A motor vehicle side slip amount comprehensive detection method based on multi-sensor fusion

Abstract

The application discloses a motor vehicle side slip amount comprehensive detection method based on multi-sensor fusion, and relates to the technical field of motor vehicle dynamic detection; the method comprises the following steps: arranging collected multi-source data in the same section to obtain same-section original data; according to the same-section original data, trace data is obtained through change trace extraction; according to the trace data, stage data is obtained by dividing the same-section original data into change stages; according to the same-section original data and the stage data, duty data is obtained through stage adaptation duty determination; according to the same-section original data, the stage data and the duty data, side slip chain data is obtained through adjacent stage succession analysis; according to the side slip chain data, the stage data and the duty data, side slip amount comprehensive detection results are obtained through continuous integration; through joint analysis of the corresponding relationship of multi-source data, change stage boundaries and stage succession relationships, the application improves the accuracy and continuity of motor vehicle side slip amount comprehensive detection under complex working conditions.

Description

Technical Field

[0001] This application relates to the field of motor vehicle dynamic detection technology, and more specifically, to a comprehensive detection method for motor vehicle sideslip based on multi-sensor fusion. Background Technology

[0002] When a motor vehicle is turning, changing lanes, navigating slopes, making emergency maneuvers, or driving on low-friction surfaces, the actual direction of its movement deviates from its longitudinal direction. The degree of this deviation can be characterized by sideslip. Sideslip reflects the lateral stability of a motor vehicle and is a key detection target in motor vehicle dynamic monitoring, instability warning, and related control processes. Therefore, accurately obtaining the sideslip of a motor vehicle under different operating scenarios has become a pressing technical problem to be solved in the field of motor vehicle dynamic monitoring.

[0003] In existing technologies, sideslip amount is often difficult to obtain directly and reliably using a single sensor. Therefore, a comprehensive method for detecting vehicle sideslip amount based on multi-sensor fusion is commonly adopted. This method jointly processes wheel speed data, steering angle data, yaw rate data, lateral acceleration data, longitudinal acceleration data, and other data related to vehicle operating status to output the corresponding sideslip amount detection result. This type of method can utilize the complementary information between different sensors, thereby improving the feasibility of sideslip amount detection to a certain extent.

[0004] However, existing multi-sensor fusion-based methods for comprehensive detection of vehicle sideslip typically treat multiple information sources as parallel inputs for fusion processing. This approach assumes that the information sources have established a stable correspondence within the current detection interval and can collectively characterize the dynamic changes of the same vehicle. In practical applications, the sampling start and end times, sampling periods, and effective recording durations of data from different sensors are often inconsistent. When existing technologies directly call upon each information source for processing within a unified detection interval, it is easy for temporally mismatched data to be included in the same detection process simultaneously, or for data that should correspond to the dynamic changes of the same vehicle to fail to establish a stable correspondence. This results in subsequent detection results being based on misaligned or incomplete data.

[0005] Meanwhile, the state changes related to sideslip during vehicle operation are usually not presented in a single, smooth, and synchronous manner, but rather occur in conjunction with acceleration, deceleration, steering input, sustained yaw, and alternating lateral and longitudinal responses. The response time and duration of the same vehicle's dynamic changes may also differ between different information sources. Most existing technologies directly fuse data around the original detection range, lacking clear distinction between the boundaries of different change processes within that range. This leads to multiple change processes that should be identified separately easily becoming mixed up in the same processing chain, making it difficult for the detection results to accurately reflect the start and end ranges and sequential relationships of each change process.

[0006] Furthermore, under complex operating conditions, different information sources react differently to the sideslip state. Some information sources respond early in the initial stage of change, some maintain a longer and more continuous response during the change, and some show strong continuity between adjacent time periods. Existing technologies typically treat different information sources as uniform inputs at the same level, lacking further differentiation of the differences in the effects of different information sources at different time periods. Consequently, in scenarios such as low-speed to high-speed switching, low-excitation to high-excitation switching, low-attachment to high-attachment switching, and normal operation to short-term disturbance recovery, existing technologies are prone to problems such as unclear dominant criteria and unclear relationships between the two sides of the boundary, thus affecting the accurate judgment of the actual sideslip change process.

[0007] Furthermore, existing technologies, when outputting detection results, typically focus on directly providing results at a single moment or for a single segment, lacking effective constraints on whether adjacent segments belong to the same continuous change process. This easily leads to situations where adjacent segments belonging to the same continuous change process are fragmented, or adjacent segments not belonging to the same continuous change process are incorrectly continued, resulting in insufficient continuity of detection results or blurred segment boundaries. These problems cannot be solved simply by adjusting fusion parameters, modifying fusion weights, or changing calculation formulas. The root cause lies in the lack of clear, continuous, and effective processing mechanisms for the correspondence between multi-source data, the boundary relationships between different change segments, and the continuity relationships between successive change processes.

[0008] Therefore, existing technologies have at least the following shortcomings: For comprehensive detection methods of vehicle sideslip based on multi-sensor fusion, there is a lack of a processing mechanism capable of accurately determining the correspondence between multi-source data, the boundary relationships between adjacent change segments, and the continuity between preceding and subsequent change processes under complex operating conditions. This leads to problems such as data mismatch, mixing of different change processes, misconnection or omission of adjacent segments, and insufficient continuity of results in existing comprehensive sideslip detection results. Based on this, it is necessary to propose a new comprehensive detection method for vehicle sideslip based on multi-sensor fusion.

[0009] In view of this, this application proposes a comprehensive detection method for motor vehicle sideslip based on multi-sensor fusion to solve the above problems. Summary of the Invention

[0010] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, this application provides the following technical solution: a comprehensive detection method for motor vehicle sideslip based on multi-sensor fusion, comprising: The collected multi-source data is processed into the same segment to obtain the original data of the same segment; Based on the original data in the same segment, indicator data is obtained by extracting signs of change. Based on the indication data, stage data is obtained by dividing the original data in the same segment into stages of change; Based on the original data and stage data of the same segment, responsibility data is obtained through stage adaptation responsibility determination; Based on the original data, stage data and responsibility data of the same segment, the side slip chain data is obtained through the analysis of adjacent stage succession. Based on the slip chain data, stage data, and responsibility data, a comprehensive detection result of the slip volume is obtained through continuous integration.

[0011] Furthermore, methods for processing collected multi-source data into the same segment to obtain the original data of that segment include: Multi-source data includes wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data; Based on the sampling time sequence of each of the multi-source data, determine the overlapping time period covered by the multi-source data; Based on overlapping time periods, the multi-source data is segmented and time-series aligned to obtain the original data for the same segment.

[0012] Furthermore, methods for obtaining identifiable data include: Extract signs of speed change from the wheel rotation speed data in the same segment of raw data; Extract signs of turning changes based on the steering angle data in the same segment of raw data; Based on the yaw rate data in the same segment of raw data, extract signs of continuous directional change; Based on the lateral and longitudinal acceleration data in the same segment of raw data, extract signs of linkage changes; The signs are arranged in chronological order of appearance based on the signs of rapid or slow changes, turning points, continuous changes in direction, and interconnected changes, thus obtaining the sign data.

[0013] Furthermore, methods for obtaining stage data include: Based on the indicator data, determine the start and end times of each change indicator in the same segment of raw data to obtain indicator time data; Based on the indicator time data, the time sequence of each change indicator is merged to determine the start and end times of each change stage, thus obtaining the stage boundary data. Based on the stage boundary data, the original data in the same segment is divided into segments to obtain stage data.

[0014] Furthermore, methods for obtaining responsibility data through phase adaptation responsibility determination include: Based on the original data and stage boundary data of the same segment, extract the response start time and response end time corresponding to the multi-source data in each change stage to obtain the stage response data; Based on the phase response data, the multi-source data with the earliest response start time is determined as the initial responsibility data, the multi-source data with the longest response duration is determined as the continuous responsibility data, and the multi-source data that crosses the boundary of adjacent change phases is determined as the successor responsibility data. The initial responsibility data, continuous responsibility data, and inherited responsibility data are aggregated to obtain responsibility data.

[0015] Furthermore, methods for obtaining side slip chain data through adjacent stage continuity analysis include: Based on the stage data, extract the stage boundaries corresponding to adjacent change stages; Based on the responsibility data, extract the corresponding responsibility data on both sides of the stage boundary for each adjacent change stage; Based on the original data and the data on the responsibilities of each segment, determine the lateral sliding connection relationship at the stage boundary of each adjacent change stage; Based on the sideslip connection relationship between each adjacent change stage, a chain association is performed according to the stage boundaries to obtain sideslip chain data.

[0016] Furthermore, methods for determining the sideslip continuity relationship between adjacent change stages at stage boundaries include: Based on the data of the assumed responsibilities, determine the target multi-source data corresponding to each adjacent change stage before and after the stage boundary; Based on the original data of the same segment, extract the boundary change data corresponding to the front and back of the stage boundary of the target multi-source data; Based on the boundary change data, determine whether the change directions corresponding to the front and back sides of the stage boundary are consistent and whether the change is continuous, and obtain the boundary connection result. Based on the boundary acceptance results, the lateral sliding acceptance relationship at the stage boundary of each adjacent change stage is determined.

[0017] Furthermore, methods for obtaining boundary acceptance results include: Based on the boundary change data, determine the change direction and end time corresponding to the front side of the stage boundary; Based on the boundary change data, determine the change direction and start time corresponding to the back side of the stage boundary; Determine whether the change directions are consistent based on the change direction corresponding to the front side of the stage boundary and the change direction corresponding to the back side of the stage boundary; Determine whether the change is continuous based on the end time corresponding to the front side of the stage boundary and the start time corresponding to the back side of the stage boundary. The boundary connection result is obtained based on whether the direction of change is consistent and whether the change is continuous.

[0018] Furthermore, methods for obtaining comprehensive sideslip detection results through continuous integration include: Based on the stage data, determine the order of each stage of change to obtain stage sequence data; Based on the side slip chain data and stage sequence data, the side slip chain connection relationships corresponding to adjacent change stages in the order of arrangement are extracted to obtain the connection relationship data; Based on the responsibility data and the phase sequence data, the priority takeover phase corresponding to the adjacent change phases in the order of arrangement is determined, and the priority takeover data is obtained. Based on the data on the connection relationships and the data on priority connections, the sideslip changes corresponding to each stage of change are continuously integrated to obtain the comprehensive detection results of the sideslip amount.

[0019] Furthermore, methods for obtaining priority data acceptance include: Based on the stage sequence data, extract the stage boundaries corresponding to each adjacent change stage at the stage boundary. Based on the responsibility data, extract the corresponding responsibility data at the stage boundary of each adjacent change stage; Based on the data on the assumed responsibilities, compare the duration of the assumed responsibilities at the stage boundaries of adjacent change stages. Based on the duration of the acceptance, the change phase with a longer duration is identified as the priority acceptance phase, and priority acceptance data is obtained.

[0020] Compared with existing technologies, the technical effects and advantages of the multi-sensor fusion-based comprehensive detection method for motor vehicle sideslip in this application are as follows: This application obtains original data for each segment by organizing collected multi-source data into segments; extracting change indicators from these segments to obtain indicator data; dividing the original data into change stages based on the indicator data to obtain stage data; determining stage-specific responsibilities based on the original and stage data to obtain responsibility data; performing adjacent stage continuity analysis based on the original, stage, and responsibility data to obtain side slip chain data; and continuously integrating the side slip chain, stage, and responsibility data to obtain a comprehensive side slip measurement result. This allows for the continuous determination of the correspondence between multi-source data, the boundary relationships between different change processes, and the continuity between preceding and subsequent change processes, thereby mitigating the problems of data misalignment, mixed change processes, misconnection or omission of adjacent segments, and insufficient continuity of detection results in existing technologies.

[0021] This application identifies overlapping time periods covered by multiple data sources and performs segment extraction and time alignment within these overlapping time periods. This ensures that wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data form a unified and stable correspondence before entering subsequent processing. This alleviates the data misalignment problem caused by inconsistencies in sampling start time, sampling end time, sampling period, and effective recording duration in existing technologies, and provides a unified data foundation for subsequent analysis of sideslip-related changes.

[0022] This application extracts signs of rapid and slow changes, turning changes, continuous changes in direction, and linked changes based on the original data of the same segment. Based on these signs, it determines the start and end times of each sign, merges the time sequence, and forms change stages. This allows multiple types of changes that cross each other within the original detection interval to be transformed into change stages with clear start and end ranges. This alleviates the problems of unclear boundaries between different change processes and multiple change processes being mixed in the same processing chain in the prior art, and allows the start and end ranges and the connection relationships between sideslip-related changes to be clearly characterized.

[0023] This application extracts the response start time and response end time corresponding to multi-source data in each change stage, and further distinguishes the relevant multi-source data into initial responsibility data, continuous responsibility data, and succession responsibility data. This distinguishes the differences in the roles of different information sources in the change formation stage, change continuation stage, and stage boundary extension stage. This alleviates the problem of unclear dominant basis and mixed roles in the stage caused by different information sources participating in the processing only as inputs at the same level in the prior art. It is conducive to improving the pertinence and hierarchy of the judgment of sideslip-related changes under complex working conditions.

[0024] This application extracts the corresponding responsibility data of adjacent change stages on both sides of the stage boundary, and combines the target multi-source data with the corresponding boundary change data on the front and back sides of the stage boundary to determine whether the change direction is consistent and whether the change is continuous. This allows it to distinguish whether adjacent change stages belong to the same continuous lateral sliding change process at the stage boundary, thereby alleviating the problem in the prior art that it is difficult to determine whether adjacent segments should continue continuously even though they are connected in time. This helps to avoid incorrectly connecting different change processes or incorrectly splitting change processes that should be continuous.

[0025] This application obtains sideslip chain data by chaining the sideslip connection relationships corresponding to each adjacent change stage, and continuously integrates the sideslip changes corresponding to each change stage based on the connection relationship data and priority connection data. This allows the comprehensive sideslip detection results to no longer remain as isolated single-segment results, but to characterize the temporal continuity, stage dominance, and integration status of sideslip-related changes. This alleviates the problems of insufficient continuity of detection results or blurred segment boundaries in existing technologies, and helps to improve the ability of comprehensive sideslip detection results to reflect the continuous change process of complex working conditions.

[0026] In summary, this application does not merely adjust local parameters or single fusion rules in the existing multi-sensor fusion sideslip detection process. Instead, it forms a continuous processing link around multi-source data correspondence, differentiation of change stages, differentiation of stage effects, boundary acceptance judgment, and continuous integrated output. This enables the comprehensive sideslip detection results to more accurately reflect the formation process, continuation process, and integration results of sideslip-related changes under complex working conditions, thereby improving the accuracy, continuity, and scene adaptability of the comprehensive sideslip detection results. Attached Figure Description

[0027] Figure 1 This is a flowchart of a comprehensive detection method for motor vehicle sideslip based on multi-sensor fusion, according to an embodiment of this application. Figure 2 This is a schematic diagram illustrating the division of the multi-source data processing and change stages in an embodiment of this application. Figure 3 This diagram illustrates the stage adaptation responsibility determination, adjacent stage succession, and continuous integration in an embodiment of this application. Detailed Implementation

[0028] The technical solutions of this application will be described in detail, clearly, and completely below with reference to the accompanying drawings of the embodiments. It should be particularly noted that the specific embodiments described below are only used to better illustrate and explain the technical solutions of this application, and are intended to enable those skilled in the art to better understand and implement this application, and should not be construed as limiting the scope of protection of this application. Without departing from the spirit and substance of this application, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in this application, and these modifications, adjustments, or equivalent substitutions should all be considered within the scope of protection of this application.

[0029] Example 1: Please see Figure 1 As shown, this embodiment provides a comprehensive detection method for vehicle sideslip based on multi-sensor fusion, including: Step S1: Organize the collected multi-source data into segments to obtain the original data of the same segment.

[0030] Step S11: Collect multi-source data.

[0031] Collect multi-source data, including wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data.

[0032] The wheel speed data is output by wheel speed sensors installed at each wheel of the vehicle. This data includes the speeds of the left front wheel, right front wheel, left rear wheel, and right rear wheel. Each wheel speed record includes at least the wheel identifier, sampling time, and speed value. Steering angle data is output by a steering wheel angle sensor or steering gear angle sensor. Each steering angle record includes at least the sampling time and steering angle value. Yaw rate data is output by a yaw rate sensor. Each yaw rate record includes at least the sampling time and yaw rate value. Lateral and longitudinal acceleration data are output by acceleration sensors in the inertial measurement unit. Each lateral acceleration record includes at least the sampling time and lateral acceleration value, and each longitudinal acceleration record includes at least the sampling time and longitudinal acceleration value.

[0033] When each sensor outputs its corresponding data value, the vehicle controller reads the current system clock and writes it into the corresponding data record, which serves as the sampling time for that data record. All data records of the same type are arranged from earliest to latest according to their sampling time, forming a corresponding sampling sequence.

[0034] Step S12: Determine the overlapping time period covered by the multi-source data based on the sampling time sequence corresponding to each of the multi-source data.

[0035] The sampling start and end times of the wheel speed sampling time sequence, steering angle sampling time sequence, yaw rate sampling time sequence, lateral acceleration sampling time sequence, and longitudinal acceleration sampling time sequence are read respectively. The latest sampling start time among the five types of sampling time sequences is determined as the common start time, and the earliest sampling end time among the five types of sampling time sequences is determined as the common end time. The time interval between the common start time and the common end time is used as the initial overlapping period.

[0036] Continuity checks were performed on the five types of sampling time sequences within the initial overlapping period. First, the nominal sampling period for each of the five types of sampling time sequences was calculated, and the maximum value was determined as the base sampling period. To determine the criteria for identifying consecutive missing sampling sections, calibration samples were pre-collected. These calibration samples included samples of stable straight-line driving, left-turn, right-turn, braking, acceleration / steering, deceleration / steering, and constant-speed steering. Each calibration sample was recorded synchronously from multiple sources using a unified system clock. Samples involved in determining consecutive missing sampling were manually marked. The manual marking method was as follows: multiple technicians independently marked the samples based on the original reported logs, the synchronization curve under the unified clock, and the corresponding vehicle video playback. When the marking results were consistent, the marked result was directly adopted; when the marking results were inconsistent, a review technician verified the results, and the verification result was used as the final marking result.

[0037] Normal continuous sampling is defined as follows: within the observation window, although there is natural jitter between adjacent sampling times of the same data source, it does not cause the data source to lose its continuous alignment capability within the observation window. No broken sampling is defined as follows: there is a single or small sampling delay, but continuous valid records are retained before and after the delay, and this delay does not cause the current data source to lose its continuous overlap with other data sources within the observation window. True continuous missing sampling is defined as: a data source experiences a significant sampling interruption within the observation window, and this interruption breaks the continuous overlap between the current data source and other data sources.

[0038] The distribution of the time interval between two consecutive adjacent sampling times relative to the basic sampling period was statistically analyzed for each of the three types of samples to determine the upper bound of the time interval under normal continuous sampling conditions and the maximum tolerable upper bound under conditions of no sampling interruption. The method for setting the preset period stratification threshold is as follows: the time interval multiples corresponding to normal continuous sampling samples, the time interval multiples corresponding to samples without sampling interruption, and the time interval multiples corresponding to true continuous missing sampling samples are merged and sorted. The midpoint of adjacent and unequal multiples is taken to form candidate period stratification thresholds. Period level rounding is performed on each candidate period stratification threshold, and the rounding result is used for continuous missing sampling segment identification. The number of false positives and false negatives corresponding to each candidate period stratification threshold are then counted to calculate the total number of misjudgments. The candidate period stratification threshold with a small total number of misjudgments and a small change in the rounding result after the addition of new samples is determined as the preset period stratification threshold.

[0039] The larger of the upper bound of the time interval under normal continuous sampling conditions and the upper bound of the maximum tolerance under conditions of no sampling interruption is rounded upwards to the period level corresponding to the nearest preset period stratification threshold, and determined as the continuous sampling omission judgment multiple. Then, when the time interval between two consecutive adjacent sampling times is greater than the base sampling period corresponding to the continuous sampling omission judgment multiple, this time interval is determined as the continuous sampling omission segment. If the calculation result is exactly in the middle of two period levels, the larger period level is taken as the rounding result.

[0040] When any type of data has a continuous missing mining segment within the initial overlapping period, the initial overlapping period is divided into multiple candidate continuous overlapping sub-periods according to the continuous missing mining segment, and the duration of each candidate continuous overlapping sub-period is calculated; the candidate continuous overlapping sub-period with the longest duration is determined as the final overlapping period; when none of the five types of data have a continuous missing mining segment within the initial overlapping period, the initial overlapping period is directly determined as the final overlapping period.

[0041] A minimum time period threshold is preset. The method for setting the minimum time period threshold is as follows: In the calibration sample, extract the actual dynamic change sample segments corresponding to wheel speed, steering angle, yaw rate, lateral acceleration, and longitudinal acceleration, respectively, and calculate the shortest continuous sampling time required for each type of data from the start of change to stably show effective change; then take the maximum value among the shortest continuous sampling times corresponding to the five types of data as the minimum time period threshold. When the common start time is earlier than the common end time, and the duration of the final overlapping period is not less than the minimum time period threshold, it is determined that there is a valid overlapping period in the multi-source data collected this time; when the common start time is later than or equal to the common end time, or the duration of the final overlapping period is less than the minimum time period threshold, it is determined that there is no valid overlapping period in the multi-source data collected this time, the original data of the same segment of this wheel is not generated, and the comprehensive detection process of the sideslip of this wheel ends.

[0042] Step S13: Based on the overlapping time periods, the multi-source data is segmented and time-series aligned to obtain the original data of the same segment.

[0043] Based on the final overlapping time period, the wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data are each extracted in the same segment, and only the data records whose sampling time is within the final overlapping time period are retained.

[0044] After segmenting the data into segments, time-series alignment is performed on the segmented data. The method for setting the unified alignment time interval is as follows: the nominal sampling period of each of the five data categories is read, and the largest nominal sampling period is determined as the unified alignment time interval. The start time of the final overlapping period is used as the alignment starting point, and a standard time series is generated according to the unified alignment time interval.

[0045] For each alignment point in the standard time series, the corresponding data value is extracted from the five categories of truncated data. When an actual sample value exists for a certain category of data at the alignment point, that actual sample value is directly determined as the data value corresponding to that alignment point; when no actual sample value exists, the nearest sample value before and after the alignment point is retrieved, and the corresponding time difference is calculated.

[0046] A maximum permissible time deviation threshold is preset. The method for setting the maximum permissible time deviation threshold is as follows: In the calibration sample, multi-source data that has been synchronized and recorded under a unified system clock is extracted. Substitution tests are performed on sampling points that can be used as mapping candidate values ​​before and after the same real time. The time offset between the candidate sampling points and the standard time point is gradually increased, and it is observed whether the substitution causes a change in the direction of change, a change amplitude that crosses the corresponding change threshold, or a drift at the stage boundary. The maximum time offset that has not caused the above changes is determined as the maximum permissible time deviation threshold.

[0047] When there is a sample value between the most recent sample value on the front side and the most recent sample value on the back side with a time difference not greater than the maximum allowable time deviation threshold, the sample value with the smaller time difference is mapped to the data value corresponding to the alignment time. When the time differences between the most recent sample value on the front side and the most recent sample value on the back side are equal and not greater than the maximum allowable time deviation threshold, the sample value with the earlier time is mapped to the data value corresponding to the alignment time. When the time difference between the most recent sample value on the front side and the alignment time is greater than the maximum allowable time deviation threshold, it is determined that this type of data has no valid mapping value at the alignment time.

[0048] For each alignment time, check whether all five data categories have valid mapping values. If any data category has no valid mapping value, delete all data groups corresponding to that alignment time. If all five data categories have valid mapping values, combine the alignment time, left front wheel speed, right front wheel speed, left rear wheel speed, right rear wheel speed, steering angle, yaw rate, lateral acceleration, and longitudinal acceleration into a single valid data group. Arrange all valid data groups in ascending order of alignment time to generate the original data for the same segment.

[0049] Step S2: Based on the original data of the same segment, extract the signs of change to obtain the sign data.

[0050] For real samples, non-real samples, samples to be merged, samples not to be merged, synchronization start samples, samples with continuous effect approaching, samples with accepted results, real non-accepted samples, and invalid samples used for threshold setting and rule setting in steps S2 to S6, a unified manual labeling process is performed. The manual labeling method is as follows: multiple technicians independently label samples based on the multi-source synchronization curve under a unified clock, the actual vehicle video playback, and the driving operation record. When the labeling results are consistent, the labeling result is directly adopted; when the labeling results are inconsistent, the review technicians review them, and the review result is used as the final labeling result. When the judgment conclusions corresponding to the multi-source synchronization curve, the actual vehicle video playback, and the driving operation record conflict, the judgment conclusion corresponding to the multi-source synchronization curve is given priority; if the multi-source synchronization curve is insufficient to make a direct judgment, the judgment conclusion corresponding to the driving operation record is adopted; if the driving operation record is still insufficient to make a judgment, the judgment conclusion corresponding to the actual vehicle video playback is adopted; if the review technicians still cannot form a clear judgment according to the above order, the sample is determined as an invalid sample and will not enter the corresponding threshold setting process.

[0051] The following definitions apply to sample changes: A true speed change sample is defined as follows: At least one wheel or all four wheels within a sample segment exhibit a continuous change in the same direction, and the number of continuous alignment points in the continuous change segment is not less than the minimum continuous alignment point threshold. A true acceleration sample is defined as follows: At least one wheel or all four wheels within a sample segment exhibits a continuous positive change in average speed, and the number of continuous alignment points in the continuous change segment is not less than the minimum continuous alignment point threshold. A true deceleration sample is defined as follows: At least one wheel or all four wheels within a sample segment exhibits a continuous negative change in average speed, and the number of continuous alignment points in the continuous change segment is not less than the minimum continuous alignment point threshold. A true non-speed change sample is defined as follows: Wheel speed changes within a sample segment fluctuate around a local baseline, and there are no continuous change segments in the same direction that satisfy the aforementioned conditions. A true turning action sample is defined as follows: The direction of steering angle change within a sample segment switches from increasing to decreasing or from decreasing to increasing, and the number of continuous alignment points in the continuous segments before and after the switch is not less than the minimum continuous alignment point threshold. A true continuous yaw sample is defined as follows: the yaw angular velocity sign within the sample segment remains in the same non-zero direction, and the number of continuous alignment points in the continuous segment is not less than the minimum continuous alignment point threshold. A true linked change sample is defined as follows: the representative moments of the lateral acceleration change segment and the longitudinal acceleration change segment within the sample segment both fall within the same control window, and both correspond to the start and end time ranges defined by the driving control record for that control window. The control window is defined as follows: the start time of the same steering, acceleration, or braking operation in the driving control record is used as the window start time, and the end time of that operation is used as the window end time; when the same steering operation and the same acceleration or braking operation overlap in time, the earliest start time is used as the window start time, and the latest end time is used as the window end time, and the steering operation record and the acceleration or braking operation record are combined to form the driving control record for that control window. A true effective change segment is defined as follows: the corresponding change phenomenon within the sample segment has reached the continuous point requirement specified in this step, and the change direction and change amplitude can be stably identified. Samples that should be merged are defined as: two or more indicator time records originating from the same continuous vehicle dynamic change process. Samples that should not be merged are defined as: two or more indicator time records originating from different vehicle dynamic change processes. Synchronous initiation samples are defined as: two or more response sources forming an initiating response to the same change process within the same change phase, and the only difference between their response initiation times is normal alignment. Samples with sustained close interaction are defined as: two or more response sources within the same change phase whose response duration differences are insufficient to distinguish primary and secondary responses. Samples with valid continuation are defined as: two adjacent change phases belonging to the same continuous sideslip change process at the phase boundary. Samples without true continuation are defined as: two adjacent change phases, although temporally adjacent, whose changes on either side of the boundary no longer belong to the same continuous sideslip change process.

[0052] In this embodiment, whenever determining the target value position according to a preset position ratio threshold, the corresponding samples are first sorted from smallest to largest, then the total number of samples is multiplied by the preset position ratio threshold, and the resulting integer is used as the target value position. Whenever sample removal is performed according to a preset sample removal ratio threshold, the total number of samples is first multiplied by the preset sample removal ratio threshold, and the resulting integer is used as the sample removal quantity. When the integer is zero, no sample removal is performed, and the sorted boundary value is used directly. The preset position ratio threshold is set as follows: for the target sample set corresponding to the current step, the cumulative sample ratio corresponding to each possible position in the sorted samples is used as a candidate position ratio; for each candidate position ratio, the corresponding target value is calculated, and the target value is used for the validity determination of the current step; then the number of false validitys and false invalidities corresponding to each candidate position ratio is counted, and the total number of false judgments is calculated; the candidate position ratio with a small total number of false judgments and whose corresponding value fluctuates little after the new sample enters is determined as the preset position ratio threshold. The method for setting the preset sample elimination ratio threshold is as follows: For the target sample set corresponding to the current step, the cumulative percentage of eliminated samples corresponding to the samples eliminated one by one from the edge inwards in the sorted samples is used as the candidate sample elimination ratio; for each candidate sample elimination ratio, sample elimination is performed and the parameters required for the current step are recalculated; then the recalculated parameters are used for the validity determination of the current step, the number of falsely valid and falsely invalid results are counted, the total number of misjudgments is calculated, and the change range of this parameter under the condition of continuously added samples is counted; the candidate sample elimination ratio with a small total number of misjudgments and a small change range of the parameter is determined as the preset sample elimination ratio threshold. Subsequent actions involving the determination of target value positions and sample elimination shall be performed according to the above rules.

[0053] In steps S21 to S24, whenever a separation interval exists between the upper bound of the background fluctuation and the lower bound of the minimum effective change, a unified target sub-interval determination rule is adopted. The unified target sub-interval determination rule is as follows: First, the separation interval is divided into multiple continuous sub-intervals along the amplitude direction according to a preset segmentation threshold; then, the upper boundary of each sub-interval is used as a candidate threshold, which is used for the validity determination of the current step; subsequently, the number of false validitys and false invalidities corresponding to each candidate threshold are counted, and the total number of false judgments is calculated; the sub-interval with a small number of false judgments and strong ability to suppress background fluctuations is determined as the target sub-interval position, and the upper boundary of the corresponding sub-interval is taken as the threshold of the current step. The method for setting the preset segment number threshold is as follows: All feasible segments that can divide the separation interval into multiple continuous sub-intervals are taken as the candidate segment number. Separation interval division is performed for each candidate segment number, and the upper boundary of the target sub-interval corresponding to different candidate segment numbers is used for the validity determination of the current step. Then, the number of falsely valid and falsely invalid results corresponding to each candidate segment number are counted, and the total number of false positives is calculated. The candidate segment number with a small total number of false positives and whose corresponding threshold fluctuates little after new samples are added is determined as the preset segment number threshold. Subsequent steps S22 to S24, involving separation interval division and target sub-interval extraction, all follow the same unified target sub-interval determination rule.

[0054] Please see Figure 2 As shown, in step S21, the wheel speed data in the same segment of original data is used to extract signs of speed change.

[0055] Read the speed values ​​of the left front wheel, right front wheel, left rear wheel, and right rear wheel from morning to night according to the alignment time. Use the wheel speed difference between two adjacent alignment times to represent the speed change at the current alignment time, and obtain the speed change sequence of each of the four wheels.

[0056] The system presets thresholds for wheel speed variation amplitude and minimum number of continuous alignment points. The method for setting the wheel speed variation amplitude threshold is as follows: First, extract a background sample segment of straight-line constant-speed driving from the calibration sample. This segment is defined as follows: In the sample segment, the vehicle trajectory remains straight as shown in the real-vehicle video playback; no new start or end records are found for steering, braking, and acceleration in the driving operation record; the average speed of the four wheels does not exhibit a continuous unidirectional change segment in the multi-source synchronization curve; and the steering angle does not exhibit a direction switching process that meets the conditions for a turning change. Within this straight-line constant-speed driving background sample segment, the upper bound of the background fluctuation of the absolute value of speed change between adjacent alignment moments of the four wheels is calculated. Then, real acceleration and deceleration samples are extracted, and the speed change within the corresponding effective change segment is calculated. The minimum effective lower bound of absolute value change is determined as follows: When there is a separation interval between the upper bound of background fluctuation and the minimum effective lower bound of change, the aforementioned unified target sub-interval determination rule is used to obtain the threshold for wheel speed change amplitude; when there is overlap between the two, the average value and standard deviation of the absolute value of speed change in the background fluctuation sample are statistically analyzed, and multiple candidate multiple thresholds are pre-set. Each candidate threshold is formed by adding the average value to the standard deviation of the corresponding candidate multiple threshold. The total number of misjudgments of each candidate threshold on the real speed change sample and the real non-speed change sample is compared, and the candidate threshold with the smaller number of misjudgments is taken as the threshold for wheel speed change amplitude. The method for setting the candidate multiple threshold is as follows: different amplification factors of the standard deviation of the absolute value of change in the background fluctuation sample are used as candidate multiples, and the target candidate multiple threshold is determined according to the principle of a smaller total number of misjudgments and higher threshold stability. The method for setting the minimum sustained alignment point threshold is as follows: the shortest effective wheel response segment is extracted from the calibration sample, the duration covered by the shortest wheel response segment from start to end is calculated, and the duration is divided by the unified alignment time interval and rounded up to determine the minimum sustained alignment point threshold.

[0057] For each of the four wheel speed change sequences, a speed change determination is performed. When the speed change of a certain wheel exceeds a certain threshold for speed change amplitude for multiple consecutive alignment moments, the continuous segment is identified as a candidate segment for acceleration. When the speed change of a certain wheel is less than the negative of the threshold for speed change amplitude for multiple consecutive alignment moments, the continuous segment is identified as a candidate segment for deceleration. The number of consecutive alignment points is counted for each candidate segment. If the number of consecutive alignment points is not less than the minimum continuous alignment point threshold, the candidate segment is determined as a valid speed change segment; otherwise, it is discarded.

[0058] Speed ​​change indicators are generated for each valid speed change segment. Each speed change indicator includes at least the indicator type, source wheel identification, indicator direction, representative time, and indicator intensity. The representative time is the aligned time of the largest absolute value of speed change within the corresponding valid speed change segment. The indicator intensity is the average of the absolute values ​​of speed change within that valid speed change segment.

[0059] When multiple wheels exhibit valid speed change segments simultaneously near the same alignment time, the speed change signs corresponding to these multiple wheels are merged and generated. A preset speed change merging time threshold is established. The method for setting the speed change merging time threshold is as follows: extract the valid speed change segments corresponding to the four wheels during the same vehicle dynamic change process from the calibration sample, and statistically analyze the time difference between the representative times of the signs corresponding to different wheels. The method for determining abnormal time differences is as follows: for the representative times of the signs corresponding to the four wheels during the same vehicle dynamic change process, first calculate the time difference between the representative time of any wheel's sign and the representative times of the signs of the other wheels; when the time difference corresponding to a certain wheel meets the preset abnormal difference multiple threshold condition, and the direction of the sign of the speed change segment corresponding to that wheel is inconsistent with the common direction of the majority of other wheels, the time difference corresponding to that wheel is determined as an abnormal time difference and removed. The method for setting the preset abnormal difference multiple threshold is as follows: statistically analyze the time difference multiple distribution corresponding to abnormal time difference samples and normal time difference samples in the calibration sample, and then compare the total number of misjudgments by the candidate boundary value to determine the preset abnormal difference multiple threshold. After removing abnormal time differences, the maximum value among the remaining time differences is determined as the merging time threshold for speed changes. When the time difference between the moments represented by speed change indicators corresponding to multiple wheels is not greater than the merging time threshold for speed changes, and the indicators have the same direction or belong to the same speed change category, the speed change indicators corresponding to multiple wheels are merged into a single combined speed change indicator.

[0060] Step S22: Extract signs of turning changes based on the steering angle data in the original data of the same segment.

[0061] The steering angle values ​​are read from early to late according to the alignment time, and the steering angle difference between two adjacent alignment times is calculated to obtain the steering angle change sequence. The steering angle change sequence is then divided into directions according to the direction of steering angle change between adjacent alignment times.

[0062] The system presets thresholds for turning change amplitude, turning direction switching, and minimum continuous alignment point count. The method for setting the turning change amplitude threshold is as follows: First, extract a background sample segment for steering fine-tuning from the calibration sample. This background sample segment is defined as follows: no new steering operation start / end records are found in the driving operation record within the sample segment; the vehicle trajectory playback shows no continuous steering process; although there are local small fluctuations in the steering angle in the multi-source synchronization curve, no direction switching process meets the conditions for forming a turning change sign. In the steering fine-tuning background sample segment, the upper bound of the background oscillation of the absolute value of the steering angle change is statistically analyzed. Then, a sample segment of the actual turning action is extracted, and the lower bound of the minimum effective change corresponding to the weaker side of the continuous change segment before and after the turning action is statistically analyzed. When there is a separation interval between the upper bound of the background oscillation and the lower bound of the minimum effective change, the aforementioned unified target sub-interval determination rule is used to obtain the turning change amplitude threshold. When there is an overlap between the two, the average value and standard deviation of the absolute value of the steering angle change in the background oscillation sample are statistically analyzed. Multiple candidate multiple thresholds are preset, and candidate thresholds are formed and compared with the total number of misjudgments to determine the turning change amplitude threshold. The method for setting the candidate multiple threshold follows the method used in step S21. The method for setting the turning direction switching threshold is as follows: extract real turning action sample segments from the calibration samples; for each turning action sample, identify the continuous change segment before and after the switching; and take the smaller of the absolute values ​​of their average changes as the weaker side change level corresponding to that turning action sample; then sort all the weaker side change levels in ascending order, and take the smallest value that can completely cover the lower bound of the weaker side change level in the real turning action as the turning direction switching threshold. The minimum continuous alignment point number threshold follows the setting result in step S21.

[0063] When the number of continuous alignment points in a continuously increasing steering angle segment is not less than the minimum continuous alignment point threshold, and a continuously decreasing segment with a number of continuous alignment points not less than the minimum continuous alignment point threshold appears after the continuously increasing segment, and the direction switching amplitude between the two segments is not less than the turning direction switching threshold, the change process from increasing to decreasing is determined as a turning change sign; when a continuously increasing segment that meets the same conditions appears after a continuously decreasing steering angle segment, the change process from decreasing to increasing is determined as another turning change sign; otherwise, no turning change sign is generated.

[0064] Each indicator of a turning point includes at least the indicator type, indicator direction, indicator representative time, and indicator strength. The indicator representative time is the alignment time corresponding to the direction change. The indicator strength is the sum of the average absolute values ​​of the changes in the two consecutive segments before and after the direction change.

[0065] Step S23: Extract signs of continuous change in direction based on the yaw rate data in the original data of the same segment.

[0066] Read the yaw rate values ​​from morning to night according to the alignment time, and calculate the difference in yaw rate between two adjacent alignment times to obtain the yaw rate change sequence. Based on the sign of the yaw rate values ​​and the direction of change in the yaw rate change sequence, determine the direction of yaw motion and the trend of yaw change.

[0067] The system presets a threshold for continuous yaw change amplitude, a threshold for directional consistency, and a threshold for the minimum number of continuous alignment points. The method for setting the threshold for continuous yaw change amplitude is as follows: First, extract a stable yaw background sample segment from the calibration samples. This stable yaw background sample segment is defined as follows: within the sample segment, the actual vehicle video playback does not show a continuous turning process, there are no new continuous steering operation start and end records in the driving operation record, and the yaw angular velocity in the multi-source synchronization curve does not form a continuous segment with a single non-zero direction. In the stable yaw background sample segment, calculate the upper bound of the background fluctuation of the absolute value of the yaw angular velocity change. Then, extract sample segments where a real continuous yaw process exists, and calculate the lower bound of the minimum effective change of the absolute value of the yaw angular velocity change within the effective yaw change segment. When there is a separation interval between the upper bound of the background fluctuation and the lower bound of the minimum effective change, the aforementioned unified target sub-interval determination rule is used to obtain the threshold for continuous yaw change amplitude. When there is an overlap between the two, compare the total number of misjudgments corresponding to the candidate thresholds formed by the candidate multiple thresholds to determine the threshold for continuous yaw change amplitude. The method for setting the candidate multiple threshold follows the method used in step S21. The method for setting the direction consistency threshold is as follows: extract real continuous yaw sample segments from the calibration samples, and calculate the proportion of the number of times the yaw angular velocity symbol with the same direction as the main direction appears in each sample segment relative to the total number of effective alignment points in that sample segment, thus obtaining the direction consistency ratio; then sort the direction consistency ratios corresponding to the real continuous yaw samples from smallest to largest, remove low-end samples according to the preset sample removal ratio threshold, and take the minimum direction consistency ratio among the remaining samples as the direction consistency threshold. The minimum continuous alignment point number threshold follows the setting result in step S21.

[0068] If, within a certain continuous segment, the absolute value of the yaw rate is greater than the yaw rate amplitude threshold for multiple consecutive alignment moments, and the sign of the yaw rate in the continuous segment meets the direction consistency threshold requirement, and the number of continuous alignment points is not less than the minimum number of continuous alignment points threshold, then the continuous segment is determined as a valid direction continuous change segment; otherwise, no direction continuous change indication is generated.

[0069] Each sign of continuous directional change includes at least the sign type, direction marker, sign representative time, and sign intensity. The sign representative time is the alignment time with the largest absolute value of yaw rate within the valid continuous directional change segment. The sign intensity is the average of the absolute values ​​of yaw rate within the valid continuous directional change segment.

[0070] Step S24: Extract signs of linkage change based on the lateral acceleration data and longitudinal acceleration data in the original data of the same segment.

[0071] Read the lateral acceleration values ​​and longitudinal acceleration values ​​from early to late according to the alignment time. Calculate the difference in lateral acceleration and longitudinal acceleration between two adjacent alignment times to obtain the lateral acceleration change sequence and the longitudinal acceleration change sequence. Then, pair the lateral acceleration change and longitudinal acceleration change at the same alignment time.

[0072] The system presets thresholds for lateral change amplitude, longitudinal change amplitude, linkage time matching, and linkage weight. The method for setting the lateral change amplitude threshold is as follows: In the calibration samples, first extract sample segments with no obvious steering and no obvious lateral load transfer. Specifically, "no obvious steering" is defined as the absolute value of the steering angle within the sample segment never exceeding the turning change amplitude threshold, and the absolute value of the steering angle change between adjacent alignment moments never exceeding the turning change amplitude threshold; "no obvious lateral load transfer" is defined as the absolute value of the yaw rate within the sample segment never exceeding the continuous yaw change amplitude threshold, and at the same alignment moment, the absolute values ​​of the speed difference between the left and right front wheels and the absolute values ​​of the speed difference between the left and right rear wheels are both no greater than the average speed of the two wheels corresponding to the preset wheel difference ratio threshold. The method for setting the preset wheel difference ratio threshold is as follows: In the calibration samples, extract samples with no obvious lateral load transfer and samples with obvious lateral load transfer; calculate the ratio of the absolute value of the difference in speed between the left and right wheels to the average speed of the corresponding two wheels in each sample to obtain the wheel difference ratio samples; then merge and sort the two types of wheel difference ratio samples, and construct candidate boundary values ​​for adjacent ratio values; for each candidate boundary value, count the number of false positives and false negatives, and calculate the total number of misjudgments; determine the candidate boundary value with the smaller total number of misjudgments as the preset wheel difference ratio threshold. In the above sample segment, calculate the upper bound of the background fluctuation of the absolute value of lateral acceleration change; then extract the real linkage change sample segment, and calculate the lower bound of the minimum effective change of the absolute value of lateral acceleration change; when there is a separation interval between the upper bound of the background fluctuation and the lower bound of the minimum effective change, use the aforementioned unified target sub-interval determination rule to obtain the lateral change amplitude threshold; when there is an overlap between the two, compare the total number of misjudgments by the candidate multiple threshold to determine the lateral change amplitude threshold. The method for setting the candidate multiple threshold follows the method for setting the candidate multiple threshold in step S21.

[0073] The method for setting the longitudinal change amplitude threshold is as follows: In the calibration samples, first extract sample segments with no obvious acceleration and no obvious braking. Here, "no obvious acceleration and no obvious braking" is defined as follows: The absolute value of the change in the average speed of the four wheels within the sample segment between adjacent alignment moments is always no greater than the wheel speed change amplitude threshold, and the absolute value of the difference in the average speed of the four wheels between the start and end points of the sample segment is no greater than the average speed of the four wheels within the sample segment corresponding to the preset average speed deviation ratio threshold. The method for setting the preset average speed deviation ratio threshold is as follows: In the calibration samples, extract samples with no obvious acceleration and no obvious braking and samples with obvious acceleration or obvious braking; calculate the ratio of the absolute value of the difference in the average speed of the four wheels between the start and end points of the sample segment to the average speed of the four wheels within the sample segment, obtaining the average speed deviation ratio sample; then merge and sort the two types of average speed deviation ratio samples, constructing candidate boundary values ​​for adjacent ratio values; for each candidate boundary value, count the number of false positives and false negatives, calculating the total number of misjudgments; determine the candidate boundary value with the smaller total number of misjudgments as the preset average speed deviation ratio threshold. In the aforementioned sample segment, the upper bound of the background fluctuation of the absolute value of longitudinal acceleration change is statistically determined; then, the sample segment of actual linkage change is extracted, and the lower bound of the minimum effective change of the absolute value of longitudinal acceleration change is statistically determined; when there is a separation interval between the upper bound of the background fluctuation and the lower bound of the minimum effective change, the aforementioned unified target sub-interval determination rule is used to obtain the longitudinal change amplitude threshold; when there is an overlap between the two, the longitudinal change amplitude threshold is determined by comparing the total number of misjudgments using candidate multiple thresholds. The method for setting the candidate multiple threshold follows the method used in step S21.

[0074] The method for setting the linkage time matching threshold is as follows: extract real linkage change sample segments from the calibration samples, calculate the time difference between the representative time of the lateral acceleration change segment and the representative time of the longitudinal acceleration change segment, sort them from smallest to largest, remove high-end time difference samples according to the preset sample removal ratio threshold, and take the maximum value of the remaining time difference as the linkage time matching threshold.

[0075] The method for setting the linkage weights is as follows: In real linkage samples, the correlation strength between lateral acceleration changes and lateral sway changes, and the correlation strength between longitudinal acceleration changes and lateral sway changes are calculated separately; the stronger the correlation between lateral acceleration changes and lateral sway changes, the greater the weight of lateral change; the stronger the correlation between longitudinal acceleration changes and lateral sway changes, the greater the weight of longitudinal change; when the correlation strengths are equal, the weights of lateral change and longitudinal change are set to the same weight.

[0076] To obtain specific weight values, the absolute value sequences of lateral acceleration change, longitudinal acceleration change, and yaw rate change are extracted separately and mapped one-to-one according to alignment times. First, the average values ​​of each alignment time in the lateral acceleration change absolute value sequence are calculated to obtain the average lateral change value; then, the average values ​​of each alignment time in the longitudinal acceleration change absolute value sequence are calculated to obtain the average longitudinal change value; finally, the average values ​​of each alignment time in the yaw rate absolute value sequence are calculated to obtain the average yaw rate change value. Subsequently, for each alignment time value in the lateral acceleration change absolute value sequence, the average lateral change value is subtracted to obtain the lateral shift value; for each alignment time value in the longitudinal acceleration change absolute value sequence, the average longitudinal change value is subtracted to obtain the longitudinal shift value; and for each alignment time value in the yaw rate absolute value sequence, the average yaw rate change value is subtracted to obtain the yaw shift value. Next, the lateral shift offset value is multiplied moment-by-moment by the yaw shift offset value corresponding to the same alignment moment, and the average of all products is calculated to obtain the lateral correlation value. Similarly, the longitudinal shift offset value is multiplied moment-by-moment by the yaw shift offset value corresponding to the same alignment moment, and the average of all products is calculated to obtain the longitudinal correlation value. The absolute values ​​of the lateral and longitudinal correlation values ​​are then taken to obtain the basic weights of the lateral and longitudinal shifts. When the sum of the basic weights of the lateral and longitudinal shifts is greater than zero, the basic weight of the lateral shift is divided by the sum of the two to obtain the lateral shift weight; the basic weight of the longitudinal shift is divided by the sum of the two to obtain the longitudinal shift weight. When the sum of the basic weights of the lateral and longitudinal shifts is equal to zero, both the lateral and longitudinal shift weights are set to the same weight value.

[0077] When the absolute value of the lateral acceleration change within a certain lateral acceleration change segment exceeds a lateral change amplitude threshold for multiple consecutive alignment moments, and there exists a longitudinal change segment within the linkage time matching threshold range before and after the representative moment of that segment where the absolute value of the longitudinal acceleration change exceeds a longitudinal change amplitude threshold for multiple consecutive alignment moments, the lateral change segment and the longitudinal change segment are identified as a group of linkage change candidate segments. When the number of continuous alignment points in both the lateral and longitudinal change segments is not less than the minimum continuous alignment point number threshold, the linkage change candidate segment is identified as a valid linkage change segment.

[0078] For each group of valid linked change segments, a linked change indicator is generated. Each linked change indicator includes at least the indicator type, the indicator's representative time, and the indicator intensity. The indicator intensity is calculated as follows: first, calculate the average absolute value of the lateral acceleration change within the lateral change segment, then calculate the average absolute value of the longitudinal acceleration change within the longitudinal change segment; then multiply by the lateral change weight and the longitudinal change weight respectively; finally, add the two together to obtain the indicator intensity of the linked change indicator.

[0079] Step S25: Arrange the signs according to their order of appearance based on the signs of change in speed, signs of change in turning point, signs of continuous change in direction, and signs of coordinated change, to obtain the sign data.

[0080] The signs of rapid and slow changes obtained in step S21, the signs of turning changes obtained in step S22, the signs of continuous directional changes obtained in step S23, and the signs of coordinated changes obtained in step S24 are summarized to form a candidate sign set. The representative time of each change sign in the candidate sign set is read and sorted from earliest to latest according to the representative time.

[0081] A preset parallel ranking threshold is established. The method for setting the parallel ranking threshold is as follows: In the calibration sample, the time differences between the representative moments of different change indicators that are manually determined to occur synchronously are statistically analyzed, and then sorted from smallest to largest. The time difference corresponding to a preset rank ratio threshold is used as the parallel ranking threshold. When two or more change indicators represent the same moment, or when the time difference between the representative moments is not greater than the parallel ranking threshold, they are first sorted from largest to smallest in intensity. When the intensity is the same, they are then sorted by indicator type, with the sorting order being: fast / slow change indicators, turning point change indicators, continuous directional change indicators, and linked change indicators.

[0082] All sorted change indicators are sequentially numbered to generate an indicator sequence, which is then identified as indicator data. By uniformly converting change phenomena from different sources into indicator data arranged in chronological order, subsequent change phase divisions are established on a unified indicator layer rather than the original chaotic data layer, thus providing a consistent entry point for the temporal classification of changes from different sources.

[0083] Step S3: Based on the indication data, the original data in the same segment is divided into stages of change to obtain stage data.

[0084] Please see Figure 3 As shown, in step S31, based on the sign data, the start and end times of each change sign in the same segment of original data are determined to obtain the sign time data.

[0085] Read the sign data. Using the sign representative time of each change sign as the center, locate the corresponding alignment time sequence within the same segment of raw data, and determine the start and end times for different sign types. The start and end boundaries of fast / slow change signs, turning point change signs, continuous direction change signs, and linked change signs are obtained by backtracking and searching backward according to the continuous valid conditions established in steps S21 to S24, respectively. For turning point change signs, if the number of alignment points continuously satisfying the conditions before or after the sign representative time is less than the minimum continuous alignment point threshold in step S22, the boundary is no longer extended to the side that is below the threshold; instead, only the continuous valid segment that participated in forming the turning point change sign in step S22 is used as the boundary range for that side.

[0086] A preset tracing threshold for the symptom boundary is established. The method for setting the tracing threshold is as follows: Extract truly valid change segments from the calibration samples. Statistically calculate the longest duration during which consecutive boundary conditions are not met due to local noise, instantaneous fluctuations, or short-term recovery when extending outwards from the inside of the change segment, but subsequently return to valid change conditions. This longest duration is then determined as the symptom boundary tracing threshold. When tracing forward or backward, if the duration of consecutive boundary conditions not met for the corresponding symptom type exceeds the tracing threshold, the tracing or search stops, and the alignment time of the last time the boundary conditions were met before stopping is taken as the start or end time of that symptom.

[0087] The sign number, sign type, sign representative time, start time, end time, and sign intensity are combined to form a sign time record; all sign time records are arranged from earliest to latest according to the sign representative time to generate sign time data.

[0088] Step S32: Based on the indicator time data, perform time-series merging of each change indicator to determine the start and end times corresponding to each change stage, and obtain stage boundary data.

[0089] Read the sign time data and reorder them from earliest to latest according to the start time in each sign time record. Take the first sign time record after sorting as the initial sign record of the current merging phase, and determine its start time as the candidate start time of the current merging phase, and its end time as the candidate end time of the current merging phase.

[0090] The system presets a stage merging time interval threshold and a stage overlap ratio threshold. The stage merging time interval threshold is set as follows: Multiple sign time records from different sensors belonging to the same continuous vehicle dynamic change process are extracted from the calibration sample. The idle time interval between adjacent sign time records is calculated, and high-frequency time interval samples are removed according to a preset sample removal ratio threshold. The maximum value among the remaining time intervals is taken as the stage merging time interval threshold. The stage overlap ratio threshold is set as follows: Sign time record pairs manually marked as belonging to the same change stage are extracted from the calibration sample. The proportion of the time overlap interval length between each sign time record pair to the shorter duration segment is calculated. The minimum overlap ratio that can stably occur in the samples to be merged is then taken as the stage overlap ratio threshold.

[0091] For the next sign time record following the current merging phase, a phase merging determination is performed. If the time interval between the start time of the next sign time record and the candidate end time of the current merging phase is less than or equal to the phase merging time interval threshold, it is determined that the time is adjacent to the current merging phase, and merging is performed; or, if there is a time overlap ratio of not less than the phase overlap ratio threshold between the next sign time record and any merged sign time record in the current merging phase, it is determined that the time overlaps with the current merging phase, and merging is performed; or, if there is a time interval inclusion relationship between the two, it is determined that there is an inclusion relationship with the current merging phase, and merging is performed; if none of the above conditions are met, the next sign time record is determined to be the initial sign record of the next change phase.

[0092] When a sign time record is merged into the current merging stage, the earlier of the sign time record's start time and the current merging stage's stage start candidate time is updated as the current merging stage's stage start candidate time; the later of the sign time record's end time and the current merging stage's stage end candidate time is updated as the current merging stage's stage end candidate time. This process continues until subsequent sign time records no longer meet the merging conditions, at which point the current merging stage's stage start candidate time is determined as the start time of the change stage, and the current merging stage's stage end candidate time is determined as the end time of the change stage.

[0093] Each stage of change is assigned a unique stage number, and the stage number, stage start time, stage end time, and the set of indicator numbers belonging to that stage are combined to form a stage boundary record. All stage boundary records are arranged from earliest to latest according to the stage start time to generate stage boundary data. By grouping discrete indicator time records into change stages with start and end times, subsequent responsibility determination and stage succession analysis can be carried out around stable stage units, reducing the boundary drift problem caused by directly performing subsequent analysis around discrete indicators.

[0094] Step S33: Based on the stage boundary data, the original data in the same segment is segmented to obtain stage data.

[0095] Read the stage boundary data and the original data of the same segment. For each stage boundary record, use the stage start time and stage end time as the truncation boundary, and extract all valid data groups whose sampling time is within the time range from the original data of the same segment to form the corresponding stage data segment.

[0096] A minimum data set threshold is preset for each stage. The method for setting this threshold is as follows: extract the shortest, truly valid stage data segment from the calibration sample, and calculate the total duration covered by this shortest stage data segment; then divide this total duration by the uniform alignment time interval and round up to determine the minimum data set threshold for each stage. When the number of valid data segments in a stage data segment is less than the minimum data set threshold, the stage data segment is determined to be an invalid stage data segment, and its corresponding stage boundary record is deleted; when the number of valid data segments in a stage data segment is not less than the minimum data set threshold, the stage data segment is determined to be a valid stage data segment.

[0097] For each valid stage data segment, record the stage number, stage start time, stage end time, the sequence of valid data groups contained in that stage, and the set of sign numbers corresponding to that stage. Arrange all valid stage data segments in ascending order of stage start time to generate stage data.

[0098] Step S4: Based on the original data and stage data of the same segment, obtain the responsibility data through stage adaptation responsibility determination.

[0099] Step S41: Based on the original data of the same segment and the stage boundary data, extract the response start time and response end time corresponding to the multi-source data in each change stage to obtain the stage response data.

[0100] Read the stage boundary data and stage data. For each stage, use wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data as five types of response sources, and extract the corresponding response start time and response end time for each response source within that stage.

[0101] For wheel speed data, the changes in the speed of all four wheels are read sequentially at each alignment moment in the valid data set sequence corresponding to the current stage. A preset wheel response consistency ratio threshold is established. The method for setting the wheel response consistency ratio threshold is as follows: extract real wheel response sample segments for straight-line stable driving, steering, and braking from the calibration samples, and count the proportion of the number of wheels in each sample segment that meet the same direction of change and whose absolute value of speed change is not less than the wheel speed change amplitude threshold to the total number of four wheels; then sort the wheel coverage ratios corresponding to the real wheel response samples from smallest to largest, remove low-end samples according to the preset sample removal ratio threshold, and take the smallest wheel coverage ratio in the remaining samples as the wheel response consistency ratio threshold. When the proportion of the number of wheels in the four wheels that meet the same direction of change and whose absolute value of speed change is not less than the wheel speed change amplitude threshold to the total number of four wheels is not less than the wheel response consistency ratio threshold, the wheel speed data at that alignment moment is determined to be in a valid response state. Statistical analysis is performed on multiple consecutive alignment moments in a valid response state within the current stage. When the number of consecutive alignment points is not less than the minimum continuous alignment point number threshold, the continuous segment is determined as the wheel speed response segment. The start alignment time of the earliest wheel speed response segment in the current stage is determined as the response start time of the wheel speed data, and the end alignment time of the latest wheel speed response segment is determined as the response end time.

[0102] For steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data, the amplitude threshold and minimum continuous alignment point threshold corresponding to steps S22, S23, and S24 are used respectively to identify the corresponding effective response segments in the current stage, and the start alignment time of the earliest effective response segment is extracted as the response start time, and the end alignment time of the latest effective response segment is extracted as the response end time.

[0103] A threshold for merging response segments within a preset stage is established. The method for setting this threshold is as follows: In the calibration sample, multiple response sub-segments from the same response source appearing within the same change stage are statistically analyzed. The interruption time between adjacent response sub-segments is calculated and sorted from smallest to largest. The interruption time corresponding to a preset threshold value is taken as the threshold for merging response segments within the stage. When multiple valid response segments from the same response source appear within the same stage, and the time interval between two adjacent valid response segments is less than or equal to the threshold for merging response segments within the stage, these two adjacent valid response segments are merged into a single response segment.

[0104] For each stage and each response source, the response duration and response intensity are calculated. The response duration is calculated by subtracting the response start time from the response end time and adding a uniform alignment time interval. The response intensity is calculated by extracting the absolute value of the corresponding response index for all valid data points of the response source between its response start time and response end time, then calculating the average of all corresponding response index absolute values, and dividing the average by the response threshold corresponding to the response source to obtain the normalized response intensity. Specifically, the response index for wheel speed data is obtained as follows: For each alignment time between the response start time and response end time, wheels that meet the wheel response consistency ratio threshold condition are first screened out, then the absolute value of the speed change corresponding to all screened wheels at that alignment time is averaged to obtain the wheel speed response value corresponding to that alignment time; subsequently, the wheel speed response values ​​corresponding to all alignment times between the response start time and response end time are averaged to obtain the response index for wheel speed data. The response index for steering angle data is the absolute value of the steering angle change; for yaw rate data, it is the absolute value of the yaw rate; for lateral acceleration data, it is the absolute value of the lateral acceleration change; and for longitudinal acceleration data, it is the absolute value of the longitudinal acceleration change. The response source identifier, stage number, response start time, response end time, response duration, and response intensity are combined to form a stage response record. All stage response records are then sorted by stage number and response start time to generate stage response data.

[0105] Step S42: Based on the phase response data, determine the multi-source data that is earliest at the start time of the response as the initial responsibility data.

[0106] The phase response data is grouped and processed according to phase number. Within the same phase, the response start time is extracted from the response records of each phase. The response source corresponding to the phase response record with the earliest response start time is determined as the source of the initial responsibility for that phase.

[0107] A threshold for determining parallel start-of-duty status is preset. The method for setting this threshold is as follows: In the calibration sample, the response start time differences between different response sources manually determined to have synchronized start are statistically analyzed and sorted from smallest to largest. The time difference corresponding to a preset threshold is used as the threshold for determining parallel start-of-duty status. When the time difference between the response start times of two or more response sources is less than or equal to the threshold, it is determined to be a parallel start. When parallel start occurs, the response intensity is first compared, and the response source with the larger intensity is determined as the source of the starting duty. If the response intensity is still the same, the source of the starting duty is determined in the following order: wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data.

[0108] The initial responsibility record is formed by combining the phase number, the source identifier of the initial responsibility, the start time of the response, the end time of the response, the duration of the response, and the intensity of the response.

[0109] Step S43: Based on the phase response data, determine the multi-source data with the longest response duration as the continuous responsibility data.

[0110] The response data is grouped and processed according to the phase number. Within the same phase, the response duration is read from the phase response record corresponding to each response source. The response source corresponding to the phase response record with the longest response duration is determined as the source of continuous responsibility for that phase.

[0111] A threshold for determining parallel persistent responsibilities is preset. The method for setting this threshold is as follows: In the calibration sample, the duration differences between different response sources manually judged to have similar levels of persistent effect are statistically analyzed and sorted from smallest to largest. The duration difference corresponding to a preset threshold is used as the threshold for determining parallel persistent responsibilities. When the difference in response duration between two or more response sources is less than or equal to the threshold, they are determined to be parallel persistent responsibilities. In the event of a parallel persistent responsibilities, the response intensity is first compared, and the response source with the greater intensity is identified as the source of persistent responsibility. If the response intensity is still the same, the response start time is compared, and the response source with the earlier start time is identified as the source of persistent responsibility.

[0112] The continuous responsibility record is formed by combining the phase number, the source identifier of the continuous responsibility, the response start time, the response end time, the response duration, and the response intensity.

[0113] Step S44: Based on the phase response data, the multi-source data that crosses the boundaries of adjacent change phases is identified as the data for taking over responsibilities.

[0114] Based on the stage boundary data, adjacent stage pairs corresponding to two adjacent change stages are extracted. For each adjacent stage pair, the stage boundary time is determined based on the end time of the preceding stage and the start time of the following stage.

[0115] For calibration samples involving the setting of continuity relationships, a unified manual labeling process was implemented. The manual labeling method was as follows: multiple technicians independently labeled the samples based on the synchronization curves before and after the stage boundary, real vehicle video playback, driving operation records, and the direction and duration of change on both sides of the boundary. When the labeling results were consistent, the labeling results were directly adopted; when the labeling results were inconsistent, they were reviewed by review technicians, and the review results were used as the final labeling results. Among them, the true cross-stage boundary continuity relationship was defined as: the same response source maintained continuous and effective change before the previous stage boundary and continued to maintain continuous and effective change in the same direction after the subsequent stage boundary, and there were no independent new operation triggers on both sides of the boundary; the continuity sample was defined as: two adjacent change stages belonged to the same continuous sideslip change process at the stage boundary, and at least one target multi-source data on both sides of the boundary satisfied the consistency of the front and rear directions and the continuous connection of the front and rear changes in time; the true non-continuity sample was defined as: although two adjacent change stages were adjacent in time, the changes on both sides of the boundary no longer belonged to the same continuous sideslip change process. When there is a conflict between the judgment conclusions corresponding to the synchronization curves before and after the stage boundary, the actual vehicle video playback, and the driving operation record, the judgment conclusion corresponding to the synchronization curves before and after the stage boundary shall be given priority; if the synchronization curves are insufficient to make a direct judgment, the judgment conclusion corresponding to the driving operation record shall be used; if the driving operation record is still insufficient to make a judgment, the judgment conclusion corresponding to the actual vehicle video playback shall be used; if the review technicians still cannot form a clear judgment according to the above order, the sample shall be determined as an invalid sample and shall not enter the acceptance threshold setting process.

[0116] A preset threshold for proximity to the acceptance boundary is established. The threshold is set as follows: In the calibration samples, the time difference between the end times of the preceding stage response and the following stage response is calculated. These two types of time differences are then merged and sorted. After removing samples with high-value time differences according to a preset sample removal ratio threshold, the maximum value among the remaining time differences is taken as the threshold for proximity to the acceptance boundary. For each adjacent stage pair, the stage response records of the preceding and following stages are read. Acceptance determination is performed for each of the five types of response sources. When the same response source has a stage response record in the preceding stage, and the time difference between its response end time and the end time of the preceding stage is less than or equal to the threshold for proximity to the acceptance boundary, and simultaneously, the same response source has a stage response record in the following stage, and the time difference between its response start time and the start time of the following stage is less than or equal to the threshold for proximity to the acceptance boundary, then the response source is determined to cross the stage boundary of the adjacent stage pair and belongs to the acceptance responsibility source of that adjacent stage pair.

[0117] For each response source that meets the acceptance criteria, the duration of acceptance in the preceding and following stages is calculated. The adjacent stage pair identifiers, stage boundary times, response source identifiers, preceding stage numbers, following stage numbers, preceding stage response start times, preceding stage response end times, preceding stage acceptance durations, following stage response start times, following stage response end times, and following stage acceptance durations are combined to form an acceptance responsibility record. By further distinguishing multi-source data within the same change stage into initial responsibility data, continuous responsibility data, and acceptance responsibility data, the roles of different source data in the change formation stage, change duration stage, and stage boundary continuation stage are separated and determined. This provides a clear basis for responsibility in the boundary acceptance analysis of subsequent adjacent change stages, mitigating the problem of mixed roles and unclear boundary attribution when multi-source data participates in subsequent judgments solely based on synchronous change phenomena.

[0118] Step S45: Summarize the initial responsibility data, continuous responsibility data, and inherited responsibility data to obtain responsibility data.

[0119] Read the initial responsibility data, continuous responsibility data, and inherited responsibility data separately. For the initial and continuous responsibility data, use the stage number as an index to map them; for the inherited responsibility data, use the adjacent stage pair identifier as an index to map them. Generate stage responsibility records and boundary responsibility records, and arrange them in chronological order to obtain the responsibility data.

[0120] Step S5: Based on the original data, stage data, and responsibility data of the same segment, obtain the side slip chain data through adjacent stage continuity analysis.

[0121] In step S53, the determination of whether the change directions corresponding to the front and rear sides of the stage boundary are consistent is completed within the same target multi-source data type, without forcibly unifying the directional meaning between different types of target multi-source data. Specifically, the change direction of wheel speed data on the front side of the stage boundary is compared based on the sign of the last continuous effective change segment in the front average wheel speed change sequence; the change direction of wheel speed data on the rear side of the stage boundary is compared based on the sign of the earliest continuous effective change segment in the rear average wheel speed change sequence; the change direction of steering angle data is compared based on the change directions of the front and rear steering angle change sequences; the yaw rate data is compared based on the positive and negative directions of the front and rear yaw rate change sequences; the change direction of lateral acceleration data is compared based on the change directions of the front and rear lateral acceleration change sequences; and the change direction of longitudinal acceleration data is compared based on the change directions of the front and rear longitudinal acceleration change sequences. Subsequent comparisons only determine whether the directions of the same target multi-source data are consistent on the front and rear sides of the stage boundary.

[0122] Step S51: Extract the stage boundaries corresponding to adjacent change stages based on the stage data.

[0123] Read the stage data, sort it from earliest to latest according to the stage start time to obtain the stage sequence, and construct adjacent stage pairs. For each adjacent stage pair, determine the stage boundary time and form a stage boundary record.

[0124] Step S52: Based on the responsibility data, extract the corresponding responsibility data on both sides of the stage boundary for each adjacent change stage.

[0125] Read the responsibility data and match the stage boundary records with the boundary responsibility records according to the identifiers of adjacent stage pairs. For each adjacent stage pair, form a set of candidate responsibility records.

[0126] Calculate the acceptance support for each candidate responsibility assignment record. The acceptance support is calculated as follows: sum the duration of the preceding stage and the duration of the following stage corresponding to the candidate responsibility assignment record, and then divide by the sum of the total duration of the preceding and following stages to obtain the normalized acceptance support.

[0127] A preset threshold for the retention ratio of target multi-source data is established. The method for setting this threshold is as follows: In the calibrated sample of accepted responsibilities, the concentration of support among candidate responsibility acceptance records is statistically analyzed. The cumulative support ratio that can be covered after retaining the main sources of responsibility acceptance is observed. The minimum cumulative coverage ratio that effectively covers the main responsibility acceptance while excluding weak sources is then taken as the target multi-source data retention ratio threshold. The method for determining the main sources of responsibility acceptance is as follows: The support in the candidate responsibility acceptance record set corresponding to this sample is sorted from largest to smallest, and source elimination tests are performed one by one. The method for source elimination testing is as follows: Each time, one source of responsibility is eliminated from the current candidate set of responsibility acceptance records. Then, the ratio of the sum of the support of the remaining sources corresponding to the valid responsibility acceptance to the total sum of the support is recalculated, and the boundary acceptance result corresponding to the original sample is reassessed to determine if the acceptance is still valid. When eliminating a source causes the boundary acceptance result corresponding to the original sample to change from valid to invalid, that source is identified as the primary source of responsibility acceptance. When eliminating two or more sources individually would cause the boundary acceptance result corresponding to the original sample to change from valid to invalid, these two or more sources are then identified as the primary sources of responsibility acceptance. All sources mentioned above are identified as primary sources of responsibility, and no longer are retained by just one. When there are more than two primary sources of responsibility, the internal sorting and pruning of the primary sources within that group is discontinued; instead, the entire group of primary sources is used as the basis for subsequent coverage ratio calculations. When removing any single source does not change the original responsibility establishment result, sources are accumulated in descending order of responsibility support, and an overall retention test is performed after each accumulation. When the accumulated set of retained sources can, for the first time, independently maintain the original responsibility establishment result, this accumulated set of retained sources is identified as the primary sources of responsibility. The proportion of the sum of the responsibility support corresponding to the primary sources of responsibility to the sum of the total responsibility support is calculated, and this proportion is summarized in sample order. Then, all proportions are sorted in ascending order, and the proportion corresponding to the preset threshold is taken as the coverage ratio of the primary sources of responsibility.

[0128] For each adjacent stage, the support for undertaking responsibilities in the corresponding candidate responsibility record set is sorted from largest to smallest, and the support is accumulated sequentially. When the proportion of the accumulated support to the total support is not less than the target multi-source data retention ratio threshold for the first time, all the responsibility sources currently covered by the accumulated support are retained as target multi-source data. Using the accumulated support coverage method as the retention rule is to prioritize retaining response sources that can cover the main undertaking roles within the current adjacent stage, while suppressing candidate sources with significantly weak support.

[0129] The corresponding stage boundary time, the preserved target multi-source data set, and the support degree corresponding to each target multi-source data are combined to form the support analysis input record.

[0130] Step S53: Based on the original data and the data on the responsibility of the same segment, determine the lateral sliding relationship at the stage boundary of each adjacent change stage.

[0131] Step S531: Based on the data of the assumed responsibilities, determine the target multi-source data corresponding to each adjacent change stage on the front and back sides of the stage boundary.

[0132] Read the input records of the continuation analysis, extract the target multi-source data sets corresponding to each adjacent stage, and use them as the observation objects for subsequent boundary change analysis.

[0133] Step S532: Based on the original data of the same segment, extract the boundary change data corresponding to the front and back sides of the stage boundary of the target multi-source data.

[0134] Read the original data from the same segment, and also read the target multi-source data and the corresponding stage boundary times. Preset the observation duration before and after the boundary. The method for setting the observation duration before and after the boundary is as follows: extract adjacent stage pairs with a real connection from the calibration sample, and calculate the shortest continuous observation time required to stably represent the direction of change before and after the stage boundary; then take the larger of the shortest continuous observation time before and after the boundary as the observation duration before and after the boundary. Subsequently, extract the observation data segment before the stage boundary time, extract the observation data segment after the stage boundary time, and extract the corresponding boundary change data according to the target multi-source data type.

[0135] Step S533: Based on the boundary change data, determine whether the change directions corresponding to the front and back sides of the stage boundary are consistent and whether the change is continuous, and obtain the boundary connection result.

[0136] For each target multi-source data, determine the change direction and end time corresponding to the front side of the stage boundary, and the change direction and start time corresponding to the rear side of the stage boundary.

[0137] The system presets a boundary direction determination threshold and a minimum consecutive point threshold for the boundary. The boundary direction determination threshold is not reset separately; instead, it adopts the change amplitude threshold already determined during the change sign extraction stage of the corresponding target multi-source data. Specifically, wheel speed data uses the wheel speed change amplitude threshold, steering angle data uses the turning change amplitude threshold, yaw rate data uses the continuous yaw change amplitude threshold, lateral acceleration data uses the lateral change amplitude threshold, and longitudinal acceleration data uses the longitudinal change amplitude threshold. The minimum consecutive point threshold for the boundary is set as follows: based on the continuous observation time corresponding to the observation duration before or after the boundary, this continuous observation time is divided by the uniform alignment time interval and rounded up to determine the minimum consecutive point threshold for the boundary.

[0138] If the number of consecutive alignment points in the last consecutive segment within the current observation data segment is not less than the minimum consecutive point threshold of the boundary, and the absolute value of the corresponding change in the consecutive segment is not less than the boundary direction determination threshold, then the consecutive segment is determined as the last consecutive valid change segment on the front side; the same rule is used to determine the earliest consecutive valid change segment on the back side for the subsequent observation data segment.

[0139] Based on the last continuous effective change segment on the front side and the earliest continuous effective change segment on the rear side, the direction of change on the front side, the end time of change on the front side, and the start time of change on the rear side are determined respectively. Then, according to the direction comparison caliber within the respective target multi-source data types, it is determined whether the direction of change on the front side and the direction of change on the rear side are consistent. For wheel speed data, when the average wheel speed change within the corresponding continuous effective change segment is greater than zero, it is determined as a positive direction; when the average wheel speed change within the corresponding continuous effective change segment is less than zero, it is determined as a negative direction; when the average wheel speed change within the corresponding continuous effective change segment is equal to zero, it is determined as a zero direction. For steering angle data, lateral acceleration data, and longitudinal acceleration data, when the average change within the corresponding continuous effective change segment is greater than zero, it is determined as a positive direction; when the average change within the corresponding continuous effective change segment is less than zero, it is determined as a negative direction; when the average change within the corresponding continuous effective change segment is equal to zero, it is determined as a zero direction. For yaw rate data, a positive direction is defined as an average yaw rate within a corresponding continuous effective change segment when the average yaw rate is greater than zero; a negative direction is defined as an average yaw rate less than zero; and a zero direction is defined as an average yaw rate equal to zero. When both the direction of change corresponding to the area before and after the stage boundary are zero, the multi-source data for that target is determined to lack a valid direction at that stage boundary. When either the direction of change corresponding to the area before or after the stage boundary is zero, and the other is non-zero, the directions of change are considered inconsistent. Only when both the direction of change corresponding to the area before and after the stage boundary are the same non-zero direction are the directions of change considered consistent.

[0140] A preset boundary continuity time threshold is established. The method for setting the boundary continuity time threshold is as follows: In the calibration sample, among adjacent stages with a real continuity relationship, the time difference between the end time of the preceding effective change segment and the start time of the following effective change segment is statistically analyzed. These differences are then sorted from smallest to largest, and the time difference corresponding to a preset threshold value is taken as the boundary continuity time threshold. When the time difference between the two sides of the boundary is not greater than the boundary continuity time threshold, the change is considered continuous; otherwise, the change is considered discontinuous.

[0141] For each target multi-source data, a source-level boundary acceptance result is generated based on whether the direction of change is consistent and whether the change is continuous. When the direction of change is consistent and the change is continuous, the source-level boundary acceptance result is marked as successful; otherwise, it is marked as unsuccessful.

[0142] A preset threshold for the acceptance / support ratio is established. The method for setting this threshold is as follows: Extract true acceptance samples and true non-acceptance samples from the calibration sample. Calculate the ratio of the sum of acceptance support corresponding to the source of acceptance in each sample to the total sum of acceptance support. Then, merge and sort the ratios of the two types of samples. For every two adjacent and unequal ratios, take the midpoint as a candidate cutoff value. If adjacent ratios are equal, no corresponding candidate cutoff value is generated. When only one unique ratio exists after merging and sorting, use that unique ratio as the candidate cutoff value. For each candidate cutoff value, count the number of false positives and false negatives, and calculate the total number of false positives. Determine the candidate cutoff value with the minimum number of false positives as the acceptance / support ratio threshold. When multiple candidate cutoff values ​​correspond to the same minimum number of false positives, select the candidate cutoff value with fewer false positives as the acceptance / support ratio threshold. When multiple candidate cutoff values ​​correspond to the same minimum number of false positives and also have the same number of false positives, select the candidate cutoff value with the larger value as the acceptance / support ratio threshold. When a certain adjacent stage corresponds to multiple target multi-source data, a weighted summary is performed on the boundary acceptance results of multiple source levels; when the sum of acceptance support corresponding to the source that is accepted is not less than the total acceptance support ratio threshold, the boundary acceptance result of the adjacent stage is determined to be accepted; otherwise, it is determined to be unacceptable.

[0143] Step S534: Based on the boundary connection results, determine the lateral sliding connection relationship at the stage boundary of each adjacent change stage.

[0144] For each pair of adjacent stages, extract its boundary connection result. When the boundary connection result is valid, the sideslip connection relationship of the adjacent stage pair at the stage boundary is determined to be valid; when the boundary connection result is invalid, the sideslip connection relationship of the adjacent stage pair at the stage boundary is determined to be invalid. A sideslip connection record is formed. By performing boundary change extraction, direction consistency determination, and change continuity determination on the target multi-source data corresponding to the front and back sides of the stage boundary, it is possible to clearly distinguish whether adjacent change stages belong to the same continuous sideslip change process at the stage boundary. This provides a boundary connection basis for the generation of sideslip chain data, alleviating the problem in the prior art that adjacent stages are temporally connected but it is impossible to determine whether they should be continuously merged.

[0145] Step S54: Based on the lateral slip connection relationship corresponding to each adjacent change stage, perform chain association according to the stage boundaries to obtain lateral slip chain data.

[0146] Read all sideslip connection records and sort them from earliest to latest according to the stage boundary time. Preset sideslip chain start rules and sideslip chain continuation rules. The first change stage after sorting is determined as the starting stage of the first sideslip chain; when an adjacent stage is marked as having an invalid sideslip connection, the subsequent stage in that adjacent stage pair is determined as the starting stage of a new sideslip chain; when an adjacent stage is marked as having an valid sideslip connection, the subsequent stage is merged into the same sideslip chain as the preceding stage. Generate a chain record for each sideslip chain and arrange all chain records from earliest to latest according to the chain start time to obtain the sideslip chain data.

[0147] Step S6: Based on the slip chain data, stage data, and responsibility data, the comprehensive detection result of the slip amount is obtained through continuous integration.

[0148] Step S61: Based on the stage data, determine the order of each change stage to obtain stage sequence data.

[0149] Read the stage data, sort all the changing stages stably by stage start time, stage end time and stage number, and generate stage sequence records to obtain stage sequence data.

[0150] Step S62: Based on the side slip chain data and the stage sequence data, extract the side slip connection relationships corresponding to adjacent change stages in the order of arrangement to obtain connection relationship data.

[0151] Read the side slip chain data and stage sequence data, construct sequentially adjacent stage pairs for two adjacent change stages in sequence position, query their corresponding side slip chain connection relationships, form connection relationship records, and obtain connection relationship data.

[0152] Step S63: Based on the responsibility data and the stage sequence data, determine the priority receiving stage corresponding to the adjacent change stages in the order of arrangement, and obtain the priority receiving data.

[0153] Based on the stage sequence data, extract the stage boundary time corresponding to each sequentially adjacent stage pair at the stage boundary; then, based on the responsibility data, extract the corresponding responsibility data corresponding to each sequentially adjacent stage pair at the stage boundary. For each sequentially adjacent stage pair, calculate the sum of the duration of the preceding stage and the sum of the duration of the following stage.

[0154] A preset priority acceptance comparison threshold is established. The method for setting the priority acceptance comparison threshold is as follows: In the calibration sample, among adjacent stage pairs with similar acceptance strength on both sides, the absolute value of the difference between the sum of the acceptance duration of the preceding stage and the sum of the acceptance duration of the following stage is calculated. These values ​​are then sorted from smallest to largest, and the absolute value of the difference corresponding to a preset positional threshold is taken as the priority acceptance comparison threshold. When the difference between the sum of the acceptance duration of the following stage and the sum of the acceptance duration of the preceding stage is greater than the priority acceptance comparison threshold, the following stage is determined to have a stronger acceptance duration characteristic at the stage boundary. When the difference between the sum of the acceptance duration of the preceding stage and the sum of the acceptance duration of the following stage is greater than the priority acceptance comparison threshold, the preceding stage is determined to have a stronger acceptance duration characteristic at the stage boundary. When the absolute value of the difference between the sum of the acceptance duration of the preceding and following stages is less than or equal to the priority acceptance comparison threshold, the preceding and following stages are determined to be equal in terms of acceptance duration.

[0155] When ties occur, first compare the number of responsibility sources in the candidate responsibility record set for the preceding and following stages, and determine the stage with more responsibility sources as the priority stage. If the number of responsibility sources is still the same, then compare the average responsibility support corresponding to the preceding and following stages, and determine the stage with the higher average responsibility support as the priority stage. If the average responsibility support is still the same, further compare the response strength of the continuous responsibility sources corresponding to the preceding and following stages, and determine the stage with the higher response strength of the continuous responsibility sources as the priority stage. If the response strength of the continuous responsibility sources is still the same, then determine the following stage as the priority stage. The following stage is used as the final fallback rule because the output of this step will be directly used for subsequent continuous integration, and the goal of continuous integration is to prioritize maintaining consistent coverage of the continuity of changes in subsequent stages when the change conditions on both sides of the boundary are completely consistent.

[0156] The previous stage number, the next stage number, the stage boundary time, the priority stage number, the sum of the duration of the previous stage and the sum of the duration of the next stage are combined to form the priority stage record, thus obtaining the priority stage data.

[0157] Step S64: Based on the data of the receiving relationship and the data of the priority receiving, the sideslip changes corresponding to each stage of change are continuously integrated to obtain the comprehensive detection result of the sideslip amount.

[0158] Read the succession relationship data, priority succession data, stage data, and responsibility data. Use each change stage as an initial integration unit. When a sequentially adjacent stage pair is marked as having a succession relationship in the succession relationship data, and a priority succession stage exists in the priority succession data, merge the integration unit containing the previous stage and the integration unit containing the next stage in that sequentially adjacent stage pair to form a new continuous integration unit; continue until all sequentially adjacent stage pairs have been processed.

[0159] For each continuous integration unit, extract its phase number sequence, corresponding side slip chain number set, corresponding initial responsibility source sequence, continuous responsibility source sequence, and boundary acceptance source sequence. Preset selection rules for the core integration output phase are as follows: When a continuous integration unit contains only one change phase, directly determine that change phase as the core phase of the continuous integration unit; when a continuous integration unit contains multiple change phases, count the occurrences of the phase numbers identified as priority acceptance phases in all priority acceptance records within the continuous integration unit, and determine the phase with the highest occurrences as the core phase of the continuous integration unit; when the occurrence counts are tied, determine the phase with the larger sum of response intensities corresponding to continuous responsibility sources as the core phase of the continuous integration unit; when the sum of response intensities is still the same, determine the phase with the later sequential position as the core phase of the continuous integration unit.

[0160] For each continuous integration unit, a lateral slip change record is generated. The lateral slip change record includes at least the integration unit number, integration start time, integration end time, sequence of integration stage numbers, number of integration stages, core stage number, starting responsibility source corresponding to the core stage, continuing responsibility source corresponding to the core stage, core stage response strength, and integration status identifier. The core stage identifies the dominant change stage in the continuous integration unit in terms of boundary continuity and stage continuity response. The number of integration stages characterizes the stage range covered by the continuous integration unit, and the core stage response strength characterizes the strength of the response of the dominant change stage within the continuous integration unit; both together constitute the change degree characterization field. The core stage response strength is determined by reading the response strength of the continuing responsibility source corresponding to the core stage in the stage response data and identifying that response strength as the core stage response strength. The integration status identifier includes a continuous integration identifier and an independent stage integration identifier. When a continuous integration unit is formed by merging multiple change stages after a succession relationship is established, the integration status identifier is determined as a continuous succession integration identifier; when the continuous integration unit contains only a single change stage, or there is no succession relationship between adjacent change stages, the integration status identifier is determined as an independent stage integration identifier.

[0161] The output format of the comprehensive sideslip detection results is preset. The output format is set to a continuous integration result output format, meaning the integrated set of sideslip change records serves as the comprehensive sideslip detection result; each integrated sideslip change record is considered a recording unit for the comprehensive sideslip detection result. The continuous integration result output format is adopted because the main focus of this method revolves around the division of change stages, determination of stage adaptation responsibilities, analysis of adjacent stage continuity, and continuous integration. Its final output represents the comprehensive detection result of sideslip-related changes across continuous stages. The comprehensive sideslip detection result is the comprehensive detection output result of sideslip-related changes across continuous stages; the sideslip quantity is the result quantity formed after comprehensively representing the continuity relationship, integration status, and degree of change characterization fields of sideslip-related changes across continuous stages. Its output format is the integrated set of sideslip change records, not limited to a single numerical form.

[0162] All integrated sideslip change records are arranged from earliest to latest according to the integration start time to obtain the comprehensive sideslip detection result. When a certain change stage does not form a succession relationship with other change stages, it is still output as an independent continuous integration unit in the comprehensive sideslip detection result; when all sequentially adjacent stage pairs contained in a continuous integration unit have priority succession records but the succession relationship data are all marked as non-succession, stage merging is not performed, and the independent integration results of each change stage are retained separately.

[0163] To verify the ability of this method to reflect continuous sideslip processes, continuous sideslip samples and discontinuous sideslip samples from the calibration sample were selected as comparative validation data. A method that directly performs single-stage fusion judgment based solely on the original data of the same segment was selected as the comparative method. Validation metrics included error segmentation rate, error merging rate, and continuous process judgment accuracy. Specifically, the error segmentation rate was calculated as the proportion of continuous sideslip samples incorrectly segmented into multiple result units out of the total number of continuous sideslip samples; the error merging rate was calculated as the proportion of discontinuous sideslip samples incorrectly merged into the same result unit out of the total number of discontinuous sideslip samples; and the continuous process judgment accuracy was calculated as the proportion of samples correctly identified as either continuous or discontinuous processes out of the total number of validation samples. The comparison showed that this method had lower error segmentation and error merging rates than the comparative method, but higher continuous process judgment accuracy. This demonstrates that this method can more stably identify continuous sideslip processes while maintaining the ability to distinguish stage boundaries.

[0164] By continuously integrating the sideslip changes corresponding to each stage of change based on the data of the succession relationship and the priority succession data, the comprehensive detection results of sideslip amount are no longer limited to the results of each stage, but can be characterized in the form of continuous integrated units to represent the continuity relationship, dominant stage and integration state of sideslip-related changes in time sequence, thereby improving the ability of the detection results to reflect the continuous sideslip process.

[0165] Through the above steps, multi-source data no longer participates in sideslip detection in a parallel input manner, but can form a continuous detection link in the order of change signs, change stages, division of responsibilities, boundary acceptance and continuous integration. This alleviates the problems of sideslip change being easily fragmented at stage boundaries, unclear roles of data from different sources, and insufficient continuity of detection results.

[0166] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A comprehensive detection method for vehicle sideslip based on multi-sensor fusion, characterized in that, include: The collected multi-source data is processed into the same segment to obtain the original data of the same segment; Based on the original data in the same segment, indicator data is obtained by extracting signs of change. Based on the indication data, stage data is obtained by dividing the original data in the same segment into stages of change; Based on the original data and stage data of the same segment, responsibility data is obtained through stage adaptation responsibility determination; Based on the original data, stage data and responsibility data of the same segment, the side slip chain data is obtained through the analysis of adjacent stage succession. Based on the slip chain data, stage data, and responsibility data, a comprehensive detection result of the slip volume is obtained through continuous integration.

2. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 1, characterized in that, Methods for processing collected multi-source data into the same segment to obtain the original data of the same segment include: Multi-source data includes wheel speed data, steering angle data, yaw rate data, lateral acceleration data, and longitudinal acceleration data; Based on the sampling time sequence of each of the multi-source data, determine the overlapping time period covered by the multi-source data; Based on overlapping time periods, the multi-source data is segmented and time-series aligned to obtain the original data for the same segment.

3. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 2, characterized in that, Methods for obtaining symptom data include: Extract signs of speed change from the wheel rotation speed data in the same segment of raw data; Extract signs of turning changes based on the steering angle data in the same segment of raw data; Based on the yaw rate data in the same segment of raw data, extract signs of continuous directional change; Based on the lateral and longitudinal acceleration data in the same segment of raw data, extract signs of linkage changes; The signs are arranged in chronological order of appearance based on the signs of rapid or slow changes, turning points, continuous changes in direction, and interconnected changes, thus obtaining the sign data.

4. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 1, characterized in that, Methods for obtaining stage data include: Based on the indicator data, determine the start and end times of each change indicator in the same segment of raw data to obtain indicator time data; Based on the indicator time data, the time sequence of each change indicator is merged to determine the start and end times of each change stage, thus obtaining the stage boundary data. Based on the stage boundary data, the original data in the same segment is divided into segments to obtain stage data.

5. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 4, characterized in that, Methods for obtaining responsibility data through phase adaptation responsibility determination include: Based on the original data and stage boundary data of the same segment, extract the response start time and response end time corresponding to the multi-source data in each change stage to obtain the stage response data; Based on the phase response data, the multi-source data with the earliest response start time is determined as the initial responsibility data, the multi-source data with the longest response duration is determined as the continuous responsibility data, and the multi-source data that crosses the boundary of adjacent change phases is determined as the successor responsibility data. The initial responsibility data, continuous responsibility data, and inherited responsibility data are aggregated to obtain responsibility data.

6. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 5, characterized in that, Methods for obtaining side slip chain data through adjacent stage continuity analysis include: Based on the stage data, extract the stage boundaries corresponding to adjacent change stages; Based on the responsibility data, extract the corresponding responsibility data on both sides of the stage boundary for each adjacent change stage; Based on the original data and the data on the responsibilities of each segment, determine the lateral sliding connection relationship at the stage boundary of each adjacent change stage; Based on the sideslip connection relationship between each adjacent change stage, a chain association is performed according to the stage boundaries to obtain sideslip chain data.

7. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 6, characterized in that, Methods for determining the sideslip continuity relationship between adjacent change stages at stage boundaries include: Based on the data of the assumed responsibilities, determine the target multi-source data corresponding to each adjacent change stage before and after the stage boundary; Based on the original data of the same segment, extract the boundary change data corresponding to the front and back of the stage boundary of the target multi-source data; Based on the boundary change data, determine whether the change directions corresponding to the front and back sides of the stage boundary are consistent and whether the change is continuous, and obtain the boundary connection result. Based on the boundary acceptance results, the lateral sliding acceptance relationship at the stage boundary of each adjacent change stage is determined.

8. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 7, characterized in that, Methods for obtaining boundary acceptance results include: Based on the boundary change data, determine the change direction and end time corresponding to the front side of the stage boundary; Based on the boundary change data, determine the change direction and start time corresponding to the back side of the stage boundary; Determine whether the change directions are consistent based on the change direction corresponding to the front side of the stage boundary and the change direction corresponding to the back side of the stage boundary; Determine whether the change is continuous based on the end time corresponding to the front side of the stage boundary and the start time corresponding to the back side of the stage boundary. The boundary connection result is obtained based on whether the direction of change is consistent and whether the change is continuous.

9. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 1, characterized in that, Methods for obtaining comprehensive sideslip measurement results through continuous integration include: Based on the stage data, determine the order of each stage of change to obtain stage sequence data; Based on the side slip chain data and stage sequence data, the side slip chain connection relationships corresponding to adjacent change stages in the order of arrangement are extracted to obtain the connection relationship data; Based on the responsibility data and the phase sequence data, the priority takeover phase corresponding to the adjacent change phases in the order of arrangement is determined, and the priority takeover data is obtained. Based on the data on the connection relationships and the data on priority connections, the sideslip changes corresponding to each stage of change are continuously integrated to obtain the comprehensive detection results of the sideslip amount.

10. The method for comprehensive detection of vehicle sideslip based on multi-sensor fusion according to claim 9, characterized in that, Methods for obtaining priority data acceptance include: Based on the stage sequence data, extract the stage boundaries corresponding to each adjacent change stage at the stage boundary. Based on the responsibility data, extract the corresponding responsibility data at the stage boundary of each adjacent change stage; Based on the data on the assumed responsibilities, compare the duration of the assumed responsibilities at the stage boundaries of adjacent change stages. Based on the duration of the acceptance, the change phase with a longer duration is identified as the priority acceptance phase, and priority acceptance data is obtained.

Images