three-dimensional indicator system encompassing safety, efficiency, and compliance is constructed, with each dimension comprehensively covering the core evaluation elements of left-turn operation for electric bicycles, enabling a systematic assessment of left-turn operation status at intersections in commercial areas. The comprehensive conflict intensity is calculated using a spatiotemporal coupling conflict intensity quantification formula, the mathematical expression of which is: ,in To assess the overall intensity of the conflict, Indicates the conflict type identifier. 1, 2, and 3 correspond to horizontal conflict, vertical conflict, and intersectional conflict, respectively. For conflict type weights, This is the basic risk value for horizontal conflict. This is the basic risk value for vertical conflict. The basic risk value for convergent conflicts, Weights are based on the time dimension. This is a time correction factor. For spatial dimension weights, This is the spatial correction factor. For dynamic adjustment of flow rate, For real-time left-turning electric bicycle traffic flow, Using the baseline flow rate, the degree of conflict under different spatiotemporal conditions is accurately quantified; combined with the three-dimensional coupling applicability scoring formula, a comprehensive score is obtained, achieving a comprehensive and objective quantification of left-turn applicability. The mathematical expression of the three-dimensional coupling applicability scoring formula is as follows: ,in For comprehensive scoring, For security dimension weighting, As the weight for the efficiency dimension, As a weighting factor for compliance dimensions, To score for the security dimension, To score on the efficiency dimension, To score in the compliance dimension, For three-dimensional coupling coefficients, To assess the overall intensity of the conflict, The threshold for maximum conflict intensity. To optimize the potential correction coefficient.

[0033] The applicability level was determined using FCM fuzzy clustering technology, with a cluster size of 5. A weighted fuzzy clustering method was used, and the iteration termination condition and maximum number of iterations were set according to standards. The intersection was assessed as having low applicability, clearly identifying serious problems with current left-turn operations. Apriori technology was then used to identify high-risk spatiotemporal condition combinations. Minimum support was set at 0.2, minimum confidence at 0.7, and maximum itemset length at 4. A comprehensive score of low applicability and a comprehensive conflict intensity higher than 3.5 were used as screening criteria. The identified spatiotemporal conditions included compact intersections in commercial areas, off-peak hours, high left-turn traffic flow for electric bicycles and motor vehicles, long signal cycle duration, and narrow approach lanes. A pruning strategy was used to retain the top 20% of rules with the highest confidence, forming a set of high-risk spatiotemporal condition combinations that precisely identified the key factors causing poor left-turn operation.

[0034] S4, Adaptive Solution Development: Based on the type, time-of-day characteristics, and functional differences of the compact intersection in this commercial area, targeted optimization measures were formulated according to the applicability level and high-risk combination. In terms of geometric design optimization, a dedicated left-turn lane for non-motorized vehicles was set up, with a width of 1.6 meters and a guide line curvature of 16 meters. The guide lines are marked with solid white lines, 15 centimeters wide, adapting to the limited space of the commercial area intersection, standardizing the left-turn path of electric bicycles, and reducing interference with other traffic flows. A pre-waiting area was set up, with dimensions of 5.5 meters long and 2.2 meters wide, paved with red anti-slip pavement, and surrounded by 10-centimeter-wide yellow warning lines, clearly defining the waiting area and improving safety and order during waiting periods. Regarding signal control optimization, a dedicated left-turn phase for non-motorized vehicles was set up with a duration of 33 seconds. During peak hours, the green light duration for left turns by electric bicycles is no less than 30 seconds to ensure peak traffic demand. During off-peak hours, the green light ratio is adjusted according to real-time traffic flow, with an adjustment range of 13%, achieving dynamic optimization of signal timing and improving traffic efficiency. In terms of optimizing the control strategy, a left-turn restriction is implemented during peak hours, from 7:00 to 9:00 AM and from 5:00 to 7:00 PM. Warning signs for the left-turn restriction are set up 50 meters from the intersection entrance, spaced 10 meters apart, and measuring 1.2 meters by 0.8 meters. This guides traffic flow in advance and reduces left-turn conflicts during peak hours. Through the coordinated optimization of geometric design, signal control, and control strategies, the operation of left turns at the intersection is comprehensively improved.

[0035] S5, Closed-loop performance verification: Video retests were conducted at the compact intersection in the commercial area using the same equipment and processing methods to ensure consistency between the retest data and the original data, thus guaranteeing the objectivity of the comparison. Optimized feature parameters and indicators were obtained, and the overall conflict intensity, overall score, left-turn delay, and violation rate before and after optimization were compared to verify the optimization effect. The results show that the overall conflict intensity decreased significantly after optimization, the overall score improved to a lower applicable level, left-turn delay was significantly reduced, and the violation rate was significantly lowered. This clearly demonstrates the improvement effect of the optimization measures on left-turn operation at the compact intersection in the commercial area, verifying the effectiveness and adaptability of the optimization scheme.

[0036] In summary, this embodiment focuses on compact intersections in commercial areas and implements the method step by step according to regulations. Starting with data collection from intersections that meet the screening criteria, high-quality data is obtained through standardized target detection, tracking, coordinate transformation, and quality control processes. Three-dimensional applicability modeling clarifies the current situation with low applicability and high-risk spatiotemporal combinations. Adaptive geometric, signal, and control optimization schemes are developed based on the characteristics of the commercial area scenario. After retesting and verification, the overall conflict intensity is significantly reduced after optimization, the overall score is improved to a low applicability level, and left-turn efficiency and compliance are significantly improved. This embodiment fully verifies the adaptability of the method to compact intersections in commercial areas and provides a feasible path for optimizing left turns for electric bicycles in space-constrained and complex traffic scenarios.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints, characterized in that, The specific steps of this method are as follows: S1, Multi-time period data acquisition: Screen typical intersections covering different types and functional scenarios, deploy high-definition cameras, and collect video of electric bicycles, pedestrians and motor vehicles during morning peak, evening peak, off-peak and nighttime, as well as spatiotemporal parameters such as signal cycle, intersection geometry and time period information. S2, Trajectory Reconstruction and Quality Control: Based on the collected data, target recognition and association are completed through target detection and multi-target tracking. High-precision trajectory is reconstructed through coordinate transformation and smoothing. Feature parameters such as turning radius, instantaneous speed, path offset, collision time, intrusion time, and minimum spacing are extracted. Data quality control is completed by setting trajectory continuity thresholds and outlier filtering rules. S3, 3D Applicability Modeling: Based on the extracted feature parameters, a 3D indicator system of safety, efficiency and compliance is constructed. The comprehensive conflict intensity is calculated through the spatiotemporal coupling conflict intensity quantification formula. The comprehensive score is obtained by combining the 3D coupling applicability scoring formula, applicability levels are divided and high-risk spatiotemporal condition combinations are identified. S4, Adaptive Solution Development: Based on the applicability level and high-risk combination, and combined with the differences in intersection type, time period characteristics and functional scenarios, develop differentiated optimization measures for the coordination of geometric design, signal control, and management strategies. S5, Closed-loop effect verification: Video retesting is conducted at the same intersection using the same equipment and processing methods to obtain optimized feature parameters and indicators. The overall conflict intensity, overall score, left turn delay, and violation rate are compared to complete the optimization effect verification.

2. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S1, the selection criteria for typical intersections include main road-to-main road crossroads, main road-to-secondary road T-junctions, secondary road-to-secondary road crossroads, and compact crossroads in commercial areas, with at least one intersection of each type selected. The selected intersections must meet the following requirements: non-motorized vehicle traffic accounts for 20% to 30% of the total traffic flow; the proportion of left-turning electric bicycles in the total number of non-motorized vehicles is not less than 25% on average; and the frequency of left-turning accidents involving electric bicycles is higher than the average accident frequency of urban intersections in the past year. The installation of fixed-point high-definition cameras must also meet the following requirements: the horizontal viewing angle of the lens is not less than 90 degrees, the vertical viewing angle is not less than 60 degrees, the ratio of the distance between the lens and the center of the intersection to the radius of the maximum circumscribed circle of the intersection is 1:5 to 1:8, the installation location is free from high-voltage power lines and trees, and the vertical distance from ground obstacles is not less than 2 meters.

3. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S2, the target detection uses the YOLOv8 algorithm, and the multi-target tracking uses the BoT-SORT algorithm. The confidence threshold of the YOLOv8 algorithm is set to no less than 0.8, the IOU threshold is set to no less than 0.5, the input image size is set to 640×640 pixels, the batch processing size is set to 8, the number of training iterations is set to 1000, the initial learning rate is set to 0.01, and a cosine annealing learning rate decay strategy is adopted. In the BoT-SORT algorithm, the process noise covariance matrix Q of the Kalman filter is set to diag([1e-2, 1e-2, 1e-2, 1e-2, 1e-4]), and the observation noise covariance matrix R is set to diag([1e-1, 1e-1, 1e-1, 1e-1]). When the target occlusion duration exceeds 3 frames, the trajectory prediction interpolation method is used to supplement the trajectory points.

4. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S2, the coordinate transformation adopts the homography mapping method. Specifically, a square calibration board with a side length of 1 meter is set up at the data acquisition site. The calibration board is set with 10×10 alternating black and white squares, with a side length of 10 centimeters. During the setup, it is ensured that the horizontal distance between the center of the calibration board and the center of the intersection does not exceed 5 meters, and the plane of the calibration board is parallel to the ground. At least 4 ground control points are selected to establish the mapping relationship between pixel coordinates and ground real coordinates. The coordinate measurement of the ground control points is carried out using a GPS positioning instrument, with a positioning accuracy of no more than ±2 centimeters. The smoothing process adopts Savitzky-Golay smoothing technology. The window size of this technology is set to 7 frames, and the polynomial order is set to 3rd order. By performing weighted fitting processing on abnormal fluctuation points in the original trajectory data, noise interference caused by shooting shake and slight target offset is eliminated, and continuous smooth high-precision trajectory data is output. The positional deviation of the trajectory data is controlled within 0.05 meters.

5. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S2, the specific rules for data quality control are as follows: the number of valid frames for a single trajectory is not less than 30 frames, and the time interval between adjacent valid trajectory points does not exceed 0.1 seconds; the 3σ principle is used to filter out abnormal values ​​of feature parameters, the normal range of instantaneous speed is set to 0.5 km / h to 25 km / h, and the normal range of acceleration is set to -3 m / s² to 3 m / s²; when the distance between trajectory points does not exceed 0.1 meters or the speed of 3 consecutive frames is not less than 25 km / h, it is judged as an abnormal point and is removed.

6. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S3, the specific contents of the three-dimensional indicator system of safety, efficiency and compliance are as follows: the safety dimension indicators include speed fluctuation for characterizing speed stability, path deviation for characterizing path compliance, conflict density for characterizing conflict frequency, collision time for characterizing collision urgency, intrusion time for characterizing intrusion severity, and minimum spacing for characterizing space margin. Efficiency indicators include left-turn delay (to represent time loss), travel time (to represent total travel time), number of stops and starts (to represent smoothness of travel), green light utilization rate (to represent green light time utilization efficiency), and throughput (to represent traffic capacity). Compliance indicators include guideline deviation rate (to represent route compliance) and violation rate (to represent traffic behavior violations).

7. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S3, the mathematical expression of the spatiotemporal coupling conflict intensity quantification formula is: ,in To assess the overall intensity of the conflict, Indicates the conflict type identifier. 1, 2, and 3 correspond to horizontal conflict, vertical conflict, and intersectional conflict, respectively. For conflict type weights, This is the basic risk value for horizontal conflict. This is the basic risk value for vertical conflict. The basic risk value for convergent conflicts, Weights are based on the time dimension. This is a time correction factor. For spatial dimension weights, This is the spatial correction factor. For dynamic adjustment coefficient of flow rate, For real-time left-turning electric bicycle traffic flow, This is the baseline flow rate.

8. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S3, the mathematical expression of the three-dimensional coupling applicability scoring formula is: ,in For comprehensive scoring, For security dimension weighting, As the weight of the efficiency dimension, As a weighting factor for compliance dimensions, To score for the security dimension, To score on the efficiency dimension, To score in the compliance dimension, For three-dimensional coupling coefficients, To assess the overall intensity of the conflict, The maximum conflict intensity threshold, To optimize the potential correction coefficient.

9. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S3, the specific implementation of classifying applicability levels and identifying high-risk spatiotemporal condition combinations is as follows: Applicability levels are classified using FCM fuzzy clustering technology, with a cluster size of 5. Weighted fuzzy clustering is used, and the clustering iteration termination condition is that the distance between the cluster centers of two iterations does not exceed 0.

001. The maximum number of iterations is set to 100. The classification criteria are determined based on a comprehensive score: a comprehensive score between 0.8 and 1.0 indicates high applicability; between 0.6 and 0.8 indicates relatively high applicability; between 0.4 and 0.6 indicates moderate applicability; and between 0.2 and 0.4 indicates relatively low applicability. Low applicability is defined as a score between 0 and 0.

2. Apriori technology is used to mine high-risk spatiotemporal condition combinations, with a minimum support of 0.2, a minimum confidence of 0.7, and a maximum itemset length of 4. The selection criteria are a comprehensive score of low applicability and a comprehensive conflict intensity higher than 3.

5. The mined spatiotemporal conditions include intersection type, time period, left-turning electric bicycle flow range, left-turning motor vehicle flow range, signal cycle duration, and approach lane width. Redundant rules are removed using a pruning strategy, retaining the top 20% of rules by confidence level to form a set of high-risk spatiotemporal condition combinations.

10. The method for evaluating and optimizing the applicability of electric bicycles turning left at intersections under spatiotemporal constraints according to claim 1, characterized in that, In step S4, the specific content of the differentiated optimization measures is as follows: Geometric design optimization measures include setting up a dedicated left-turn lane for non-motorized vehicles, with a lane width of 1.5 to 2.0 meters, a guide line curvature of 15 to 20 meters, and white solid lines used to mark the guide lines with a line width of 15 centimeters; setting up a pre-waiting area, with dimensions of 5 to 8 meters long and 2 to 3 meters wide, paved with red anti-slip pavement, and surrounded by yellow warning lines with a warning line width of 10 centimeters; signal control optimization measures include... The measures include setting up dedicated left-turn phases for non-motorized vehicles, with a dedicated left-turn phase duration of 30 to 40 seconds; during peak hours, the green light duration for electric bicycles turning left should not be less than 30 seconds; and during off-peak hours, the green light ratio should be adjusted according to real-time traffic flow, with an adjustment range of 10% to 15%. The optimized control strategy includes implementing peak-hour left-turn restrictions, with the restricted hours set from 7:00 to 9:00 AM and from 5:00 PM to 7:00 PM. No-left-turn warning signs will be installed 50 meters from the intersection entrance, with a sign interval of 10 meters and a sign size of 1.2 meters × 0.8 meters.

Images