Autonomous identification method, system, medium and product for overtaking without maintaining a safe distance

Abstract

This invention relates to an autonomous identification method, system, medium, and product for overtaking without maintaining a safe following distance, within the field of driving behavior analysis. The method includes acquiring video data, location data, and speed data of the target vehicle; extracting time-series data slices containing complete behavioral context through preset driving scenario filtering and risk driving behavior feature identification; calculating the headway based on the video data; generating driving behavior feature parameters by combining the location and speed data; inputting the feature parameters into a preset risk assessment model to obtain a risk category identifier; and determining that a failure to maintain a safe following distance is an act when the risk category is high and the headway remains below a preset threshold for a preset duration. This solution achieves a shift from analyzing independent behavioral features to analyzing overall risk patterns, and improves the accuracy of identifying failure to maintain a safe following distance through a dual-judgment mechanism, providing precise and reliable technical support for fleet safety management.

Description

Technical Field

[0001] This application relates to the field of driving behavior analysis, and in particular to an autonomous identification method, system, medium, and product for overtaking without maintaining a safe following distance. Background Technology

[0002] With the continuous expansion of logistics transportation and commercial fleets, fleet safety management has become an important part of enterprise operation and management. In long-distance transportation scenarios such as highways and national roads, drivers' failure to maintain a safe following distance is one of the main causes of rear-end collisions. Therefore, effective supervision and management of fleet vehicle driving behavior and timely identification of dangerous driving behaviors such as failure to maintain a safe following distance are of great significance for reducing accident rates and improving the overall safety level of the fleet.

[0003] Current fleet safety management technologies primarily employ a solution where the fleet management system collects vehicle speed and location data via onboard terminals. This data is then analyzed in the background, and dangerous driving behaviors are identified using a set of pre-defined rules. For example, a sudden and significant drop in vehicle speed exceeding a preset threshold is classified as emergency braking; multiple braking maneuvers recorded within a specific time window are classified as continuous braking. The system flags and records these identified events, generating driving behavior reports for fleet management review, thereby monitoring driver compliance.

[0004] However, in real-world applications, the detection results of single behavioral features such as sudden braking and continuous braking only reflect the instantaneous state of the driving operation. When a driver makes continuous small-amplitude decelerations to adjust distance, or frequently accelerates and decelerates in traffic flow, these scattered behavioral segments may individually conform to preset rules, but they do not necessarily constitute deliberate dangerous driving behavior. Therefore, rule-based methods face difficulties in associating these segments to identify underlying persistent risk patterns, thereby reducing the accuracy of driving behavior recognition and the effectiveness of fleet safety management. Summary of the Invention

[0005] This application provides an autonomous identification method, system, medium, and product for overtaking without maintaining a safe following distance, which is used to improve the accuracy and reliability of identifying such behavior.

[0006] Firstly, this application provides an autonomous identification method for overtaking without maintaining a safe following distance, applied to a vehicle distance identification system, comprising: acquiring driving data collected by the target vehicle during its driving process, the driving data including video data, positioning data, and speed data; identifying a first data segment in the driving data as a preset driving scenario based on the positioning data and speed data, the preset driving scenario including at least highways or national roads; identifying a second data segment containing risky driving behavior characteristics based on the speed data in the first data segment, the risky driving behavior characteristics including at least sudden deceleration, continuous braking, or accelerated approach; and extracting multiple time-series data slices based on the second data segment, each time-series data slice corresponding to a... The system includes a continuous time period containing characteristics of risky driving behavior; based on video data, it calculates the following distance between the target vehicle and the vehicle in front, and calculates the headway within the corresponding time period of the time series data slice based on the following distance and the speed data of the target vehicle; based on positioning data, speed data, and headway, it calculates the driving behavior characteristic parameters of the time series data slice; it inputs the driving behavior characteristic parameters into a preset risk assessment model to obtain the risk category identifier corresponding to the time series data slice, which includes high risk, medium risk, and low risk; when the risk category identifier is high risk and the headway is continuously lower than a first preset threshold within a first preset time period, it is determined that the time series data slice contains driving behavior of not maintaining a safe following distance.

[0007] By adopting the above technical solutions, the vehicle distance recognition system achieves accurate identification of behaviors that fail to maintain a safe following distance. First, by collecting multi-dimensional driving data and filtering preset driving scenarios, the system effectively focuses on high-risk highway or national road scenarios, avoiding interference from complex environments such as urban roads. Second, by identifying risky driving behavior characteristics and extracting time-series data slices, the system transforms continuous data streams into standardized analytical units, preserving complete behavioral context information. Third, by combining the headway calculated from video data with a risk assessment model, the system achieves a shift from analyzing single behavioral features to analyzing overall risk patterns, overcoming the limitation of traditional fixed-rule methods in identifying persistent risk patterns. Finally, through a dual determination mechanism of risk category identification and headway persistence, the system significantly improves the accuracy and reliability of identifying behaviors that fail to maintain a safe following distance, providing more precise technical support for fleet safety management.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, after extracting multiple time-series data slices based on the second data segment, the method further includes: based on the video data in the time-series data slices, identifying the number of vehicles in the video frames using a target detection algorithm, and calculating the vehicle density; based on the video data in the time-series data slices, identifying traffic flow movement patterns through continuous frame analysis, the traffic flow movement patterns including free-flow mode and intermittent flow mode, the free-flow mode being a mode of continuous vehicle movement, and the intermittent flow mode being a mode of alternating movement and stopping of vehicles; when the vehicle density is lower than a preset density threshold and the traffic flow movement mode is a free-flow mode, determining that the local traffic state of the target vehicle in the time-series data slice is a smooth-flowing state; when the vehicle density is higher than the preset density threshold or the traffic flow movement mode is an intermittent flow mode, determining that the local traffic state is a congested state.

[0009] By adopting the above technical solution, the vehicle distance recognition system achieves intelligent perception and analysis of the traffic environment. By using target detection algorithms to identify the number of vehicles in video frames and calculate vehicle density, combined with continuous frame analysis to identify traffic flow patterns, the system can comprehensively grasp the characteristics of the current traffic environment, providing crucial environmental reference for subsequent safe following distance determination. The combined analysis of vehicle density and traffic flow patterns makes the local traffic state determination more comprehensive and accurate, avoiding misjudgments that may arise from a single indicator. In summary, this solution improves the adaptability of the vehicle distance recognition system in complex and changing road environments, enabling the judgment criteria for failing to maintain a safe following distance to be dynamically adjusted according to actual traffic conditions, thereby improving the rationality and accuracy of the judgment results.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, before determining that the time-series data slice contains driving behavior of not maintaining a safe following distance when the risk category is identified as high risk and the headway between vehicles is continuously lower than the first preset threshold within a first preset duration, the method further includes: selecting a first preset threshold and a first preset duration corresponding to the current local traffic state from a preset set of parameters based on the local traffic state.

[0011] By adopting the above technical solution, the vehicle distance recognition system dynamically selects a first preset threshold and a first preset duration from a preset parameter set that match the current environment based on the determined local traffic conditions, allowing the judgment criteria to be flexibly adjusted according to environmental changes. Therefore, this solution solves the problem that fixed thresholds are difficult to adapt to different traffic conditions. It adopts stricter safety standards in smooth traffic conditions and appropriately relaxes requirements in congested conditions, conforming to actual driving patterns, thereby improving the environmental adaptability and robustness of the process for recognizing behaviors that fail to maintain a safe following distance.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of inputting driving behavior feature parameters into a preset risk assessment model to obtain the risk category identifier corresponding to the time series data slice, the method further includes: when the risk category identifier is medium risk or low risk, selecting a second preset threshold and a second preset duration corresponding to the current local traffic state from a preset parameter set according to the local traffic state, wherein the second preset threshold is less than the first preset threshold; if the headway between vehicles is continuously lower than the second preset threshold within the second preset duration, it is determined that the time series data slice contains driving behavior of not maintaining a safe following distance.

[0013] By adopting the above technical solution, the vehicle distance recognition system dynamically selects a second preset threshold and a second preset duration based on local traffic conditions for medium- or low-risk situations, and uses these as the judgment criteria. Through hierarchical judgment, the vehicle distance recognition system can perform refined identification of driving behaviors with different risk levels, avoiding a simplistic one-size-fits-all judgment method and improving the accuracy and reliability of the system's recognition results.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after determining that a time-series data slice contains driving behavior of not maintaining a safe following distance, the method further includes: extracting the time-series data slice determined to contain driving behavior of not maintaining a safe following distance; generating a visualization report based on the time-series data slice, the visualization report including at least a video segment and a risk category identifier.

[0015] By adopting the above technical solution, the vehicle distance recognition system achieves a visual presentation of the recognition results. The system extracts time-series data slices that are identified as containing driving behaviors that fail to maintain a safe following distance, and generates a visual report based on this data that includes video clips and risk category labels, thereby improving the interpretability of the recognition results.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, driving behavior feature parameters of time-series data slices are calculated based on positioning data, speed data, and headway. Specifically, this includes: calculating driving operation parameters based on positioning data and speed data, wherein the driving operation parameters include at least acceleration, deceleration, and rate of change of speed; calculating the time rate of change of headway to obtain following status parameters; and combining the driving operation parameters and following status parameters to form a feature vector to obtain driving behavior feature parameters.

[0017] By employing the above technical solution, the vehicle distance recognition system calculates driving operation parameters based on positioning and speed data, capturing driving operation characteristics. Simultaneously, it calculates the time change rate of the vehicle's headway to obtain following status parameters, accurately quantifying following behavior characteristics. This solution combines these two types of parameters into a feature vector, achieving a mathematical expression of driving behavior. This provides sufficient feature information for subsequent risk clustering analysis, thereby improving the classification accuracy and stability of the risk assessment model and providing a precise data foundation for identifying behaviors that fail to maintain a safe following distance.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after inputting driving behavior feature parameters into a preset risk assessment model to obtain the risk category identifier corresponding to the time-series data slice, the method further includes: when the risk category identifier is high risk, identifying the vehicle ahead based on the video data in the time-series data slice; extracting the motion trajectory data of the vehicle ahead in the time-series data slice, the motion trajectory data being used to represent the position information of the vehicle ahead; calculating the acceleration sequence of the vehicle ahead based on the motion trajectory data; when a negative acceleration value appears in the acceleration sequence of the vehicle ahead, and the absolute value of the negative acceleration value exceeds a preset acceleration threshold, determining that the high-risk state of the time-series data slice is caused by a sudden braking event of the vehicle ahead, and not classifying the time-series data slice as a driving behavior of the target vehicle actively failing to maintain a safe following distance.

[0019] By adopting the above technical solution, the distance recognition system achieves intelligent attribution of risk sources. When a high-risk situation is identified, the system analyzes the trajectory and acceleration sequence of the vehicle ahead to determine whether the high-risk situation is caused by a sudden braking event of the vehicle ahead or by the target vehicle's deliberate failure to maintain a safe following distance. In summary, this solution avoids unfair misjudgments of the driver, improves the fairness and credibility of the system's judgment results, and provides a more objective and accurate assessment of driving behavior for fleet management.

[0020] Secondly, this application provides a vehicle distance recognition system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the vehicle distance recognition system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a vehicle distance recognition system, cause the vehicle distance recognition system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, this application provides a computer program product that, when run on a vehicle distance recognition system, causes the vehicle distance recognition system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the vehicle distance recognition system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. By employing multi-dimensional driving data collection and analysis, preset driving scenario screening, risk driving behavior feature identification, time-series data slice extraction, vehicle headway calculation, driving behavior feature parameter calculation, risk assessment model classification, and a dual judgment mechanism, the vehicle distance recognition system can associate discrete driving behavior segments into complete risk patterns, realizing the transformation from single behavioral features to overall risk assessment. This effectively solves the problem in related technologies where it is difficult to identify persistent risk patterns based on fixed rules, thereby achieving high-precision autonomous identification of behaviors that fail to maintain a safe distance, significantly improving the accuracy and effectiveness of fleet safety management.

[0026] 2. By employing vehicle density calculation based on video data, traffic flow movement pattern recognition, and local traffic state determination mechanisms, the vehicle distance recognition system can comprehensively perceive and analyze the characteristics of the current traffic environment, accurately distinguish between smooth traffic and congested conditions, effectively solving the problem of neglecting traffic environment differences and resulting in a single judgment standard in related technologies. This enables the system to adaptively adjust the safe vehicle distance judgment standard in the environment, maintaining high accuracy in recognition in complex and ever-changing road environments and avoiding misjudgments and omissions under different traffic conditions.

[0027] 3. By employing mechanisms for identifying vehicles ahead, extracting motion trajectory data, calculating acceleration sequences, and determining emergency braking events, the distance recognition system can analyze the root causes of high-risk situations, distinguish between passive risk responses and proactive dangerous driving behaviors. This effectively solves the problem of misjudging driver responsibility due to the inability to determine the source of risk in related technologies, thereby achieving accurate attribution of risk states. It ensures that the system only judges the driver's proactive failure to maintain a safe distance, significantly improving the fairness and credibility of the system's judgment results and providing a more objective and accurate basis for assessing driving behavior for fleet management. Attached Figure Description

[0028] Figure 1This is a flowchart illustrating an autonomous identification method for overtaking without maintaining a safe following distance, as described in this application.

[0029] Figure 2 This is another flowchart illustrating an autonomous identification method for overtaking without maintaining a safe following distance, as described in this application embodiment.

[0030] Figure 3 This is a schematic diagram of the physical device structure of a vehicle distance recognition system in the embodiments of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0034] In related technologies, vehicle speed and location data can be analyzed using a fixed set of rules to identify single-risk driving behavior characteristics such as sudden deceleration and continuous braking. The following scenario illustrates an autonomous identification method for overtaking without maintaining a safe following distance, based on this technology: A long-haul truck is traveling on a highway. Due to changes in traffic density ahead, the driver performs multiple small-amplitude deceleration maneuvers within a few minutes to adjust the following distance. A fixed-rule-based solution would identify these scattered deceleration maneuvers as multiple independent "continuous braking" events and report them. Fleet managers see a series of isolated braking warnings in the backend, but cannot determine whether there is a dangerous pattern of continuous close following behind these maneuvers, making it difficult to accurately assess the driver's behavior.

[0035] The autonomous identification method for overtaking without maintaining a safe following distance, as described in this application, acquires video data, positioning data, and speed data. It identifies risky driving behavior characteristics and extracts time-series data slices, calculates the headway and multi-dimensional driving behavior characteristic parameters, uses a risk assessment model to evaluate the overall risk pattern, and finally combines the risk category identifier and the persistence of the headway for a dual determination. This achieves accurate identification of the behavior of failing to maintain a safe following distance, not only capturing the driver's instantaneous operations but also correlating these operations to assess their potential persistent risks. The following describes a scenario using this autonomous identification method for overtaking without maintaining a safe following distance: For the same scenario described above, the method extracts a complete time-series data slice containing multiple deceleration operations. By analyzing the video data within this slice, it calculates that the headway is consistently below a safe value. Combined with the high-frequency acceleration and deceleration driving behavior characteristic parameters calculated from the speed data, the risk assessment model identifies this slice as high-risk. Ultimately, because the two conditions of high risk and the headway being consistently below the first preset threshold were met, the system determined that the behavior was a complete failure to maintain a safe following distance event and generated a report containing video footage, providing managers with a complete event context and reliable basis for judgment.

[0036] As can be seen, the autonomous identification method for overtaking without maintaining a safe following distance in the embodiments of this application can not only accurately identify the behavior of not maintaining a safe following distance, but also effectively solve the technical problems in related technologies that it is difficult to associate scattered behavior fragments based on fixed rules and cannot identify continuous risk patterns, thereby improving the accuracy and reliability of identifying the behavior of not maintaining a safe following distance.

[0037] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating an autonomous identification method for overtaking without maintaining a safe following distance, as described in this application.

[0038] S101. Acquire driving data collected by the target vehicle during its driving process. The driving data includes video data, positioning data, and speed data.

[0039] Among them, the target vehicle refers to the specific vehicle for which driving behavior monitoring and analysis is required, usually a commercial vehicle or freight vehicle in a fleet management system; driving data refers to the multi-dimensional information collected by the target vehicle through various sensors and devices during its operation, which reflects the vehicle's operating status and the driver's operating behavior; video data refers to the digital image sequence obtained by continuous shooting through camera equipment installed on the target vehicle, which includes visual information about the road environment, traffic conditions and other vehicles in front of the vehicle; positioning data is used to represent the precise location information of the target vehicle in the Earth coordinate system, which usually includes longitude, latitude, altitude and corresponding timestamp; speed data refers to the instantaneous speed value of the target vehicle during its operation, reflecting the changes in the vehicle's motion state.

[0040] The vehicle distance recognition system performs this step when the target vehicle starts and begins driving. Specifically, the system synchronously collects multi-dimensional driving data of the target vehicle through various sensor modules integrated in the onboard terminal equipment. First, the system activates the front-facing camera module to continuously capture the road scene ahead of the vehicle at a frequency of 30 frames per second, acquiring high-definition video stream data including timestamp information. Simultaneously, the system activates the high-precision GNSS positioning module to acquire the real-time geographical coordinates of the target vehicle at a frequency of no less than 1Hz, recording the UTC timestamp of each positioning point. Furthermore, the system reads real-time speed information provided by the vehicle's electronic control unit via the CAN bus interface or OBD-II port, ensuring the accuracy and real-time nature of the speed data. Finally, the system performs time synchronization processing on all collected data, establishing a unified time reference, and performs preliminary formatting and encapsulation of the data.

[0041] S102. Based on the positioning data and speed data, identify the driving scenario in the driving data as the first data segment of the preset driving scenario. The preset driving scenario includes at least highways or national roads.

[0042] The first data segment represents a subset of data selected from the complete driving data that meets the conditions of a specific driving scenario. This subset is used to narrow the data range for subsequent analysis and improve processing efficiency. The preset driving scenario refers to a predefined type of road environment with specific risk characteristics. These scenarios typically have high vehicle speeds and relatively simple traffic flow patterns. The preset driving scenario setting scheme is based on traffic accident statistics and risk assessment results. It selects road types with a high incidence of rear-end collisions and significant impact on following distance behavior. Its purpose is to filter out interference from low-speed and complex scenarios such as urban congested roads and focus on analyzing high-risk long-distance driving scenarios.

[0043] The vehicle distance recognition system performs this step after completing the driving data collection. Specifically, the system first performs map matching processing on the collected positioning data, comparing and analyzing the GPS trajectory point sequence with a high-precision electronic map database. The system then calls the map service API to query the attribute information of the current road based on the target vehicle's latitude and longitude coordinates, including key parameters such as road grade, speed limit, and number of lanes. The system establishes a road type discrimination rule set; when it detects a vehicle continuously traveling on a section marked as a highway or national road, it extracts all driving data within the corresponding time period as candidate segments. The system further verifies this by combining speed data, analyzing indicators such as average speed and maximum speed within the time period to confirm that the vehicle is indeed traveling at high speed, eliminating false judgments caused by GPS drift or map data errors. The system then marks the verified data segment as the first data segment and records its start time, end time, geographical location range, and other metadata information.

[0044] Optionally, in some embodiments, the vehicle distance recognition system can utilize a machine learning model for intelligent recognition of driving scenarios. The vehicle distance recognition system pre-trains a scene classification model based on a random forest algorithm. This model takes multi-dimensional features such as speed statistics, position change rate, and road curvature as input. The vehicle distance recognition system converts positioning and speed data within a continuous time window into feature vectors and inputs them into the trained model. The model outputs the classification result and confidence score of the current driving scenario. When the model determines that it is a highway or national road scenario and the confidence score exceeds a preset threshold, the vehicle distance recognition system marks the corresponding data segment as the first data segment.

[0045] S103. Based on the speed data in the first data segment, identify a second data segment containing characteristics of risky driving behavior, which includes at least sudden deceleration, continuous braking, or accelerated approach.

[0046] The second data segment represents a subset of data containing potentially risky driving operations that is further filtered from the first data segment, used to accurately locate time periods where potential safety hazards may exist; risky driving behavior characteristics refer to vehicle movement patterns that can reflect abnormal driver operation or safety risks, and these characteristics are usually strongly correlated with the occurrence of rear-end collisions.

[0047] The distance recognition system executes this step immediately after acquiring the first data segment. Specifically, the system performs time-series analysis and feature extraction on the speed data in the first data segment. First, it calculates the first derivative of the speed data to obtain an acceleration sequence. Then, using a sliding window algorithm, it detects time points where the acceleration value is below the rapid deceleration threshold and records the occurrence time and duration of these rapid deceleration events. Next, the system establishes a braking behavior detection algorithm. By analyzing the frequency and amplitude characteristics of speed changes, it identifies continuous braking patterns. When multiple small decelerations or a single long deceleration is detected within a preset time window, the corresponding time period is marked as a continuous braking event. Finally, the system implements a pattern matching algorithm for acceleration and approach behavior. By detecting a rising-falling characteristic pattern in the speed curve, it identifies the sequence of driver-initiated acceleration followed by forced deceleration. The system then timestamps and classifies all detected risky driving behavior feature events, extracting the time periods containing these events and a certain range before and after them as the second data segment, ensuring the capture of complete risky behavior context information.

[0048] The setting scheme for the rapid deceleration threshold is based on vehicle dynamics principles and human physiological limits, and is usually set between -4.0 m / s² and -6.0 m / s². Its function is to distinguish between normal deceleration and emergency braking behavior. The judgment parameters for continuous braking include braking frequency threshold and duration threshold, which are set to brake more than 3 times within 10 seconds or a single braking lasting more than 5 seconds, respectively, to identify abnormal braking patterns. The recognition parameters for acceleration approach include acceleration change rate threshold and speed change amplitude threshold, to capture typical risky behavior patterns of accelerating first and then decelerating.

[0049] S104. Based on the second data segment, extract multiple time-series data slices, each time-series data slice corresponding to a continuous time period containing characteristics of risky driving behavior;

[0050] Among them, time-series data slices represent continuous data segments with fixed durations divided from the second data segment according to the time dimension. Each slice contains complete driving behavior context information. The duration setting scheme of time-series data slices is based on the typical duration of driving behavior and the characteristics of human reaction time, and is usually set to 10-20 seconds. Its purpose is to ensure that the slice can contain the complete behavior sequence without introducing too much irrelevant information. The overlap parameter of time-series data slices is usually set to 50% to avoid important behavioral features being segmented by slice boundaries and causing information loss.

[0051] After identifying risky driving behavior characteristics, the vehicle distance recognition system performs this step, decomposing the continuous data stream into standardized data units that are easy to analyze. Specifically, the system first traverses all labeled risky driving behavior characteristic events in the second data segment, obtaining the precise timestamp and event type information for each event. Using the occurrence time of each risky event as a reference point, the system traces back a preset pre-set duration (e.g., 8 seconds) and extends forward a preset post-set duration (e.g., 7 seconds), forming a time window with a total length of 15 seconds. Within the defined time window, the system extracts all relevant driving data, including video frame sequences, GPS location point sequences, speed data sequences, etc., ensuring that each time-series data slice contains complete multimodal information. The system then performs a quality check on the extracted time-series data slices, verifying the integrity and continuity of the data and removing incomplete slices caused by sensor malfunctions or signal interruptions. The vehicle distance recognition system assigns a unique identifier to each time-series data slice, records its corresponding risk behavior characteristic type, time range, data source and other metadata information, and establishes a slice index for subsequent batch processing and analysis.

[0052] Optionally, in some embodiments, the vehicle distance recognition system can utilize an adaptive window extraction algorithm for intelligent segmentation of time-series data slices. The system first analyzes the duration and intensity characteristics of risky driving behaviors, dynamically adjusting the slice window size based on the typical duration of different types of behaviors. A shorter time window is used for rapid deceleration events, while a longer time window is used for continuous braking and acceleration approaches. The system automatically identifies the start and end boundaries of behaviors using a signal change point detection algorithm, ensuring that each slice contains the complete behavior cycle. The system also controls the overlap of generated slices, merging adjacent slices when the overlap exceeds a preset ratio to avoid data redundancy.

[0053] S105. Based on video data, calculate the following distance between the target vehicle and the vehicle in front, and calculate the headway within the corresponding time period of the time series data slice based on the following distance and the speed data of the target vehicle.

[0054] The following distance refers to the straight-line physical distance between the front of the target vehicle and the rear of the vehicle in front, and is an important indicator for measuring the safe distance between vehicles; the vehicle in front refers to the other motor vehicle that is closest to the target vehicle in the same lane and in the direction of travel; the headway is used to indicate the time required for the front of the following vehicle to reach a fixed point just after the rear of the vehicle in front has passed, and is a standardized time indicator for assessing following safety.

[0055] The vehicle distance recognition system performs this step immediately after obtaining the time-series data slice, quantifying the spatiotemporal relationship between the target vehicle and the vehicle in front and establishing a basis for safety assessment. Specifically, the vehicle distance recognition system first analyzes the video data in the time-series data slice frame by frame, using a pre-trained target detection model to identify vehicle targets in the video frames and generate bounding box coordinates. The vehicle distance recognition system establishes a forward vehicle filtering algorithm, identifying the nearest vehicle in the same lane directly in front of the target vehicle as the forward vehicle by analyzing the position, size, and trajectory of the detected vehicle targets in the image. The vehicle distance recognition system uses a monocular vision ranging algorithm, combined with the camera's intrinsic calibration parameters and installation geometry parameters, to convert the position information of the forward vehicle in the image coordinate system into actual physical distance values. The vehicle distance recognition system obtains the target vehicle speed data at the corresponding time in the time-series data slice, and uses the headway calculation formula THW=D / V (where D is the following distance and V is the target vehicle speed) to calculate the instantaneous headway at each moment. The vehicle distance recognition system continuously calculates the headway distance of vehicles within the entire time-series data slice, generates time-series data of headway distance, and performs smoothing filtering to eliminate calculation errors and noise interference.

[0056] S106. Based on positioning data, speed data, and headway, calculate the driving behavior characteristic parameters of the time-series data slices;

[0057] Among them, driving behavior feature parameters represent a set of numerical indicators that can quantify the driver's operating style and vehicle motion state, and are used to convert complex driving behavior into a computable and analyzable feature vector. The numerical range and normalization scheme of driving behavior feature parameters need to be standardized according to the actual data distribution to ensure that features of different dimensions have the same weight.

[0058] After calculating the following distance and headway, the vehicle distance recognition system performs this step, converting the multi-source, heterogeneous raw data into a standardized feature representation. Specifically, the system first performs statistical analysis on the speed data in the time-series data slices, calculating basic statistical features such as maximum and minimum speeds. It then calculates the acceleration sequence through numerical differentiation, further extracting dynamic features such as maximum acceleration, maximum deceleration, and root mean square acceleration. Next, the system performs spatial analysis on the positioning data, calculating geometric features such as the total length of the driving trajectory, average heading angle, and trajectory curvature. It assesses the vehicle's lateral and longitudinal motion stability by analyzing the rate of change of GPS coordinates. Finally, the system performs time-domain analysis on the headway sequence, calculating following behavior characteristics such as the mean and minimum headway, the percentage of time below the safe threshold, and the rate of change in headway. These parameters directly reflect the driver's following habits and risk tolerance. The vehicle distance recognition system standardizes all calculated feature parameters and uses Z-score standardization or maximum-minimum normalization to eliminate dimensional differences between different features. Finally, it generates a fixed-dimensional driving behavior feature parameter vector as a numerical representation of the time-series data slice.

[0059] S107. Input the driving behavior characteristic parameters into the preset risk assessment model to obtain the risk category identifiers corresponding to the time series data slices. The risk category identifiers include high risk, medium risk and low risk.

[0060] The preset risk assessment model refers to a pre-trained and validated machine learning classifier used to map driving behavior feature parameters to corresponding risk level categories; the risk category label refers to the classification label output by the model that represents the degree of risk of driving behavior, used to quantitatively assess the safety risk level of the current driving behavior.

[0061] After calculating the driving behavior feature parameters, the vehicle distance recognition system performs this step, utilizing machine learning technology to intelligently identify and classify the inherent risk patterns of driving behavior. Specifically, the system first loads a pre-trained risk assessment model, which typically employs ensemble learning algorithms such as random forests or gradient boosting decision trees, capable of processing multi-dimensional feature vectors and outputting probability distributions. The system uses the driving behavior feature parameter vector calculated in step S106 as input data to the model, ensuring that the dimension and format of the input features are completely consistent with the requirements during model training. The system then calls the model's prediction interface. Internally, the model analyzes and processes the input feature vectors through algorithms such as feature weight calculation, decision tree voting, or neural network forward propagation, outputting the probability distribution for each risk category. Based on the probability distribution output by the model, the system determines the final risk category label, selecting the category with the highest probability as the prediction result, and simultaneously recording the confidence score of the prediction. The system then performs post-processing on the model output results, including quality control steps such as confidence threshold filtering and result consistency checks, ensuring that the output risk category labels have sufficient reliability and stability.

[0062] Optionally, in some embodiments, the risk assessment model is trained in the following ways:

[0063] The vehicle distance recognition system first collects a large amount of historical driving data, including driving behavior characteristic parameters of different drivers under various road conditions and corresponding safety event records. Traffic safety experts manually annotate each time-series data slice, classifying driving behavior into three levels: high risk, medium risk, and low risk, and establishing an annotated dataset containing no less than 10,000 samples.

[0064] The vehicle distance recognition system preprocesses the labeled dataset, including feature normalization, outlier removal, and data balancing. It uses stratified sampling to divide the dataset into training, validation, and test sets in a 7:2:1 ratio to ensure that the distribution of each risk category remains consistent in each subset.

[0065] The vehicle distance recognition system selects the random forest algorithm as the basic classifier, sets the number of decision trees to 200, and the maximum depth to 15 layers. The model hyperparameters are optimized through grid search and cross-validation. During training, feature importance evaluation and recursive feature elimination techniques are used to select the optimal feature subset. The validation set is used to evaluate the model performance and an early stopping strategy is used to prevent overfitting. Finally, the generalization ability of the model is verified on the test set to ensure that the classification accuracy reaches more than 85% and the recall rate of each category is balanced.

[0066] S108. When the risk category is identified as high risk and the headway between vehicles is continuously lower than the first preset threshold within the first preset time period, it is determined that the time-series data slice contains driving behavior of not maintaining a safe following distance.

[0067] The first preset duration represents the length of the time window used to determine the continuity of the headway. It is an important time standard for assessing whether following behavior constitutes a continuous risk. The setting scheme of the first preset duration is based on the safety theory of driver reaction time and braking distance, and is usually set to 3-5 seconds. Its function is to filter instantaneous fluctuations in headway and ensure that continuous dangerous following behavior is identified. The first preset threshold is the critical value for determining whether the headway is too small. It is used to distinguish between safe following behavior and dangerous following behavior. The first preset threshold is set according to international traffic safety standards and vehicle dynamics principles, and is usually 1.5-2.0 seconds. Its function is to establish the minimum time standard for safe following. Following behavior below this threshold is considered to have a high risk of rear-end collision.

[0068] After obtaining the risk category identifier, the vehicle distance recognition system executes this step, identifying behaviors that fail to maintain a safe following distance through a multi-condition judgment mechanism. Specifically, the vehicle distance recognition system first checks the risk category identifier output in step S107 to confirm whether the current time series data slice is classified as high-risk. Only slices that meet the high-risk conditions will proceed to the subsequent headway analysis process. The vehicle distance recognition system extracts the complete headway time series within the corresponding time period of the time series data slice and preprocesses the series data, including outlier detection, data smoothing, and missing value imputation, to ensure that the data quality meets the analysis requirements. The vehicle distance recognition system implements a sliding window analysis algorithm, using a window size of a first preset duration to slide and scan the headway time series, detecting whether the headway value within each window is continuously lower than the first preset threshold. The vehicle distance recognition system establishes a persistence judgment logic; when it finds that the headway is lower than the first preset threshold for a continuous time period, and the length of that time period reaches or exceeds the first preset duration, it confirms that the persistence condition is met. The vehicle distance recognition system comprehensively evaluates two key conditions: risk category identification and vehicle headway continuity. When both conditions are met, the system formally determines that the time-series data slice contains driving behavior that fails to maintain a safe following distance and generates corresponding event records and alarm information.

[0069] In this embodiment, the vehicle distance recognition system employs a complete technical solution, including multi-dimensional driving data collection, preset driving scenario screening, risk driving behavior feature recognition, time-series data slice extraction, vehicle headway calculation, driving behavior feature parameter extraction, risk assessment model classification, and a dual judgment mechanism. This enables the system to achieve intelligent analysis throughout the entire process, from raw data to accurate identification of risky behaviors. It effectively solves the technical problems in related technologies where fixed rules make it difficult to correlate scattered behavioral fragments and identify persistent risk patterns. Consequently, it achieves high-precision autonomous identification of behaviors that fail to maintain a safe following distance, significantly improving the accuracy and effectiveness of fleet safety management and providing reliable technical support for reducing rear-end collision rates and improving the overall safety level of the fleet.

[0070] Based on the above embodiments, the method provided in this embodiment will be described in further detail below. Please refer to... Figure 2 This is another flowchart illustrating an autonomous identification method for overtaking without maintaining a safe following distance, as described in this application.

[0071] S201. Based on the second data segment, extract multiple time-series data slices, each time-series data slice corresponding to a continuous time period containing characteristics of risky driving behavior;

[0072] This step and Figure 1 The description of step S103 in the above embodiment is similar and will not be repeated here.

[0073] S202. Based on video data in time-series data slices, the number of vehicles in video frames is identified by a target detection algorithm, and the vehicle density is calculated.

[0074] Among them, vehicle density represents the number of vehicles in a unit space within a specific road area or time period, and is an important quantitative indicator for measuring the busyness and congestion of road traffic; target detection algorithm refers to machine learning algorithm that uses computer vision technology to automatically identify and locate specific target objects in images or videos, and can output the target's category, location, and confidence information.

[0075] The vehicle distance recognition system performs this step immediately after extracting the time-series data slices, quantifying the vehicle distribution density of the current traffic environment. Specifically, the system first preprocesses the video data in the time-series data slices. Then, it loads a pre-trained lightweight object detection model, such as YOLOv5 or SSD-MobileNet, which is specifically optimized for vehicle detection tasks, ensuring both detection accuracy and real-time performance. The system defines a region of interest (ROI) in each video frame, typically within a 100-150 meter range of the current lane and adjacent lanes. Perspective transformation is used to convert pixel coordinates to the actual road coordinate system, ensuring accurate distance calculations. The system performs object detection on each frame within the ROI, identifying all vehicle targets and recording their bounding box coordinates and confidence scores. Detection results with confidence scores below 0.5 are filtered out to reduce false detections. Finally, the system counts the number of valid vehicles detected in each frame and averages the number of vehicles over the entire time-series data slice to obtain the average vehicle density value for that time period.

[0076] S203. Based on video data in time-series data slices, identify traffic flow movement patterns through continuous frame analysis. Traffic flow movement patterns include free-flow mode and intermittent flow mode. Free-flow mode is a mode in which vehicles move continuously, and intermittent flow mode is a mode in which vehicles alternate between moving and stopping.

[0077] Among them, continuous frame analysis refers to computer vision technology that compares and analyzes pixel changes and target motion between adjacent video frames in a time-series data slice, and is used to extract the dynamic features and motion patterns of traffic flow; vehicle flow motion pattern represents the overall motion state and behavioral characteristics of vehicle groups on the road, reflecting the macroscopic dynamic characteristics of traffic flow; free flow pattern refers to the local traffic state in which vehicles can travel continuously at the desired speed on the road with little mutual influence between vehicles, which usually occurs in unobstructed road sections with low traffic density; intermittent flow pattern is used to represent the local traffic state in which vehicles exhibit stop-and-go motion characteristics on the road, and vehicle speed frequently switches between zero and a certain positive value, which usually occurs at traffic light controlled intersections or congested road sections.

[0078] The vehicle distance recognition system executes this step immediately after calculating vehicle density. Specifically, the system first performs optical flow analysis on consecutive video frames in the time-series data slice, using the Lucas-Kanade optical flow algorithm to calculate the pixel motion vector field between adjacent frames. The system extracts statistical features of the optical flow vectors within the region of interest, including the average amplitude, principal direction consistency, and temporal continuity of the motion vectors. By analyzing these features, the system determines the overall traffic flow trend. The system then establishes a motion pattern recognition algorithm. When the detected optical flow vector field exhibits consistent direction, stable amplitude, and continuous motion, it is classified as a free-flowing mode; when the detected optical flow vector field alternates between stationary and moving states, or when the motion vector amplitude exhibits periodic fluctuations, it is classified as an intermittent flow mode.

[0079] S204. When the vehicle density is lower than the preset density threshold and the traffic flow mode is free flow mode, the local traffic state of the target vehicle in the time series data slice is determined to be smooth flow state.

[0080] Among them, local traffic conditions represent the macroscopic traffic characteristics and congestion level of the road environment where the target vehicle is located, which is an important environmental factor affecting the standard of safe following distance; smooth traffic conditions are used to represent the ideal traffic conditions where the road traffic flow is moderate, vehicles can travel freely at a relatively high speed, and there is little interference between vehicles; the preset density threshold setting scheme is based on traffic engineering theory and actual road capacity analysis, and is usually set to 5-8 vehicles per kilometer of road, which is used to distinguish between low-density and high-density traffic conditions.

[0081] S205. When the vehicle density is higher than the preset density threshold or the traffic flow pattern is intermittent, the local traffic condition is determined to be congested.

[0082] Among them, congestion refers to a traffic condition in which road traffic flow is close to or exceeds road capacity, vehicle speed is low, the distance between vehicles is small, and stop-and-go traffic occurs frequently.

[0083] S206. Based on video data, calculate the following distance between the target vehicle and the vehicle in front, and calculate the headway within the corresponding time period of the time series data slice based on the following distance and the speed data of the target vehicle.

[0084] This step and Figure 1 The description of step S104 in the above embodiment is similar and will not be repeated here.

[0085] S207. Based on positioning data, speed data, and headway, calculate the driving behavior characteristic parameters of the time-series data slices;

[0086] This step specifically includes:

[0087] Based on positioning and speed data, driving operation parameters are calculated, which include at least acceleration, deceleration and rate of change of speed.

[0088] Calculate the rate of change of the headway to obtain the following status parameters;

[0089] The driving operation parameters and following status parameters are combined to form a feature vector, thus obtaining the driving behavior feature parameters.

[0090] Among them, driving operation parameters represent a set of numerical indicators that quantify the driver's vehicle control behavior and are used to reflect the driver's operating style and habit characteristics; following status parameters refer to numerical indicators that describe the relative motion relationship between the target vehicle and the vehicle in front and are used to quantify the safety and stability of following behavior.

[0091] After calculating the headway, the vehicle distance recognition system performs this step, converting the multi-source heterogeneous raw data into a standardized feature representation, providing structured input for subsequent risk clustering analysis. Specifically, the system first calculates driving operation parameters based on positioning and speed data. It performs time differentiation on the speed data, calculating the instantaneous acceleration value at each moment, and extracts feature parameters for acceleration and deceleration processes based on the positive or negative value of the acceleration. The system calculates statistical features of acceleration, including maximum acceleration, maximum deceleration, average acceleration, and acceleration standard deviation, which reflect the driver's operational intensity and stability. By analyzing the fluctuation characteristics of the speed curve, the system calculates speed change rate indices, including speed fluctuation frequency and the ratio of speed change amplitude to average speed, to quantify the smoothness of driving operations. Next, the system calculates the time change rate of the headway. By performing time differentiation on the headway sequence, it obtains the rate of change of the headway and extracts its maximum, minimum, and average values, forming following status parameters. Finally, the vehicle distance recognition system combines the driving operation parameters and following status parameters into a fixed-dimensional feature vector according to a predefined order and format, which serves as the mathematical representation of the driving behavior feature parameters, ensuring that the data structure is consistent with the input requirements of the subsequent risk assessment model.

[0092] S208. Input the driving behavior characteristic parameters into the preset risk assessment model to obtain the risk category identifiers corresponding to the time series data slices. The risk category identifiers include high risk, medium risk and low risk.

[0093] This step and Figure 1 The description of step S107 in the above embodiment is similar and will not be repeated here.

[0094] S209. Based on the local traffic conditions, select a first preset threshold and a first preset duration corresponding to the current local traffic conditions from a preset parameter set;

[0095] The preset parameter set represents a predefined safe following distance judgment parameter library for different local traffic conditions. It contains multiple sets of threshold parameters applicable to different road environments, including at least a first preset duration and a first preset threshold, as shown in Table 1. The first preset threshold is the critical value for judging whether the headway is too small, used to distinguish between safe following and dangerous following behaviors. It is usually set to 1.8-2.0 seconds in smooth traffic conditions and 1.2-1.5 seconds in congested conditions. Its function is to establish a safe following standard that conforms to the characteristics of the current traffic environment. The first preset duration represents the length of the time window used to judge the persistence of the headway. It is an important time standard for assessing whether following behavior constitutes a persistent risk. It is usually set to 4-5 seconds in smooth traffic conditions and 2-3 seconds in congested conditions. Its function is to adjust the judgment standard for persistent dangerous behaviors according to different local traffic conditions.

[0096] Local traffic conditions Speed ​​range (km / h) First preset threshold (seconds) First preset duration (seconds) Unobstructed travel status 60-80 2.2 3.0 80-100 2.5 3.5 100-120 2.8 4.0 Congestion 60-80 1.5 2.5 80-100 1.8 3.0 >100 (instantaneous) 2.0 3.0

[0097] Table 1

[0098] After obtaining the local traffic condition determination result and risk category identifier, the vehicle distance recognition system executes this step, dynamically adjusting the safe distance determination standard based on the current road environment characteristics. Specifically, the vehicle distance recognition system first reads the local traffic condition result determined in step S204 or S205 to confirm whether the traffic environment type corresponding to the current time-series data slice is a smooth flow or a congested state. The vehicle distance recognition system accesses a pre-configured parameter set database, which stores multiple sets of determination parameters optimized for different local traffic conditions. Each set of parameters includes two key indicators: a first preset threshold and a first preset duration. Based on the current local traffic condition, the vehicle distance recognition system retrieves the matching parameter set from the parameter set. When the local traffic condition is smooth flow, a higher first preset threshold and a longer first preset duration are selected; when the local traffic condition is congested, a lower first preset threshold and a shorter first preset duration are selected. The vehicle distance recognition system can also combine contextual information such as the target vehicle's current speed and road type to fine-tune the basic parameter values, obtaining a more refined and suitable first preset threshold and first preset duration.

[0099] Optionally, to avoid abrupt parameter changes at speed interval boundaries, the vehicle distance recognition system performs function fitting on the parameters for each local traffic state. For example, for the first preset threshold under smooth traffic conditions, a linear or quadratic function can be fitted based on the data points in the table above:

[0100] First preset threshold_Smooth Traffic (v) = a × v + b

[0101] Where v is the real-time speed of the target vehicle, and a and b are coefficients fitted from the data in the table.

[0102] Similarly, a corresponding adjustment function can be fitted for the first preset duration and the parameters under congestion conditions.

[0103] When the vehicle distance recognition system needs to select parameters, it first selects the corresponding parameter adjustment function based on the determined local traffic conditions. Then, it substitutes the target vehicle's current real-time speed into the function to calculate a precise and continuously changing first preset threshold and a first preset duration.

[0104] Optionally, the vehicle distance recognition system can adopt a dynamic parameter adjustment method based on vehicle speed. First, a baseline parameter value is selected based on the local traffic conditions. Then, a secondary adjustment is made based on the current speed of the target vehicle. The vehicle distance recognition system establishes a functional relationship model between speed and safe distance. When the vehicle speed is high, the first preset threshold is appropriately increased, and when the vehicle speed is low, the first preset threshold is appropriately decreased. At the same time, the first preset duration is adjusted accordingly to ensure that the parameter settings conform to both the characteristics of the local traffic conditions and the current vehicle speed conditions.

[0105] S210. When the risk category is identified as high risk, identify the vehicle ahead based on the video data in the time-series data slice.

[0106] The vehicle distance recognition system executes this step after confirming that the risk category identifier is high-risk. Its purpose is to accurately identify and locate vehicles ahead that may be associated with a high-risk situation. Specifically, the vehicle distance recognition system first checks the risk category identifier output in step S208 to confirm whether the current time-series data slice is classified as high-risk. Only in high-risk situations will the detailed vehicle ahead recognition process be triggered. The vehicle distance recognition system processes the video data in the time-series data slice frame by frame, applying a target detection algorithm to identify all vehicle targets in each frame. The vehicle distance recognition system implements a lane recognition algorithm, determining the lane area where the target vehicle is located by analyzing road markings or vehicle movement trajectories. The vehicle distance recognition system establishes a forward vehicle filtering logic, selecting the closest vehicle in the same lane as the target vehicle from all detected vehicles as the forward vehicle.

[0107] S211. Extract the motion trajectory data of the vehicle in front from the time-series data slice. The motion trajectory data is used to represent the position information of the vehicle in front. Calculate the acceleration sequence of the vehicle in front based on the motion trajectory data.

[0108] Among them, the motion trajectory data represents the spatial position change sequence of the vehicle in front over a continuous time period, which is used to describe the vehicle's motion state and behavioral characteristics; the position information refers to the coordinate values ​​of the vehicle in a two-dimensional or three-dimensional spatial coordinate system, which usually includes the lateral position, longitudinal position and timestamp; the acceleration sequence represents the acceleration change data set of the vehicle in front over a continuous time period, which is used to quantify the longitudinal dynamic characteristics of the vehicle and the intensity of driving operations.

[0109] After successfully identifying the vehicle ahead, the vehicle distance recognition system performs this step to obtain precise motion characteristic data of the vehicle ahead. Specifically, the vehicle distance recognition system first establishes a continuous target tracking chain based on the vehicle target identified in step S210, covering the entire time range of the time-series data slice. The vehicle distance recognition system extracts the position data of the vehicle ahead in each frame. The vehicle distance recognition system arranges the extracted position data in chronological order to form complete motion trajectory data, with each trajectory point containing lateral position, longitudinal position, and corresponding timestamp information. Based on the position trajectory data, the vehicle distance recognition system calculates the displacement change between adjacent time points and calculates the velocity sequence by combining the time interval. The vehicle distance recognition system further performs time differentiation on the velocity sequence to obtain the acceleration sequence of the vehicle ahead, which directly reflects the acceleration and deceleration behavior characteristics of the vehicle ahead.

[0110] S212. If a negative acceleration value appears in the acceleration sequence of the vehicle ahead, and the absolute value of the negative acceleration value exceeds the preset acceleration threshold, the high-risk state of the time series data slice is determined to be caused by the sudden braking event of the vehicle ahead, and the time series data slice is not determined to be the target vehicle's active driving behavior of not maintaining a safe distance.

[0111] The preset acceleration threshold represents the critical acceleration value for determining an emergency braking event. The preset acceleration threshold is set based on vehicle dynamics characteristics and emergency braking feature analysis, and is usually set between -4.0 m / s² and -6.0 m / s². Its function is to distinguish between normal deceleration and emergency braking behavior.

[0112] After acquiring the acceleration sequence of the vehicle ahead, the distance recognition system performs this step to analyze the root cause of the high-risk situation and distinguish between proactive risk behavior and reactive response behavior. Specifically, the distance recognition system first analyzes the acceleration sequence of the vehicle ahead calculated in step S211, detects the presence of negative acceleration values, and records the time point, duration, and deceleration intensity of all deceleration events. The distance recognition system compares the detected negative acceleration values ​​with a preset acceleration threshold. When the absolute value of the negative acceleration value at a certain time point or multiple consecutive time points exceeds the preset acceleration threshold, the distance recognition system determines that the vehicle ahead has experienced a sudden braking event at that moment. The distance recognition system further analyzes the temporal characteristics of the sudden braking event, including the time of occurrence, duration, and morphological characteristics of the deceleration process, and assesses its urgency and impact on following safety. The distance recognition system compares the occurrence time of the sudden braking event with the risk behavior characteristic time of the target vehicle to confirm the temporal and causal relationship between the two. When it is confirmed that the sudden braking event of the vehicle in front occurs before or simultaneously with the high-risk state of the target vehicle, the vehicle distance recognition system determines that the high-risk state of the current time-series data slice is a passive risk caused by the sudden behavior of the vehicle in front, rather than the target vehicle's active failure to maintain a safe distance.

[0113] Optionally, in some embodiments, to avoid misjudging normal overtaking behavior as failure to maintain a safe following distance, the vehicle distance recognition system also includes a false alarm suppression mechanism in overtaking scenarios.

[0114] Specifically, after determining that the high-risk condition in the time-series data slice was not caused by the sudden braking of the vehicle in front, the vehicle distance recognition system further extracts the lateral movement characteristics of the target vehicle relative to the lane boundary line based on the video data in the time-series data slice. The vehicle distance recognition system calculates the rate of change of the lateral distance between the target vehicle's centerline and the left or right lane line of the current lane to obtain the lateral movement speed.

[0115] When the lateral movement speed exceeds a preset lane change threshold (e.g., 0.2 m / s) and the target vehicle's speed is greater than the speed of the vehicle in front, the distance recognition system determines that the target vehicle is in the process of overtaking and changing lanes. At this time, the distance recognition system starts an overtaking tolerance timer and sets an overtaking exemption duration (e.g., 5 seconds).

[0116] If the time difference between the vehicle head and the vehicle head is lower than the first preset threshold for a duration within the overtaking exemption period, and the target vehicle completes a lane change (i.e., enters the adjacent lane laterally) before the end of the exemption period, the distance recognition system determines that the close following during this period is a normal overtaking behavior and does not determine it as a failure to maintain a safe distance.

[0117] Conversely, if the time difference between the vehicle head and the target vehicle is lower than the first preset threshold for a duration exceeding the overtaking exemption period, or if the target vehicle returns to the center of the original lane after making lateral movement (abandoning overtaking), the distance recognition system maintains the risk assessment and executes the subsequent logic for confirming failure to maintain a safe distance.

[0118] Finally, in order to cover the complete overtaking behavior loop, the distance recognition system can also perform a secondary following safety assessment for the "lane change and return to the original lane" phase.

[0119] When the system detects that the target vehicle is moving laterally towards the center of the original lane and is about to enter the original lane, it will lock the target as the new vehicle in front of the target vehicle in the original lane.

[0120] At this point, the vehicle distance recognition system calculates the initial following distance (return-to-lane distance) when the target vehicle enters its original lane. If the return-to-lane distance is lower than the first preset threshold, the vehicle distance recognition system activates "return-to-lane adjustment monitoring": 1) If the target vehicle exhibits significant deceleration behavior (e.g., deceleration exceeding -1.0 m / s²) within the second preset time period (e.g., 3 seconds) after returning to the lane to actively increase the headway, it is determined to be a safe overtaking and return-to-lane behavior and is not recorded as a violation; 2) If the target vehicle does not exhibit deceleration behavior after returning to the lane, or if the headway remains lower than the second preset threshold (e.g., the critical safety threshold of 1.0 second) after the return-to-lane adjustment period ends, it is determined that the time-series data slice contains driving behavior that fails to maintain a safe following distance, and its risk type is marked as dangerous insertion following.

[0121] Furthermore, to identify high-risk approaching maneuvers with overtaking intent, the distance recognition system simultaneously monitors the approach characteristics and lateral stability of the target vehicle during the overtaking tolerance timer. Specifically, this includes:

[0122] First, calculate the Time to Collision (TTC): If, during the overtaking and lane-changing process, the TTC of the target vehicle and the vehicle in front remains below the emergency braking threshold (e.g., 1.5 seconds), it indicates that the target vehicle is approaching the vehicle in front at an excessively high relative speed. The distance recognition system immediately terminates the overtaking exemption and determines it as an "aggressive approach" behavior that fails to maintain a safe distance.

[0123] Second, monitor lateral sway characteristics: If the vehicle distance recognition system detects that the lateral center position of the target vehicle sways back and forth multiple times (more than twice) within a preset time window (e.g., 3 seconds) during close following, and the sway amplitude exceeds a preset proportion of the vehicle width (e.g., 20%), it is judged as "unstable serpentine tailgating" behavior. At this time, even if the target vehicle eventually completes the lane change, the vehicle distance recognition system still determines that the time-series data slice in this process contains driving behavior that does not maintain a safe distance and marks the risk type as "high risk - malicious tailgating".

[0124] Through the above-mentioned overtaking intent recognition, malicious tailgating monitoring, and return-to-lane safety assessment, the system has achieved accurate differentiation of different types of following behaviors throughout the overtaking process.

[0125] S213. When no negative acceleration value appears in the acceleration sequence of the vehicle in front, and the headway is continuously lower than the first preset threshold within the first preset time period, it is determined that the time sequence data slice contains driving behavior of not maintaining a safe distance.

[0126] This step and Figure 1 The description of step S108 in the above embodiment is similar and will not be repeated here.

[0127] S214. When the risk category is identified as medium risk or low risk, a second preset threshold and a second preset duration corresponding to the current local traffic state are selected from the preset parameter set according to the local traffic state. The second preset threshold is less than the first preset threshold.

[0128] The second preset threshold represents the critical value used to determine whether the headway is too small under medium or low risk conditions. Its value is less than the first preset threshold. It is usually set to 1.2-1.5 seconds under smooth traffic conditions and 0.8-1.0 seconds under congested conditions. Its function is to set differentiated safety standards for different risk levels. The second preset duration represents the length of the time window used to determine the duration of headway in medium or low risk conditions. It is usually set to 6-8 seconds under smooth traffic conditions and 4-5 seconds under congested conditions. Its function is to require a longer duration under medium or low risk conditions to determine that the behavior of not maintaining a safe following distance is not being maintained.

[0129] The vehicle distance recognition system executes this step after confirming that the risk category is medium or low risk, employing differentiated judgment criteria for different risk levels. Specifically, the system first checks the risk category identifier output in step S208 to confirm whether the current time-series data slice is classified as medium or low risk. The system then reads the local traffic state result determined in steps S204 or S205 to confirm whether the traffic environment type corresponding to the current time-series data slice is smooth or congested. The system accesses a pre-configured parameter set database and, based on the current risk category identifier and local traffic state, retrieves the most matching parameter set from the parameter set, similar to step S209. When the risk category identifier is medium risk, a moderately strict second preset threshold and second preset duration are selected; when the risk category identifier is low risk, a relatively lenient second preset threshold and a longer second preset duration are selected. The system can also fine-tune the basic parameter values ​​by combining contextual information such as the target vehicle's current speed, road type, and weather conditions to achieve more refined parameter adaptation.

[0130] Optionally, when retrieving the second preset threshold and the second preset duration, risk level attenuation coefficients can be defined, as shown in Table 2. The vehicle distance recognition system predefines a set of attenuation coefficients (K_th for the threshold and K_d for the duration) bound to the risk category identifier. These attenuation coefficients are all less than or equal to 1, representing the degree of leniency of the medium- and low-risk standards compared to the high-risk standards.

[0131] The calculation of the second preset threshold and the second preset duration is no longer based on consulting a separate table, but is dynamically calculated based on the first preset threshold and the first preset duration already defined for high-risk scenarios in step S209, including:

[0132] Second preset threshold = K_th × First preset threshold

[0133] Second preset duration = K_d × First preset duration

[0134] When the vehicle distance recognition system determines the risk category to be medium or low risk, it first looks up the corresponding K_th and K_d from the attenuation coefficient table based on the risk category identifier. Then, it multiplies these two coefficients by the first preset threshold and the first preset duration, as shown in Table 1. In this way, the second preset threshold (smaller) and the second preset duration (longer) have a clear calculation source and logical relationship. For example, the threshold for a medium-risk event is 80% of the corresponding high-risk standard, but the duration required is 1.2 times that of the high-risk standard, which makes the judgment criteria both discriminative and intrinsically linked.

[0135] Risk Category Identifier Threshold decay coefficient (K_th) Duration amplification factor (K_d) Strictness description High risk 1.0 1.0 benchmark Medium risk 0.8 1.2 medium strict Low risk 0.6 1.5 Relatively lenient

[0136] Table 2

[0137] S215. If the headway between vehicles remains below the second preset threshold for a second preset duration, the time-series data slice is determined to contain driving behavior that fails to maintain a safe following distance.

[0138] The vehicle distance recognition system executes this step after selecting the second preset threshold and the second preset duration. Its purpose is to assess the headway distance under low-to-medium risk conditions and identify potential failures to maintain a safe following distance. Specifically, the system first extracts the complete headway distance time series within the corresponding time period of the time series data slice. The series data is preprocessed, and then a sliding window analysis algorithm is used to scan the headway distance sequence with a window size of the second preset duration, detecting whether the headway distance value within each window is consistently lower than the second preset threshold. When a continuous time period is found where the headway distance is consistently lower than the second preset threshold, and the length of this time period reaches or exceeds the second preset duration, the persistence condition is confirmed. The system comprehensively evaluates the numerical and temporal characteristics of the headway distance. When it is confirmed that the headway distance is consistently lower than the second preset threshold within the second preset duration, the system formally determines that the time series data slice contains driving behavior that fails to maintain a safe following distance and generates corresponding event records and alarm information.

[0139] Optionally, in some embodiments, when the risk category is identified as medium or low risk, and the headway distance remains below a second preset threshold for a second preset duration, the distance recognition system does not immediately determine it as a driving behavior of not maintaining a safe following distance. Instead, it marks the time-series data slice as a potential risk event and stores it in a temporary observation sequence. Within a preset observation time window (e.g., 10 minutes or 20 kilometers), the distance recognition system continuously accumulates and analyzes the potential risk events in the observation sequence. When the number of potential risk events accumulated in the observation sequence exceeds a preset frequency threshold (e.g., occurring 3 or more times within 10 minutes), or the total duration of headway distances below the second preset threshold in all potential risk events exceeds a preset cumulative duration threshold (e.g., accumulating to 45 seconds), the distance recognition system comprehensively determines that the driver has a continuous driving behavior of not maintaining a safe following distance within this observation time window and generates an aggregated violation report.

[0140] S216. Extract time-series data slices that are determined to contain driving behaviors that fail to maintain a safe following distance; generate a visualization report based on the time-series data slices, the visualization report including at least video clips and risk category identifiers.

[0141] Among them, the visualization report is an information presentation document that presents the identification results to users in an intuitive and easy-to-understand form, in order to support fleet management and driving behavior improvement.

[0142] The distance recognition system performs this step after determining whether a driver has failed to maintain a safe following distance. The purpose is to present the analysis results in an intuitive and easy-to-use format for fleet management review and driver improvement. Specifically, the system first filters all time-series data slices to identify those containing driving behaviors that fail to maintain a safe following distance. These slices represent potentially dangerous driving segments requiring close attention. The system extracts complete data from each selected slice, including raw video data, headway data, risk category indicators, and relevant environmental information and judgment criteria. The system processes the raw video data, extracting keyframes and adding information annotations, such as headway values, safety threshold lines, and risk level indicators, enhancing the video's information content and intuitiveness. The system designs standardized report templates, including modules for basic information (time, location, vehicle information), video display, data analysis, and risk assessment, ensuring a clear report structure and complete content. The vehicle distance recognition system organizes and arranges core content such as processed video clips, vehicle headway curves, and risk category labels according to a template format to generate a structured visual report.

[0143] In this embodiment, by employing a complete technical solution including time-series data slice extraction, vehicle density calculation, traffic flow movement pattern recognition, local traffic state determination, vehicle headway calculation, driving behavior feature parameter extraction, risk assessment model classification, dynamic parameter selection, forward vehicle recognition, motion trajectory analysis, emergency braking event determination, and risk classification determination, the vehicle distance recognition system can achieve comprehensive intelligent analysis of traffic environment perception, risk level classification, and risk source attribution. This effectively solves the technical problems of neglecting traffic environment differences, having a single risk judgment standard, and being unable to distinguish the responsible party for risk in related technologies. Furthermore, it achieves highly reliable identification of behaviors that fail to maintain a safe following distance, including environmental adaptability, risk classification, and accurate attribution of responsibility, thus improving the accuracy, fairness, and practical value of the system's judgment results. This solution also further reduces the false alarm rate and improves scenario adaptability by recognizing and exempting overtaking intentions and distinguishing between normal overtaking behavior and malicious approaching behavior.

[0144] The vehicle distance recognition system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 3 This is a schematic diagram of the physical device structure of a vehicle distance recognition system in an embodiment of this application.

[0145] It should be noted that, Figure 3 The structure of the vehicle distance recognition system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0146] like Figure 3As shown, the vehicle distance recognition system includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in the read-only memory ROM 302 or a program loaded from the storage section 308 into the random access memory RAM 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O interface 305 is also connected to the bus 304.

[0147] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0148] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0149] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0151] Specifically, the vehicle distance recognition system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the autonomous recognition method for overtaking without maintaining a safe distance provided in the above embodiment.

[0152] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the vehicle distance recognition system described in the above embodiments; or it may exist independently and not assembled into the vehicle distance recognition system. The storage medium carries one or more computer programs, which, when executed by a processor of the vehicle distance recognition system, enable the vehicle distance recognition system to implement the autonomous recognition method for overtaking without maintaining a safe following distance provided in the above embodiments.

[0153] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0154] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0155] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. An autonomous identification method for overtaking without maintaining a safe following distance, applied to a vehicle distance identification system, characterized in that, include: Acquire driving data collected during the driving process of the target vehicle, including video data, positioning data, and speed data; Based on the positioning data and the speed data, a first data segment in the driving data is identified as a preset driving scenario, wherein the preset driving scenario includes at least a highway or national road. Based on the speed data in the first data segment, a second data segment containing characteristics of risky driving behavior is identified, the characteristics of risky driving behavior including at least sudden deceleration, continuous braking or acceleration approach; Based on the second data segment, multiple time-series data slices are extracted, and each time-series data slice corresponds to a continuous time period containing the characteristics of the risky driving behavior; Based on the video data, the following distance between the target vehicle and the vehicle in front is calculated, and the headway within the corresponding time period of the time-series data slice is calculated based on the following distance and the speed data of the target vehicle. Based on the positioning data, the speed data, and the headway, the driving behavior characteristic parameters of the time-series data slice are calculated; The driving behavior characteristic parameters are input into a preset risk assessment model to obtain the risk category identifiers corresponding to the time series data slices. The risk category identifiers include high risk, medium risk, and low risk. When the risk category is identified as high risk, and the headway between vehicles remains below a first preset threshold for a first preset duration, the time-series data slice is determined to contain driving behavior that fails to maintain a safe following distance.

2. The method according to claim 1, characterized in that, After extracting multiple time-series data slices based on the second data segment, the method further includes: Based on the video data in the time-series data slices, the number of vehicles in the video frames is identified by a target detection algorithm, and the vehicle density is calculated. Based on the video data in the time-series data slices, traffic flow movement patterns are identified through continuous frame analysis. The traffic flow movement patterns include free-flow mode and intermittent flow mode. The free-flow mode is a mode in which vehicles move continuously, and the intermittent flow mode is a mode in which vehicles alternate between moving and stopping. When the vehicle density is lower than a preset density threshold and the traffic flow movement mode is a free flow mode, the local traffic state of the target vehicle in the time-series data slice is determined to be a smooth flow state. When the vehicle density is higher than the preset density threshold or the traffic flow pattern is an intermittent flow pattern, the local traffic condition is determined to be a congested state.

3. The method according to claim 2, characterized in that, Before the step of determining that the time-series data slice contains driving behavior of not maintaining a safe following distance when the risk category is identified as high risk and the headway is continuously lower than a first preset threshold for a first preset time period, the method further includes: Based on the local traffic condition, a first preset threshold and a first preset duration corresponding to the current local traffic condition are selected from a preset set of parameters.

4. The method according to claim 2, characterized in that, After the step of inputting the driving behavior characteristic parameters into a preset risk assessment model to obtain the risk category identifier corresponding to the time-series data slice, the method further includes: When the risk category is identified as medium risk or low risk, a second preset threshold and a second preset duration corresponding to the current local traffic state are selected from a preset parameter set based on the local traffic state, wherein the second preset threshold is less than the first preset threshold. If the headway between vehicles remains below a second preset threshold for a second preset duration, then the time-series data slice is determined to contain driving behavior that fails to maintain a safe following distance.

5. The method according to claim 1, characterized in that, After determining that the time-series data slice contains driving behavior of not maintaining a safe following distance, the method further includes: Extract time-series data slices that are determined to contain driving behaviors that fail to maintain a safe following distance; A visualization report is generated based on the time-series data slices, and the visualization report includes at least the video clips and the risk category identifiers.

6. The method according to claim 1, characterized in that, Based on the positioning data, the speed data, and the headway, the driving behavior characteristic parameters of the time-series data slice are calculated, specifically including: Based on the positioning data and the speed data, driving operation parameters are calculated, and the driving operation parameters include at least acceleration, deceleration and rate of change of speed; Calculate the rate of change of the time distance between the vehicle heads to obtain the following status parameters; The driving operation parameters and the following status parameters are combined to form a feature vector, thus obtaining the driving behavior feature parameters.

7. The method according to claim 1, characterized in that, After the step of inputting the driving behavior characteristic parameters into a preset risk assessment model to obtain the risk category identifier corresponding to the time-series data slice, the method further includes: When the risk category is identified as high risk, the vehicle ahead is identified based on the video data in the time-series data slice; Extract the motion trajectory data of the vehicle ahead from the time-series data slice, and the motion trajectory data is used to represent the position information of the vehicle ahead. Calculate the acceleration sequence of the vehicle ahead based on the motion trajectory data; When a negative acceleration value appears in the acceleration sequence of the vehicle ahead, and the absolute value of the negative acceleration value exceeds a preset acceleration threshold, the high-risk state of the time-series data slice is determined to be caused by the sudden braking event of the vehicle ahead, and the time-series data slice is not determined to be the target vehicle's active driving behavior of not maintaining a safe distance.

8. A vehicle distance recognition system, characterized in that, The vehicle distance recognition system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the vehicle distance recognition system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the vehicle distance recognition system, the vehicle distance recognition system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the vehicle distance recognition system, it causes the vehicle distance recognition system to perform the method as described in any one of claims 1-7.

Images