Semitrailer reversing folding real-time monitoring system based on multi-sensor fusion

Abstract

The invention relates to the technical field of semi-trailer safety monitoring, and discloses a semi-trailer reversing folding real-time monitoring system based on multi-sensor fusion. The data acquisition module acquires sensor data such as position, angle and speed, and divides normal and potential folding areas through time synchronization processing; the data fusion module generates a fusion state vector and calibrates a space coordinate system; the folding state identification module judges the folding risk according to the path trajectory density, the motion feature distribution pattern and the span of the key monitoring point; and the real-time monitoring module expands the data area to generate a continuous track, and evaluates the folding probability in real time by using a curve fitting slope change rate and a convolutional neural network model. The system integrates multiple types of sensors, and adopts dynamic filtering, multi-anchor-point reference and intelligent algorithms, so that accurate real-time monitoring of the backing folding risk of the semi-trailer is realized, and the driving safety is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of semi-trailer vehicle safety monitoring, and specifically to a real-time monitoring system for semi-trailer vehicle reverse folding based on multi-sensor fusion. Background Art

[0002] Due to its large load capacity and high transportation efficiency, semi-trailer vehicles are widely used in the logistics transportation field. However, their special articulated structure poses a significant folding risk during the reverse process (i.e., abnormal angular deflection between the tractor and the trailer, which may cause accidents such as collisions and rollovers). Traditional monitoring methods mainly rely on single sensors (such as cameras or ultrasonic sensors), and have problems such as limited monitoring range, insufficient data accuracy, and poor real-time performance, making it difficult to comprehensively and accurately capture the complex motion state of the vehicle during reverse driving.

[0003] In the prior art, some monitoring systems do not fully consider the time synchronization problem of multi-sensor data, resulting in time deviations in the data collected by different types of sensors (such as GPS positioning sensors, IMU angle sensors, and wheel speed sensors), which cannot be effectively fused, thus affecting the accuracy of the monitoring results. In addition, when judging the folding risk of the vehicle, traditional methods often only rely on the angular change in a single dimension or a simple geometric model, without combining multi-dimensional information such as the density distribution of the vehicle motion trajectory and the span analysis of motion characteristics, and it is easy to have false positives or false negatives. For example, when the vehicle motion trajectory shows a non-linear distribution or the longitudinal / transverse span exceeds the safety threshold, it is difficult for traditional systems to quickly identify potential high-risk states.

[0004] In terms of real-time performance, existing systems usually adopt filtering algorithms with fixed parameters or simple curve fitting methods, and cannot dynamically adapt to the data change characteristics at different driving speeds of the vehicle. For example, if the window length of the moving average filter cannot be dynamically adjusted according to the vehicle speed, it may lead to poor data smoothing effect, affect the calculation accuracy of the slope of the state change curve, and thus cannot detect abnormal changes in the folding state in time. At the same time, the lack of a dynamic expansion mechanism for the data area of the monitoring unit and a continuous trajectory generation mechanism makes it difficult to effectively connect the data of adjacent monitoring units and form a complete chain for analyzing the motion state.

[0005] In addition, the definition of the preset safe angle range in the prior art is relatively rough, and the specific positions and distribution rules of the anchor points are not clear, resulting in a lack of a scientific benchmark when calculating the path trajectory density and angle change. For example, if the angle peak points in the front-end area and the angle change points in the rear-end area are not used as key anchor points, it is difficult to accurately divide the normal driving area and the potential folding area. During the coordinate transformation process, if only single-axis calibration is used or the multi-axis azimuth deviation of the boundary contour line is not considered, it will lead to a large calibration error in the spatial coordinate system of the sensor data, affecting the generation accuracy of the subsequent fusion state vector. Summary of the Invention

[0006] The object of the present invention is to provide a real-time monitoring system for the reverse folding of semi-trailer trucks based on multi-sensor fusion to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions: A real-time monitoring system for the reverse folding of semi-trailer trucks based on multi-sensor fusion, the system includes:

[0008] A data acquisition module, which is used to obtain multiple sensor data of the semi-trailer truck, including position data, angle data and speed data, and perform time synchronization processing on the sensor data to divide the normal driving area and potential folding area of the vehicle;

[0009] A data fusion module, which is used to generate a fusion state vector based on the sensor data and perform coordinate transformation according to the vehicle geometric model;

[0010] A folding state recognition module, which is used to select any monitoring unit of the vehicle hinge point, calculate the starting angle value and ending angle value of each monitoring point based on the fusion state vector, both are marked as reference points, obtain several reference points closest to the geometric center point of the monitoring unit, define them as key monitoring points, sequentially associate adjacent key monitoring points to generate a path trajectory, obtain the density of the path trajectory covering a preset safety angle interval, when the density is less than the preset threshold, identify whether it is a motion feature of linear distribution or curve distribution, if not, then judge that the angle change of the monitoring unit is abnormal; if so, then detect the span of the longitudinal or transverse distribution of the motion feature, if the span is greater than the preset span value, then judge that the folding risk of this unit is high;

[0011] A real-time monitoring module, which is used to select an arbitrary one-dimensional fusion state vector sequence, longitudinally expand the data area of each monitoring unit based on the axis where the unit center point is located until the data areas in adjacent monitoring units form a continuous motion trajectory, generate a continuous state change curve, calculate the slope change rate of the curve fitting equation, when the deviation between the change rate and the standard value exceeds the preset range, then judge that the real-time performance of the folding state in this dimension does not meet the standard.

[0012] Preferably, the process of the time synchronization processing is as follows:

[0013] Perform timestamp alignment processing on the multiple sensor data to generate synchronized data, perform dynamic range adjustment on the synchronized data, convert the synchronized data into normalized data by setting data thresholds, and perform smoothing processing on the normalized data based on moving average filtering, and perform eigen-decomposition on the smoothed normalized data to separate the normal driving area and potential folding area of the vehicle.

[0014] Preferably, the preset safety angle range is composed of 6 reference anchor points, where 2 anchor points are set in the front-end area of the monitoring unit, and the other 4 anchor points are set in the rear-end area of the monitoring unit. The two points in the front-end area are positioned at the angular peak points distributed axially; the two anchor points in the rear-end area are positioned at the intersection points of the unit boundary line and the movement path, that is, the angle change points, and the remaining two points in the rear-end area are positioned at the secondary uniform distribution nodes of the connection line of the angle change points.

[0015] Preferably, the process of coordinate transformation for sensor data is as follows:

[0016] Obtain the boundary contour line of the normal driving area of the vehicle, calculate the azimuth deviation angles between the 3 main axes of the boundary and the sensor data coordinate system respectively, and calibrate the sensor data in the space coordinate system according to the weighted average value of the azimuth deviation angles.

[0017] Preferably, the process of detecting the longitudinal or transverse distribution span of the motion characteristics is as follows:

[0018] Traverse the data area according to the detection direction, where the detection direction is the longitudinal section or the transverse section, count the angle change values of each group of detected data. The angle value of the potential folding area is 180, and the angle value of the normal driving area is 0. Select the first detection group whose angle value breaks through the normal driving threshold and mark it as the risk starting group. Continue to traverse the data area. When a detection group with an angle value returning to the normal driving threshold is detected, mark it as the pending group. Obtain the spatial interval D1 between the pending group and the starting group. If the spatial interval D1 is greater than or equal to the preset span P and there is no angle abnormality when continuing to detect along the detection direction, then mark this pending group as the termination group, and then this motion is a uniformly distributed characteristic; if the spatial interval is less than D1, continue to traverse this data area. When a detection group with an angle breaking through the normal driving threshold is detected again, mark it as the second starting group. Obtain the interval D2 between the second starting group and the pending group. If the interval D2 is greater than the preset span Q, it is determined that the motion distribution is abnormal.

[0019] Preferably, the specific process of determining that the real-time performance of the folding state does not meet the standard is as follows:

[0020] For any selected monitoring unit, select the reference axis where the center point of the unit is located, obtain the offset of all angle points in this data area relative to this axis, adjust the offset of each angle point through a preset scaling factor, continuously increase the value of the scaling factor until adjacent monitoring units all form data superposition areas, obtain the continuous trajectory of the state change, obtain the corresponding fitting curve equation for this trajectory through median filtering, and obtain the slope change rate of this curve equation.

[0021] Preferably, the process of generating the path trajectory is as follows:

[0022] First, select the three reference points closest to the center point of the monitoring unit, associate adjacent reference points to generate an initial path structure, and then successively select the closest points among the remaining reference points and add them to the association of the path structure until the length of the connection path corresponding to the original reference points increases in the association path formed by the newly added reference points. Then, remove the newly added point. The association path formed by the original reference points is the path trajectory of the monitoring unit.

[0023] Preferably, the data acquisition module includes a GPS positioning sensor, an IMU angle sensor, and a wheel speed sensor. The GPS positioning sensors are arranged in a grid-like structure at key positions of the vehicle, the IMU angle sensors are axially distributed at equal intervals, and the wheel speed sensors cover all moving joint areas of the vehicle.

[0024] Preferably, the dynamic range adjustment adopts a linear transformation method to compress the range of the original data to a preset interval while retaining the data distribution characteristics. The window length of the moving average filtering is dynamically adjusted according to the vehicle's movement speed.

[0025] Preferably, the real-time monitoring module is built-in with a convolutional neural network model. This model is trained with historical vehicle folding data, uses the slope change rate of the state change curve as the input feature, and uses the folding occurrence probability as the output result. The adaptive learning rate strategy is adopted to optimize the loss function during model training.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] In the data acquisition and preprocessing link, the data acquisition module integrates a GPS positioning sensor, an IMU angle sensor, and a wheel speed sensor to respectively obtain position, angle, and speed data, and eliminates the time deviation of different sensors through time synchronization processing (including timestamp alignment, dynamic range adjustment, moving average filtering, and feature decomposition) to separate the normal driving area and the potential folding area. Among them, the dynamic range adjustment uses linear transformation to compress the data range and retain the distribution characteristics, and the window length of the moving average filtering is dynamically adjusted according to the vehicle speed to ensure the optimal data smoothing effect at different speeds, laying a foundation for subsequent fusion analysis.

[0028] The data fusion module generates a fusion state vector based on multi-source sensor data and performs coordinate transformation through the vehicle geometric model. Specifically, by calculating the azimuth deviation angles of the three main axes of the boundary contour line of the normal driving area with the sensor coordinate system, the spatial coordinate system calibration is achieved using the weighted average value, effectively reducing the calculation error caused by coordinate deviation and improving the data fusion accuracy.

[0029] The folding state recognition module realizes the accurate judgment of folding risks through an innovative path trajectory generation and risk assessment mechanism. First, a reference point of the hinge point of the monitoring unit is selected, and key monitoring points are determined through distance screening. Then, a path trajectory is associated and generated, and the density covering the preset safe angle range is calculated. When the density is lower than the threshold, the linear / curve distribution attribute of the motion characteristics is further analyzed. If it is a non-linear distribution, the angle change is directly determined to be abnormal; if it is a linear distribution, the longitudinal / transverse span is detected, and when the span exceeds the preset value, a high risk is determined. This process combines multi-dimensional features such as trajectory density, motion form, and spatial span, avoiding the limitations of single indicators and significantly improving the accuracy and robustness of risk recognition.

[0030] The real-time monitoring module generates continuous motion trajectories by dynamically expanding the data area, calculates the slope change rate using median filtering and curve fitting, and predicts the folding occurrence probability in combination with a convolutional neural network (CNN) model. Among them, the data area is longitudinally expanded based on the axis of the unit center point until adjacent units form a continuous trajectory, ensuring the integrity of the state change curve; the CNN model is trained with historical folding data and optimizes the loss function using an adaptive learning rate, which can output the risk probability in real time, providing a forward-looking warning for the driver and effectively improving the real-time performance and intelligence level of the monitoring system.

[0031] The anchor point distribution rule of the preset safe angle range (2 angle peak points at the front end, 2 angle change points at the rear end, and 2 secondary uniformly distributed nodes) provides a scientific benchmark for trajectory density calculation, ensuring a clearer boundary division between normal driving and folding risks. The motion feature span detection process (determining the risk start group, pending group, and end group by traversing the data area and comparing the spatial interval with the preset threshold) can quantitatively analyze the distribution range of angle abnormalities, further improving the credibility of risk judgment.

[0032] Through the organic combination of multi-sensor fusion, spatio-temporal data calibration, multi-dimensional risk feature extraction, and intelligent prediction models, the present invention constructs a complete real-time monitoring system for semi-trailer vehicle reverse folding, effectively solving the problems of low data fusion accuracy, single risk recognition, and insufficient real-time performance in the prior art, and providing strong technical support for improving the driving safety of semi-trailer vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is the working principle diagram of the real-time monitoring system for semi-trailer vehicle reverse folding based on multi-sensor fusion according to the present invention;

[0034] Figure 2 is the working principle diagram of sensor data time synchronization processing;

[0035] Figure 3 is the flow chart of longitudinal / transverse distribution span detection of motion characteristics;

[0036] Figure 4 It is the working principle diagram for real-time judgment of the folding state. Specific implementation manners

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] Please refer to Figures 1 - 4 , the semi-trailer vehicle reversing folding real-time monitoring system based on multi-sensor fusion involved in the present invention includes: a data acquisition module, a data fusion module, a folding state recognition module, and a real-time monitoring module. The specific implementation steps are as follows:

[0039] The data acquisition module obtains multiple sensor data of the semi-trailer vehicle, including position data, angle data, and speed data, and performs time synchronization processing on the sensor data to divide the normal driving area and the potential folding area of the vehicle.

[0040] The data fusion module generates a fusion state vector based on the sensor data and performs coordinate transformation according to the vehicle geometric model.

[0041] The folding state recognition module selects any one monitoring unit of the vehicle hinge point, calculates the starting angle value and the ending angle value of each monitoring point based on the fusion state vector, both are marked as reference points, obtains several reference points closest to the geometric center point of the monitoring unit, defines them as key monitoring points, associates the adjacent key monitoring points in sequence to generate a path trajectory, obtains the density of the path trajectory covering the preset safety angle interval. When the density is less than the preset threshold, it identifies whether it is a motion feature of linear distribution or curve distribution. If not, it determines that the angle change of the monitoring unit is abnormal; if so, it detects the span of the longitudinal distribution or the lateral distribution of the motion feature. If the span is greater than the preset span value, it determines that the folding risk of the unit is high.

[0042] The real-time monitoring module selects an arbitrary one-dimensional fusion state vector sequence, longitudinally expands the data area of each monitoring unit based on the axis where the unit center point is located until the data areas in adjacent monitoring units form a continuous motion trajectory, generates a continuous state change curve, calculates the slope change rate of the curve fitting equation. When the deviation between the change rate and the standard value exceeds the preset range, it determines that the real-time performance of the folding state in this dimension does not meet the standard.

[0043] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0044] Embodiment 1: This embodiment relates to the specific structure of the data acquisition module and the time synchronization processing process. The data acquisition module includes a GPS positioning sensor, an IMU angle sensor, and a wheel speed sensor. Among them, the GPS positioning sensors are arranged in a grid-like structure at key positions of the vehicle. This grid-like structure is designed according to the body structure characteristics of the semi-trailer truck, forming regular or irregular grid nodes at the head, middle of the body, tail, and other main parts. A GPS positioning sensor is deployed at each node to ensure that all main areas of the vehicle body can be covered, realizing the comprehensive acquisition of vehicle position data, and accurately obtaining the coordinate position and movement trajectory of the vehicle in three-dimensional space. The IMU angle sensors are axially distributed at equal intervals and are deployed at uniform intervals along the longitudinal axis of the vehicle (such as the central axis of the frame). For example, an IMU angle sensor is set at a certain distance (this distance can be set according to the vehicle length and monitoring accuracy requirements), so that it can uniformly collect the angle data of each part along the vehicle axis, including pitch angle, roll angle, yaw angle, etc., to reflect the attitude changes of each part of the vehicle in real time. The wheel speed sensors cover all moving joint areas of the vehicle, specifically including the wheel axles of the wheels, the articulated joints of the semi-trailer truck, etc. A wheel speed sensor is installed at the wheel axle of each wheel to monitor the rotational speed and direction of the wheel, and the wheel speed sensor at the articulated joint is used to monitor the movement speed of the joint, so as to ensure the accurate acquisition of the speed data of the whole vehicle and each moving joint.

[0045] In terms of time synchronization processing, the time stamps of multiple sensor data are aligned to generate synchronized data. Since the sampling frequencies and time bases of different types of sensors (such as GPS positioning sensors, IMU angle sensors, and wheel speed sensors) may be different, it is necessary to add time stamps to the data output by each sensor, and then through the method of time stamp matching, align the different sensor data at the same moment, so that the sensor data from different sources are consistent in the time dimension, forming a synchronized data set with a unified time base. Then, the dynamic range of the synchronized data is adjusted, and the data is processed by a linear transformation method. This linear transformation determines the maximum and minimum values of the original data, as well as the preset target interval range (such as [0, 1] or other intervals set according to subsequent processing requirements), and uses a linear function to compress the range of the original data into the target interval. For example, if the minimum value of the original data is and the maximum value is , and the target interval is , then for any original data , its value after linear transformation is:

[0046]

[0047] This processing method can retain the original distribution characteristics of the data while compressing the data range, and avoid affecting subsequent data processing and analysis due to excessive differences in data dimensions.

[0048] The synchronous data is converted into normalized data by setting data thresholds. Normalization processing is to unify sensor data of different types and dimensions to the same dimension, facilitating data fusion and comparative analysis. Specifically, according to the physical meaning and value range of each type of sensor data, the corresponding threshold range is set, and the data exceeding the threshold is truncated or adjusted so that all data is mapped into a standard numerical interval (such as [0,1] or [-1,1]). For example, for the longitude and latitude information in GPS positioning data, the threshold can be set according to the geographical range of vehicle travel, and the data exceeding this range is regarded as invalid or corrected; for the angle data of the IMU angle sensor, the threshold can be set according to the angle range during normal vehicle travel, and the angle value is converted into a normalized numerical value.

[0049] The normalized data is smoothed based on moving average filtering, and the window length of the moving average filtering is dynamically adjusted according to the vehicle movement speed. Moving average filtering is a commonly used signal smoothing method, which reduces noise interference by calculating the average value of the data within the window. When the vehicle movement speed is fast, the sensor data may be affected by more high-frequency noise. At this time, the window length is automatically increased to make more data points participate in the average calculation, thereby enhancing the suppression effect on high-frequency noise; when the vehicle movement speed is slow, the data changes relatively gently. At this time, the window length is reduced to respond to the real-time changes of the data faster and avoid data lag caused by too long a window. For example, when the vehicle is in a high-speed reverse state, the window length can be set to include the last 10 sampling points; when the vehicle is moving slowly at a low speed, the window length can be adjusted to include the last 5 sampling points.

[0050] Perform eigen-decomposition on the smoothed normalized data to separate the normal driving area of the vehicle from the potential folding area. Eigen-decomposition is to perform dimensionality reduction on the data through mathematical methods (such as principal component analysis, etc.) and extract the components that can reflect the main characteristics of the data. By analyzing the smoothed normalized data, identify the characteristic patterns of the vehicle in the normal driving state (such as the angle change within a certain range, the speed changes smoothly, etc.) and the characteristic patterns in the potential folding state (such as the angle suddenly changes greatly, the speed shows abnormal fluctuations, etc.), and divide the data space into the normal driving area and the potential folding area according to these characteristic patterns. For example, through eigen-decomposition, it can be determined that when the change in the articulated angle detected by the IMU angle sensor exceeds a certain threshold and the speed detected by the wheel speed sensor suddenly decreases, the data point belongs to the potential folding area; while when both the angle change and the speed change are within the normal range, the data point belongs to the normal driving area. In this way, a clear data basis is provided for the subsequent data fusion module and the folding state recognition module, enabling the system to more accurately judge the driving state and folding risk of the vehicle.

[0051] Embodiment 2: This embodiment focuses on the composition of the preset safe angle interval and the coordinate conversion process. The preset safe angle interval is composed of 6 reference anchor points, where 2 anchor points are set in the front-end area of the monitoring unit, and the other 4 anchor points are set in the rear-end area of the monitoring unit. The two points in the front-end area are located at the angle peak points of the axial distribution. This axial distribution refers to the angle change distribution along the central axis direction of the monitoring unit (such as the axis where the articulated point of the semi-trailer truck is located). The angle peak points are determined by analyzing the possible angle extreme values that may occur at the front end of the monitoring unit during the reverse process. For example, when the vehicle is turning or reversing, the front end may generate the maximum deviation angle due to inertia or steering operation. These two peak points respectively correspond to the maximum positive and negative angles in the axial distribution, which can reflect the extreme situation of the angle change at the front end of the monitoring unit in the axial direction and serve as the key reference points for judging angle abnormalities in the front-end area.

[0052] Two anchor points in the rear area are located at the intersection of the unit boundary line and the movement path, which is the angle change point. The unit boundary line is a boundary pre-determined according to the physical structure and movement range of the monitoring unit, such as a virtual boundary line extending from the hinge point to both sides; the movement path is the actual movement trajectory of the monitoring unit tracked in real time through sensor data. When the movement trajectory of the monitoring unit intersects with the unit boundary line, it indicates that the angle at the rear end of the monitoring unit begins to change significantly, and this intersection point is the starting point or turning point of the angle change, which can capture the key position of the angle change at the rear end. The remaining two points in the rear area are located at the quadratic uniform distribution nodes of the connection line of the angle change points. Specifically, first connect the two angle change points with a straight line to form a reference line, and then perform quadratic uniform division on this reference line, that is, set anchor points at the midpoint of the reference line and the midpoints between the midpoint and the two end points respectively, so that these two newly added anchor points divide the reference line into three equal segments. Through this quadratic uniform distribution setting, the details of the angle change can be more carefully captured in the area between the angle change points, avoiding missing the angle anomalies in the middle area due to sparse anchor point distribution, enabling the preset safe angle interval to comprehensively and accurately cover the angle change range at the rear end of the monitoring unit, and providing a more accurate reference basis for the density calculation and risk judgment of the path trajectory in the subsequent folding state recognition.

[0053] When performing coordinate transformation on the sensor data, the boundary contour line of the vehicle's normal driving area is obtained. This boundary contour line is the position boundary of each part of the vehicle in the normal reverse driving state obtained through statistical analysis of historical sensor data. For example, during the normal reverse driving of a semi-trailer truck without folding risk, the position ranges of each monitoring unit on the vehicle body form a closed contour. The boundary contour line can be generated by data fitting methods, such as using polygon fitting or curve fitting methods to integrate the position data of each monitoring point during normal driving to form a boundary curve or broken line that can represent the normal driving area.

[0054] Calculate the azimuth deviation angles between the three main axes of the boundary and the sensor data coordinate system respectively. The three main axes are selected as the main axes that can represent the overall structure and movement direction of the vehicle, usually including the longitudinal central axis of the vehicle (along the vehicle body length direction), the transverse axis (the horizontal axis perpendicular to the vehicle body length direction), and the vertical axis (the axis perpendicular to the ground). The sensor data coordinate system refers to the coordinate system adopted by the output data of each sensor itself, and there may be differences in the coordinate systems of different sensors (for example, the GPS positioning sensor adopts the geographic coordinate system, and the IMU angle sensor adopts the vehicle body coordinate system). The azimuth deviation angle refers to the angle difference between the actual direction and the theoretical direction of the main axis in the sensor data coordinate system. For example, the longitudinal central axis of the vehicle should theoretically be consistent with the positive direction of the x-axis of the sensor data coordinate system. If there is an included angle θ between the actually detected central axis direction and the positive direction of the x-axis, then θ is the azimuth deviation angle of the longitudinal central axis. By calculating the azimuth deviation angles of the three main axes respectively, the angular difference between the sensor data coordinate system and the actual coordinate system of the vehicle can be comprehensively reflected.

[0055] [[ID=,3]]Calibrate the sensor data in the spatial coordinate system according to the weighted average of the azimuth deviation angles. The calculation method of the weighted average is to assign different weights to the azimuth deviation angles of the three main axes. The setting of the weights is based on the importance degree of each main axis in the vehicle movement. For example, the longitudinal central axis has the greatest impact on the vehicle's reverse direction and can be assigned a higher weight, while the weights of the transverse axis and the vertical axis are relatively low. When calculating the weighted average, multiply the azimuth deviation angle of each main axis by its corresponding weight and then sum to obtain the comprehensive azimuth deviation angle. Then use this comprehensive deviation angle to perform coordinate transformation on the sensor data. For example, through the rotation matrix, the sensor data is converted from the original coordinate system to the coordinate system based on the actual main axis of the vehicle, so that the calibrated sensor data can more accurately reflect the actual spatial position and attitude of the vehicle. For example, for the GPS positioning data, after coordinate calibration, its coordinate values can exactly correspond to the position of the vehicle in the actual geographical space; for the IMU angle sensor data, after calibration, the angle deviation caused by the sensor installation position or coordinate system difference can be eliminated, and the angle data can truly reflect the actual attitude changes of each part of the vehicle. Through this coordinate transformation process, ensure that the data of different sensors are fused and analyzed in a unified coordinate system, improve the consistency and reliability of the data, and lay a foundation for the subsequent generation of the fusion state vector and the folding state recognition based on the vehicle geometric model.

[0056] Example 3: This example elaborates in detail the process of detecting the span of motion characteristics and judging the real-time folding state. When detecting the longitudinal or transverse distribution span of motion characteristics, traverse the data area according to the detection direction, where the detection direction is the longitudinal section or the transverse section. The longitudinal section refers to the vertical section along the vehicle length direction (such as from the head to the tail of a semi-trailer truck), and the transverse section refers to the vertical section along the vehicle width direction (such as from the left and right sides of the vehicle body). By comprehensively traversing the data area from these two dimensions respectively, the angular change conditions of various parts of the vehicle can be covered. Statistically analyze the angular change values of each detected data group. Among them, the angular value of the potential folding area is 180 (the theoretical angle corresponding to the fully folded state of the vehicle), and the angular value of the normal driving area is 0 (the theoretical angle corresponding to the non-folded state of the vehicle moving straight). These are used as the benchmark thresholds for judging whether the angle is in an abnormal state. In actual detection, the thresholds can be fine-tuned according to the vehicle structure and driving characteristics.

[0057] Select the detection group where the first angular value breaks through the normal driving threshold and mark it as the risk starting group. The normal driving threshold can be set to a small range close to 0 (such as ±5°). When the angular value of a certain detection group exceeds this range, it indicates that the vehicle starts to have an angular deviation that may lead to folding. For example, in the longitudinal section detection, if the angular value of a certain detection group suddenly increases from 0° to 10°, then this group is marked as the risk starting group and regarded as the potential starting point of the folding risk.

[0058] Continue to traverse the data area. When a detection group with an angular value returning to the normal driving threshold is detected, mark it as the pending group. At this time, the angular value temporarily returns to the normal range, but it is necessary to further judge whether this recovery is a real termination point rather than a short-term fluctuation. Obtain the spatial interval D1 between the pending group and the starting group. The calculation of the spatial interval D1 is based on the coordinate system of the detection direction. For example, in the longitudinal section, calculate the longitudinal distance between the two groups based on the vehicle longitudinal axis, and in the transverse section, calculate the transverse distance based on the vehicle transverse axis. If the spatial interval D1 is greater than or equal to the preset span P (the preset span P is set according to the vehicle size and safety standards, such as 2 meters), and there is no angular abnormality when continuing to detect along the detection direction, then mark this pending group as the termination group, indicating that this section of angular change is completed within a reasonable spatial range and belongs to the uniform distribution characteristic, that is, the angular change is continuous and gradually recovers without forming sudden changes or abnormal aggregations.

[0059] If the spatial interval D1 is less than the preset span P, it indicates that the angle change is restored within a short space, and there may be a risk of abnormal fluctuations. It is necessary to continue traversing this data area. When detecting a detection group where the angle breaks through the normal driving threshold again, it is marked as the second starting group, and the interval D2 between the second starting group and the pending group is obtained. If the interval D2 is greater than the preset span Q (the preset span Q can be set according to the actual situation, usually less than the preset span P, such as 1 meter), it is judged that the motion distribution is abnormal, indicating that there are multiple abnormal angle breakthroughs within a short distance, which may indicate that the vehicle motion trajectory is unstable and there is a high folding risk.

[0060] When judging whether the real-time performance of the folding state meets the standard, for any selected monitoring unit, the reference axis where the center point of the unit is located is selected. This axis is usually the geometric center axis of the monitoring unit (such as the longitudinal axis where the hinge point is located) as the reference reference line for measuring the angle change of the unit. Obtain the offset of all angle points in this data area relative to this axis. The calculation method of the offset is the absolute value of the angle difference between the angle point and the reference axis. For example, if the angle value of a certain angle point is 30° and the angle of the reference axis is 0°, the offset is 30°.

[0061] Adjust the offset of each angle point through a preset scaling factor. The scaling factor is used to magnify or reduce the scale of the offset to facilitate observing the data correlation between adjacent monitoring units. Continuously increase the value of the scaling factor until the adjacent monitoring units all form a data superposition area. The data superposition area refers to the overlap of the angle offset distribution ranges of adjacent monitoring units, indicating that the angle changes of the two are continuous and related. For example, when the scaling factor gradually increases from 1 to 3, the angle offset distribution ranges of two originally separated monitoring units begin to overlap. At this time, a continuous trajectory of the state change can be obtained, which intuitively shows the coherence and trend of the angle change during the vehicle's reverse.

[0062] Obtain the corresponding fitting curve equation for the trajectory through median filtering. Median filtering is a non-linear filtering method. By sorting the data values within the window and taking the median, it can effectively remove noise interference, especially impulse noise. For example, for trajectory data containing abnormal jump angle values, median filtering can replace them with the median of adjacent data to make the trajectory smoother. The fitting curve equation can adopt methods such as polynomial fitting and linear fitting, and select a suitable fitting method according to the shape of the trajectory. For example, when the trajectory shows an approximate linear trend, linear fitting is adopted, and when the trajectory has obvious curvature, quadratic or higher-order polynomial fitting is adopted.

[0063] Obtain the slope change rate of this curve equation. The slope change rate reflects the change trend of the angle change rate. For example, if the fitting curve is a linear function , its slope is a constant, and the slope change rate is 0, indicating a constant angular change rate; if the fitted curve is a quadratic function , its slope is , and the slope change rate is , indicating that the angular change rate changes linearly with time or space. When the deviation of the slope change rate from the standard value exceeds the preset range, it is determined that the real-time performance of the folding state of this dimension does not meet the standard, indicating that the real-time monitoring data of the vehicle's angular change fails to accurately reflect the actual state, and there may be data lag or abnormal fluctuations, and early warnings or adjustment of monitoring parameters are required. Through this process, the system can accurately extract the dynamic characteristics of angular changes from the real-time monitoring data, providing a scientific basis for the real-time assessment of folding risks.

[0064] Embodiment 4: This embodiment describes further details of the path trajectory generation and data acquisition module. When generating the path trajectory, first select the three reference points closest to the center point of the monitoring unit. The center point of the monitoring unit is the central position determined according to the geometric shape of the monitoring unit (such as the intersection of the diagonals of a rectangular monitoring unit), and the reference points are the points marked with the starting angle value and the ending angle value of each monitoring point calculated based on the fusion state vector. By calculating the Euclidean distance (i.e., the spatial straight-line distance) from each reference point to the center point, the three reference points with the smallest distance are selected. These three points serve as the initial key monitoring points, which can most directly reflect the angular change characteristics in the central area of the monitoring unit and provide a basic reference for the path trajectory.

[0065] Associate adjacent reference points to generate an initial path structure. The determination of adjacent reference points is based on spatial proximity. That is, after sorting by distance, the two points with the closest distance among the three reference points are first connected to form a line segment of the initial path, and then the third point is connected to the point with the closer distance among the two already connected points to form an initial triangular or polyline structure containing three line segments as the basic framework of the path trajectory.

[0066] Then, select the point with the closest distance among the remaining reference points one by one and add it to the association of the path structure. For each reference point not yet added to the path structure, calculate its distance to all endpoints in the current path structure, and select the endpoint with the smallest distance for connection to incorporate this reference point into the path structure. For example, if the endpoints of the current path structure are A and B, and the distance between the remaining reference point C and A is less than the distance between C and B, then connect C to A to form a new path segment AC, and at this time the path structure is extended to a polyline containing A, B, and C.

[0067] During the process of adding a new reference point, continuously determine whether the length of the connection path corresponding to the original reference point increases in the associated path formed by the newly added reference point. Specifically, when the new reference point D is added to the path structure and connected to the endpoint C, check whether the path length from the endpoint B connected to C in the original path to C changes due to the addition of D (such as whether the length of the BC segment changes due to recalculating the path order). If the length of the connection path corresponding to the original reference point increases, it indicates that the addition of the new reference point causes the path structure to deviate from the optimal connection method, possibly introducing redundancy or unreasonable turns, then remove the newly added point and maintain the original path structure; if the path length does not increase or decreases, retain the newly added point and continue to expand the path.

[0068] After the above screening and adjustment process, the associated path finally formed by the original reference points is the path trajectory of the monitoring unit. By gradually optimizing the connection order and removing invalid points, this trajectory can accurately reflect the spatial association and angular change trend between the reference points within the monitoring unit, such as presenting a smooth curve or a regular broken line, providing accurate path data for calculating the density of the path trajectory covering the preset safe angle interval in the subsequent process.

[0069] In the data acquisition module, GPS positioning sensors are arranged in a grid structure at key positions of the vehicle. The design of the grid structure is based on the body layout of the semi-trailer truck, forming crisscross grid nodes at key positions such as the top of the cab in the front of the vehicle, the middle of the vehicle chassis, and the four corners of the cargo box at the rear of the vehicle. A GPS positioning sensor is deployed at each node. For example, in the front area, a 5×5 grid is formed at an interval of 20 cm, in the middle of the vehicle body, a 10×3 grid is formed at an interval of 50 cm, and in the rear area, an 8×4 grid is formed at an interval of 30 cm, ensuring that the position data of all parts of the vehicle body are effectively collected. Each sensor transmits data to the central processing unit in real time through a wireless communication module, forming a position monitoring network covering the whole vehicle, and can accurately obtain the coordinates and displacement changes of each part of the vehicle in the geographical space.

[0070] IMU angle sensors are axially distributed at equal intervals. Starting from the front hinge point along the longitudinal central axis of the vehicle (i.e., the center line of the frame), a sensor is set every 30 cm in the direction towards the rear of the vehicle until the end of the rear of the vehicle. For the tractor and trailer parts of the semi-trailer truck, sensors are evenly deployed on the central axes of the tractor frame and the trailer frame respectively. For example, 5 sensors are deployed on the tractor part and 8 sensors are deployed on the trailer part. Each IMU angle sensor real-time collects the pitch angle, roll angle, and yaw angle data at its location, and through the collaborative work of a three-axis gyroscope and a three-axis accelerometer, accurately measures the attitude changes of each part of the vehicle, especially the angular fluctuations near the hinge point, providing key angle data support for the recognition of the folded state.

[0071] The wheel speed sensors cover all the moving joint areas of the vehicle, including the front wheel axle and the rear wheel axle of the tractor, the wheel axles of the trailer, and the articulation joint between the tractor and the trailer. A wheel speed sensor is installed at each wheel axle to monitor the rotational speed and the number of rotations of the wheel through electromagnetic induction or photoelectric coding, and calculate the traveling speed and the driving distance of the vehicle; the wheel speed sensor at the articulation joint adopts a rotary encoder structure and is installed on the articulation shaft to monitor the rotational speed and the angle change of the joint in real time, ensuring the comprehensive acquisition of the vehicle's movement speed, especially the abnormal speed (such as sudden acceleration or deceleration) at the joint part can be captured in time, providing data support for judging the vehicle's movement state and folding risk in terms of speed dimension.

[0072] Through the collaborative work of the GPS positioning sensor, the IMU angle sensor and the wheel speed sensor, the data acquisition module can synchronously obtain the position, angle and speed data of the vehicle, forming a multi-dimensional real-time monitoring data set. For example, during the backing process of a semi-trailer truck, the GPS positioning sensor tracks the position movement of each part of the vehicle body in real time, the IMU angle sensor monitors the angle change of the articulation point, and the wheel speed sensor records the movement speed of the wheels and joints. After these data are time-synchronized, they provide comprehensive and accurate raw data for the data fusion module to generate a fusion state vector, ensuring that the entire monitoring system can perform folding state recognition and real-time monitoring based on a reliable data basis.

[0073] Embodiment 5: This embodiment relates to the specific content of the dynamic range adjustment, filtering processing and real-time monitoring module. During the dynamic range adjustment process, a linear transformation method is used to process the synchronous data. This processing is based on the numerical distribution characteristics of the original sensor data. First, the maximum value and the minimum value of the original data are determined to form the original range of the data. For example, the original angle data range output by a certain IMU angle sensor is [-30°, 50°], its maximum value is 50°, the minimum value is -30°, and the original range is 80°. Subsequently, a target interval is set according to the requirements of subsequent data processing. For example, the target interval is set to [0, 1], and the original data is mapped into this interval through a linear transformation formula. The specific transformation formula is: , substituting the above example data, the original value -30° corresponds to the normalized value 0, 50° corresponds to the normalized value 1, and the intermediate value such as 0° corresponds to the normalized value 0.375. This linear transformation method can keep the relative magnitude relationship and the distribution form of the original data unchanged, only compress the numerical range of the data, and avoid deviation during fusion due to the too large difference in the data dimensions of different sensors.

[0074] The window length of the moving average filter is dynamically adjusted according to the vehicle's motion speed. The vehicle's motion speed is calculated from the wheel speed data collected in real time by the wheel speed sensor. For example, when the wheel speed sensor detects that the wheel speed is 100 revolutions per minute, the actual driving speed of the vehicle can be calculated as 5 meters per second by combining the wheel diameter. When the vehicle is in a high-speed motion state (such as a speed greater than or equal to 3 meters per second), the sensor data is easily interfered by high-frequency noise. At this time, the window length of the moving average filter is automatically increased. For example, the window length is set to include the last 15 sampling points, and the noise is smoothed by averaging more data points. When the vehicle is moving at a low speed (such as a speed less than 3 meters per second), the data changes relatively slowly and the noise has little impact. At this time, the window length is reduced to include the last 5 sampling points to improve the real-time performance of data processing and ensure that the subtle changes in angle and speed are reflected in a timely manner. The specific calculation method of the moving average filter is as follows: for the current sampling point, take the data values of several points before and after it (the number of points is determined by the window length) and calculate the arithmetic mean as the filtered value of this point. For example, when the window length is 5, the filtered value of the nth point is the average of the original data of the n - 2, n - 1, n, n + 1, and n + 2 points.

[0075] The real-time monitoring module is built-in with a convolutional neural network (CNN) model, which is trained with historical vehicle folding data. The historical data includes multi-sensor data of the vehicle in different reverse scenarios, as well as the corresponding annotation results of whether folding occurs. During the training process, the rate of change of the slope of the state change curve is used as the input feature. The state change curve is generated by the real-time monitoring module. Specifically, an arbitrary one-dimensional fusion state vector sequence (such as the vector sequence in the angle dimension) is selected, and the data region of each monitoring unit is longitudinally extended based on the axis where the center point of the unit is located until the data regions of adjacent monitoring units form a continuous motion trajectory, thereby generating the state change curve in this dimension. The rate of change of the slope is obtained by calculating the derivative of the curve fitting equation. For example, for the quadratic curve fitting equation , its slope is , and the rate of change of the slope is , which reflects the changing trend of the angle change rate.

[0076] The model takes the probability of folding as the output result, and the output value range is [0,1]. The larger the value, the higher the possibility of folding. During model training, an adaptive learning rate strategy is used to optimize the loss function. Common adaptive learning rate algorithms such as the Adam algorithm, whose core idea is to automatically adjust the learning rate according to the gradient change of the parameters during the training process. At the beginning of training, the gradient is large, and the learning rate is set large to accelerate the convergence speed. As the training progresses, the gradient gradually decreases, and the learning rate automatically decays to avoid oscillations and improve the training accuracy. The loss function uses the cross-entropy loss function, and the calculation formula is:

[0077] ,

[0078] where is the true label (0 means unfolded, 1 means folded), is the folding probability predicted by the model. [[ID={10]]

[0079] In terms of model architecture design, a convolutional neural network usually consists of several convolutional layers, pooling layers and fully connected layers. The convolutional layer extracts the spatial local correlation in the input features through a convolutional kernel, for example, extracts local trend features from the sequence of slope change rates of the state change curve; the pooling layer downsamples the feature map to reduce the number of parameters and improve the robustness of the model; the fully connected layer integrates the extracted features and outputs the final folding occurrence probability. For example, a typical CNN architecture may include 3 convolutional layers (each convolutional kernel size is 3×1, stride is 1), 2 max pooling layers (pooling window size is 2×1) and 2 fully connected layers (the number of neurons is 128 and 1 respectively).

[0080] The working process of the real-time monitoring module is as follows: First, obtain the fused state vector sequence from the data fusion module and select the data of one dimension (such as the angle or speed dimension); then, longitudinally expand the data area of each monitoring unit along the axis where the unit center point is located, and by adjusting the expansion amplitude (such as expanding 5° angle range to both sides centered on the axis), make the data areas of adjacent monitoring units overlap to form a continuous motion trajectory; then, perform curve fitting on the trajectory (such as cubic spline fitting) and calculate the slope change rate of the fitting curve; finally, input the slope change rate into the convolutional neural network model and output the folding occurrence probability in the current state. If the probability value exceeds the preset threshold (such as 0.8), the warning mechanism is triggered to send a folding risk prompt to the driver.

[0081] Through preprocessing such as dynamic range adjustment and moving average filtering, it is ensured that the data input into the model has a unified dimension and a low noise level; the convolutional neural network model can capture the implicit patterns related to folding from complex multi-sensor data through automatic feature extraction and adaptive learning, providing real-time and accurate prediction for the folding risk during the backing process of the semi-trailer truck, assisting the driver to adjust the operation in time and reducing the accident rate. The whole process does not rely on manually set rules, but realizes intelligent monitoring in a data-driven manner, improving the generalization ability and adaptability of the system.

[0082] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0083] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time monitoring system for the reversing and folding of semi-trailers based on multi-sensor fusion, characterized by: include: A data acquisition module is used to obtain multiple sensor data of the semi-trailer, including position data, angle data and speed data, and perform time synchronization processing on the sensor data to divide the vehicle into a normal driving area and a potential collapse area; a data fusion module, configured to generate a fusion state vector based on the sensor data and perform coordinate transformation according to a vehicle geometric model; The folding state recognition module is used to select any monitoring unit of the vehicle hinge point, calculate the starting angle value and ending angle value of each monitoring point based on the fusion state vector, mark them as reference points, obtain several reference points closest to the geometric center point of the monitoring unit, define them as key monitoring points, associate adjacent key monitoring points in sequence to generate a path trajectory, obtain the density of the path trajectory covering the preset safety angle range, and when the density is less than a preset threshold, identify whether it is a linear distribution or a curved distribution motion feature. If not, it is determined that the angle change of the monitoring unit is abnormal; if so, the span of the longitudinal or lateral distribution of the motion feature is detected. If the span is greater than the preset span value, it is determined that the unit has a high risk of folding; The real-time monitoring module is used to select any one-dimensional fusion state vector sequence, expand the data area of each monitoring unit longitudinally based on the axis where the unit center point is located, until the data areas in adjacent monitoring units form a continuous motion trajectory, generate a continuous state change curve, and calculate the slope change rate of the curve fitting equation. When the deviation between the change rate and the standard value exceeds a preset range, it is judged that the real-time performance of the folding state of this dimension does not meet the standard.

2. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The process of time synchronization is as follows: The multiple sensor data are timestamp aligned to generate synchronized data, the synchronized data are dynamically range adjusted, and the synchronized data are converted into normalized data by setting a data threshold. The normalized data are smoothed based on a moving average filter, and feature decomposition is performed on the smoothed normalized data to separate the normal driving area of the vehicle from the potential folding area.

3. The real-time monitoring system for reverse folding of semi-trailer vehicles based on multi-sensor fusion according to claim 1 is characterized in that: The preset safety angle interval is composed of 6 reference anchor points, of which 2 anchor points are set in the front end area of the monitoring unit, and the other 4 anchor points are set in the rear end area of the monitoring unit. The two points in the front end area are positioned at the angle peak points of the axial distribution; the two anchor points in the rear end area are positioned at the intersection of the unit boundary line and the motion path, that is, the angle change point, and the remaining two points in the rear end area are positioned at the secondary uniformly distributed nodes of the line connecting the angle change points.

4. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The process of coordinate transformation of sensor data is: The boundary contour line of the normal driving area of the vehicle is obtained, the azimuth deviation angles between the three main axes of the boundary and the sensor data coordinate system are calculated respectively, and the sensor data is calibrated in the spatial coordinate system according to the weighted average of the azimuth deviation angles.

5. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The process of detecting the vertical or horizontal distribution span of motion features is: The data area is traversed according to the detection direction, which is a longitudinal section or a transverse section. The angle change value of each group of data detected is counted. The angle value of the potential folding area is 180, and the angle value of the normal driving area is 0. The first detection group whose angle value exceeds the normal driving threshold is selected and marked as the risk starting group. The data area is continued to be traversed. When a detection group whose angle value returns to the normal driving threshold is detected, it is marked as a pending group. The spatial interval D1 between the pending group and the starting group is obtained. If the spatial interval D1 is greater than or equal to the preset span P, and no angle abnormality is found when continuing to detect along the detection direction, the pending group is marked as the termination group, and the motion is characterized by uniform distribution. If the spatial interval is less than D1, the data area is continued to be traversed. When a detection group whose angle exceeds the normal driving threshold is detected again, it is marked as the second starting group. The interval D2 between the second starting group and the pending group is obtained. If the interval D2 is greater than the preset span Q, it is determined that the motion distribution is abnormal.

6. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 5 is characterized in that: The specific process of judging whether the folding state is not up to standard in real time is as follows: For any selected monitoring unit, select the reference axis where the center point of the unit is located, obtain the offset of all angle points in the data area relative to the axis, adjust the offset of each angle point by a preset scaling factor, and continuously increase the value of the scaling factor until adjacent monitoring units form a data superposition area, obtain a continuous trajectory of state changes, obtain the corresponding fitting curve equation through median filtering of the trajectory, and obtain the slope change rate of the curve equation.

7. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The process of generating path trajectory is: First, the three reference points closest to the center point of the monitoring unit are selected, and the adjacent reference points are associated to generate an initial path structure. Then, the nearest points among the remaining reference points are selected one by one and added to the association of the path structure. When the length of the connection path corresponding to the original reference point in the associated path formed by the newly added reference point increases, the newly added point is removed, and the associated path formed by the original reference point is the path trajectory of the monitoring unit.

8. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The data acquisition module includes a GPS positioning sensor, an IMU angle sensor and a wheel speed sensor, wherein the GPS positioning sensor is arranged in a grid structure at key positions of the vehicle, the IMU angle sensor is distributed axially at equal intervals, and the wheel speed sensor covers all moving joint areas of the vehicle.

9. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 2 is characterized in that: The dynamic range adjustment adopts a linear transformation method to compress the range of the original data to a preset range while retaining the data distribution characteristics. The window length of the moving average filter is dynamically adjusted according to the vehicle movement speed.

10. The real-time monitoring system for semi-trailer reversing and folding based on multi-sensor fusion according to claim 1 is characterized in that: The real-time monitoring module has a built-in convolutional neural network model, which is trained using historical vehicle folding data, using the slope change rate of the state change curve as an input feature and the probability of folding as an output result. During model training, an adaptive learning rate strategy is used to optimize the loss function.

Images