A real-time monitoring method and device for increasing the safety of getting off a vehicle

Abstract

The present invention relates to a real-time monitoring method and device for increasing the safety of getting off a vehicle, comprising the following steps: using a collection device installed on a vehicle to capture the dynamic information of the position, speed, acceleration and other dynamic information of surrounding obstacles in real time, and outputting a safety risk score based on the dynamic information based on the dynamic information of the obstacles. Integrate the dynamic information of obstacles within a period of time into a time series, and predict the future steering angle based on it, adjust the original safety risk score according to the prediction result and output a new full risk score; collect environmental factor data such as weather, lighting, road conditions, etc., adjust the original risk score according to the environmental information and output a new full risk score, and weight the two output new full risk scores to calculate the overall safety score. When the vehicle stops, calculate the overall safety score in real time, and judge the current warning level according to the preset threshold set, and issue a corresponding warning signal.

Description

Technical Field

[0001] The present invention relates to the field of safety detection technology, and more specifically to a real-time monitoring method and device for increasing the safety of getting off a vehicle. Background Art

[0002] In the existing technology, automobiles are one of the main means of transportation for modern transportation, and the acceleration of intelligent transportation has promoted the revolutionary development of transportation. Intelligent transportation can maximize the efficiency and safety of automobile transportation. The continuous improvement of the intelligence level of automobiles is one of the important factors, and strengthening the exchange of information between automobiles and the external environment is also an indispensable factor. According to statistics from the Traffic Management Bureau of the Ministry of Public Security, the number of motor vehicles and motor vehicle drivers in the country will continue to grow rapidly in 2017. With the increase in the number of motor vehicles and the frequency of use, the potential safety hazards also increase. For example, some drivers and passengers do not have good riding habits. When parking the car, they forget to detect whether there are vehicles or pedestrians approaching from behind, causing the door to collide and scratch with the vehicle that quickly approaches the door at the moment of opening, especially motorcycles and electric bicycles, which are often hit by the door and fall to the ground and are hit and crushed by other passing cars, causing serious secondary injuries to the victims and causing tragic traffic accidents.

[0003] Therefore, the market demands a safe door opening warning and anti-collision system to reduce the occurrence of accidents. Summary of the invention

[0004] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, the present invention provides a real-time monitoring method for increasing the safety of getting off a vehicle, comprising the following steps:

[0005] Step S1: Obstacle dynamic information safety assessment

[0006] Collect data information of multiple obstacles and convert them into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , the obstacle dynamic information O t is a matrix formed by multiple obstacle feature vectors. Each row in the matrix represents a feature vector of an obstacle. O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ;

[0007] Step S2: Score adjustment based on obstacle turn prediction

[0008] Collect dynamic information of obstacles over a period of time t , and integrate them into a dynamic information sequence of obstacles, input the dynamic information sequence of obstacles into the preset LSTM model, and the LSTM model outputs the prediction of the steering angle

[0009] The safety risk score R calculated from the obstacle dynamic information is adjusted by the correction function g O , and output the adjusted security risk score R' O , the correction function g is based on the steering prediction Adjust the security risk score R O ;

[0010] Step S3: Score adjustment based on environmental information

[0011] Collect environmental data and convert it into environmental information t , to represent the environmental factors at time t, and use the correction function h to adjust the safety risk score R calculated from the obstacle dynamic information O , and output the adjusted security risk score R" O , the correction function h is based on the environmental information E t Adjust the security risk score R O ;

[0012] Step S4: Overall safety risk assessment

[0013] The safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ;

[0014] Step S5: Security risk classification

[0015] The generated multiple overall safety scores are integrated into a scoring data set. The K-means clustering algorithm is used to perform cluster analysis on the scoring data set. The number of clusters n is selected as the number of levels predetermined by the early warning system. The n cluster centers are found through the clustering algorithm. The overall safety score is divided into n segments through the n cluster centers. That is, the threshold set is determined according to the cluster center or cluster boundary. Each threshold λ i Indicates the boundary that distinguishes different security risk levels.

[0016] Step S6: Security risk assessment

[0017] Obtain the vehicle speed data and dynamic information of obstacles in the preset area. If the vehicle stops, the overall safety score R is calculated in real time. t , and by presetting the function L(R t ) determines the current warning level and issues a corresponding warning signal according to the warning level, where the function L(R t ) By comparing R t and the elements in the threshold set Λ to determine the warning level.

[0018] Furthermore, in step S1,

[0019] The obstacle feature vector includes the position (x t ,y t )、speed(v x,t ,v y,t ) and acceleration (a x,t ,a y,t ); The obstacle dynamic information O t Expressed as: t =[x t ,y t ,v x,t ,v y,t ,a x,t ,a y,t ] T ;

[0020] Evaluation function f O It is expressed as:

[0021] R O =β0+β1x t +β2y t +β3v x,t +β4v y,t +β5a x,t +β6a y,t

[0022] Among them, R O It is a safety risk score based on dynamic information of obstacles;

[0023] β0, β1, β2...,β6 are model parameters;

[0024] The calculation method of the model parameters is:

[0025] The obstacle dynamic information O t Convert to design matrix D, and convert multiple design matrices D and R O Integrate into training data where d i is the feature vector of the i-th sample, R o,i is the corresponding security risk score; construct the cost function J(β), and obtain the optimal model parameters by minimizing the cost function

[0026] Further, in step S2,

[0027] The obstacle dynamic information sequence is represented as X = {O t-n ,...,O t-1 ,O t}; The prediction function of the obstacle turning angle is expressed as:

[0028] in is the obstacle turning angle predicted at time t+Δt

[0029] The correction function g is expressed as:

[0030] Among them, R O Provide safety risk scores based on dynamic obstacle information;

[0031] is the steering angle at the future time point t+Δt;

[0032] γ is an adjustment parameter used to adjust the sensitivity of the impact of the predicted steering angle on the safety risk score;

[0033] The adjustment parameter γ is calculated by collecting and analyzing the statistical relationship between the predicted steering angle change and the accident rate.

[0034] Furthermore, in step S3, the environmental information is represented as E t =[e 1,t ,e 2,t ,e 3,t ,...,e m,t ] T , where e i,t Representing different environmental factors, the correction function h is expressed as:

[0035] R” O =R O (1+α1*e1+α2*e2+...+α m *e m )

[0036] Among them, R O Provide safety risk scores based on dynamic obstacle information;

[0037] is the steering angle at the future time point t+Δt; each α i Indicates different environmental factors i The intensity of the impact on the security risk score;

[0038] By collecting and analyzing environmental factors and security incident data, statistical methods are used to calculate each environmental factor e i The influence coefficient α i .

[0039] Furthermore, in step S4, the overall safety score is expressed as: R t =w1R' O +w2R” O , where w1 and w2 are R'O , R” O The security weight of w1 and w2 is obtained as follows:

[0040] Collect multiple samples, each sample includes a security risk score R' O , R” O , integrate multiple samples into a data set where are the safety risk scores of the i-th sample based on obstacle steering prediction and environmental information, y i is the corresponding safety status indicator;

[0041] By minimizing the objective function, the overall safety score after weighted fusion reflects the actual safety status, where the objective function is expressed as:

[0042] Among them, w1, w2 ≥ 0, w1 + w2 = 1;

[0043] The objective function is iterated multiple times according to the gradient algorithm. In each iteration, the weight is updated according to the gradient of the objective function with respect to the weight, which is expressed as:

[0044] Where μ is the learning rate, Represents the derivative of the loss function with respect to the weight w1;

[0045] Through iterative updates, the weight is determined until the difference between the overall security score and the actual security status is minimized.

[0046] Further, in step S5, the scoring dataset is represented as {R 1,t ,R 2,t ,R 3,t ,...,R M,t}, the steps to perform cluster analysis on the scoring data set using the K-means clustering algorithm are:

[0047] Initialization: Randomly select n overall security risk scores as the initial cluster centers {c1, c2, c3, ..., c n};

[0048] Cluster assignment: For each overall security risk score R i,t , calculate its relationship with each cluster center c j The distance R i,t Assign it to the cluster represented by the nearest cluster center, that is, for each overall security risk score R i,t , assigned to cluster {k=argmin||R i,t -c j || 2};

[0049] Update cluster centers: Recalculate the center of each cluster to obtain the average of all overall security risk scores in each cluster, and the new cluster center c′ j The calculation method is:

[0050] Repeat the steps of cluster assignment and updating cluster centers until the cluster centers no longer change significantly;

[0051] After clustering is completed, the midpoint between two adjacent cluster centers is used as the split point to define the threshold of the security risk level and form a set Λ={λ1,λ2,...,λ n}; through the threshold {λ1,λ2,...,λ n The overall security risk score data is divided into n segments, each segment corresponding to a security risk level.

[0052] Furthermore, in step S6, when the vehicle stops and is shifted to P gear or the driving speed exceeds 50 and then drops to 0, the overall safety risk score of the effective area for door opening is obtained according to the dynamic information of the obstacle, and a determination is made whether to send a restriction signal to the door restriction device according to the result of the overall safety risk score, and a warning message is sent to a mobile phone and an in-vehicle display module, wherein the warning message includes a warning level; the effective area for door opening includes a left area and a right area.

[0053] The present invention also provides a real-time monitoring method and device for improving the safety of getting off a vehicle. Based on the above-mentioned real-time monitoring method for improving the safety of getting off a vehicle, the method is characterized by comprising the following modules:

[0054] Camera module: set in the middle of the rear windshield, used to obtain image information of a preset area and send the image information to the display module;

[0055] A radar module, used to collect motion information of moving objects in a preset area and send the motion information to the processor module in real time, wherein the motion information includes distance information, speed information and direction information;

[0056] A wireless connection module is used to connect to the mobile phone of the occupants of the vehicle and transmit warning information and image information to the mobile phone;

[0057] The processor module obtains the vehicle speed data and the dynamic information of obstacles in the preset area. If it is detected that the vehicle stops, the warning level is obtained in real time and a corresponding warning signal is issued according to the warning level;

[0058] The display module includes a touch screen disposed in the armrest box, and the touch screen displays image information acquired by the camera when receiving a switching signal;

[0059] The door restriction device is used to control the opening amplitude of the opened door according to the restriction signal.

[0060] Furthermore, a server module is also included, and the server module includes:

[0061] Obstacle dynamic information scoring unit: used to collect data information of multiple obstacles and convert them into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , by evaluating the function f O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ;

[0062] The first adjustment unit: collects dynamic information of obstacles within a period of time. t , and integrate them into a dynamic obstacle information sequence, which is input into the preset LSTM model. The LSTM model outputs the prediction of the steering angle, and adjusts the safety risk score R calculated from the dynamic obstacle information based on the steering prediction. O , and output the adjusted security risk score R' O ;

[0063] The first adjustment unit: collects environmental information and adjusts the safety risk score R calculated from the dynamic information of obstacles based on the environmental information. O , and output the adjusted security risk score R" O ;

[0064] Safety risk overall scoring unit: the safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ;

[0065] Security risk grading unit: Integrate the generated multiple overall security scores into a scoring data set, and perform cluster analysis on the scoring data set using the K-means clustering algorithm.

[0066] The beneficial effects of the present invention are as follows: in view of the deficiencies of the prior art, the present invention relates to a real-time monitoring method and device for increasing the safety of getting off a vehicle. By dynamically collecting obstacle information and performing real-time scoring, potential risks can be discovered in a timely manner, and safety risks can be evaluated based on data such as the position, speed, and acceleration of the obstacle, effectively avoiding the occurrence of accidents. The future turning of the obstacle is predicted by the LSTM model, which enhances the system's foresight of future dynamic changes and helps to take measures in advance to deal with possible risks. Environmental information, such as weather conditions and road conditions, is integrated to adjust the safety score, making the safety assessment more comprehensive and accurate. Through multi-step safety risk assessment and diversified data input, the system can more carefully evaluate the impact of various factors on safety, enhancing the comprehensiveness and accuracy of the assessment. The K-means clustering algorithm is used to automatically classify safety risks, and the risk level can be automatically divided according to different safety scores, realizing the automation and intelligence of risk management. When the vehicle stops, according to the overall safety score calculated in real time and the preset risk level threshold, the system can instantly determine the current safety level and issue a corresponding warning signal to improve the safety of passengers and pedestrians. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work:

[0068] Figure 1 It is a flow chart of a real-time monitoring method for increasing the safety of getting off a vehicle according to an embodiment of the present invention;

[0069] Figure 2 It is a structural block diagram of a real-time monitoring device for increasing the safety of getting off a vehicle according to an embodiment of the present invention;

[0070] Figure 3 It is a side view of a vehicle mounting structure of a real-time monitoring device for increasing the safety of getting off a vehicle according to an embodiment of the present invention;

[0071] Figure 4 The present invention is a top view of a vehicle mounting structure of a real-time monitoring device for increasing the safety of getting off a vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the following will be described clearly and completely in combination with the technical solution in the embodiments of the present invention. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0073] The present invention provides a real-time monitoring method and device for increasing the safety of getting off a vehicle. The specific scheme is as follows: a collection device installed on the vehicle is used to capture the dynamic information of the position, speed and acceleration of surrounding obstacles in real time. The sensor continuously monitors and collects the dynamic information of surrounding obstacles, such as pedestrians, bicycles, other vehicles, etc. The processor module converts the collected data into obstacle feature vectors and constructs an obstacle dynamic information matrix. Based on the obstacle dynamic information, a safety risk score based on the dynamic information is output. The data is transmitted to a central server, and the warning information transmitted by the central processor is received.

[0074] The dynamic information of obstacles over a period of time is integrated into a time series, and the future steering angle is predicted based on it. The original safety risk score is adjusted according to the prediction result and a new full risk score is output. Environmental factor data such as weather, lighting, and road conditions are collected, and the original risk score is adjusted according to the environmental information and a new full risk score is output. The two output new full risk scores are weighted and fused to calculate the overall safety score.

[0075] When the vehicle stops, the overall safety score is calculated in real time, and the current warning level is determined based on the preset threshold set, and a corresponding warning signal is issued.

[0076] Through the Internet of Vehicles technology, it can communicate with surrounding vehicles and traffic infrastructure to improve the accuracy and timeliness of early warnings. At the same time, it can also receive safety warning information from other vehicles to form a mutual safety warning network.

[0077] In the specific implementation, the following embodiments are included:

[0078] Embodiment 1

[0079] like Figure 1 As shown, a real-time monitoring method for increasing the safety of getting off a vehicle comprises the following steps:

[0080] Step S1: Obstacle dynamic information safety assessment

[0081] Collect data information of multiple obstacles and convert them into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , the obstacle dynamic information O tis a matrix formed by multiple obstacle feature vectors. Each row in the matrix represents a feature vector of an obstacle. O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ;

[0082] Step S2: Score adjustment based on obstacle turn prediction

[0083] Collect dynamic information of obstacles over a period of time t , and integrate them into a dynamic information sequence of obstacles, input the dynamic information sequence of obstacles into the preset LSTM model, and the LSTM model outputs the prediction of the steering angle

[0084] The safety risk score R calculated from the obstacle dynamic information is adjusted by the correction function g O , and output the adjusted security risk score R' O , the correction function g is based on the steering prediction Adjust the security risk score R O ;

[0085] Step S3: Score adjustment based on environmental information

[0086] Collect environmental data and convert it into environmental information t , to represent the environmental factors at time t, and use the correction function h to adjust the safety risk score R calculated from the obstacle dynamic information O , and output the adjusted security risk score R" O , the correction function h is based on the environmental information E t Adjust the security risk score R O ;

[0087] Step S4: Overall safety risk assessment

[0088] The safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ;

[0089] Step S5: Security risk classification

[0090] The generated multiple overall safety scores are integrated into a scoring data set. The K-means clustering algorithm is used to perform cluster analysis on the scoring data set. The number of clusters n is selected as the number of levels predetermined by the early warning system. The n cluster centers are found through the clustering algorithm. The overall safety score is divided into n segments through the n cluster centers. That is, the threshold set is determined according to the cluster center or cluster boundary. Each threshold λ i Indicates the boundary that distinguishes different security risk levels.

[0091] Step S6: Security risk assessment

[0092] Obtain the vehicle speed data and dynamic information of obstacles in the preset area. If the vehicle stops, the overall safety score R is calculated in real time. t , and by presetting the function L(R t ) determines the current warning level and issues a corresponding warning signal according to the warning level, where the function L(R t ) By comparing R t and the elements in the threshold set Λ to determine the warning level.

[0093] In the specific implementation, high-resolution cameras and radar sensors installed in the vehicle are used to capture the position, speed and acceleration of surrounding obstacles in real time. The on-board processor module is responsible for processing the data received from the sensor and converting it into a feature vector, sending the processed data to the central control system, and receiving instructions from the central control system. The sensor continuously monitors and collects dynamic information of surrounding obstacles, such as pedestrians, bicycles, other vehicles, etc. The processor module converts the collected data into obstacle feature vectors and constructs an obstacle dynamic information matrix. Using the evaluation function f O Process the dynamic information of obstacles and output the safety risk score R based on the dynamic information O .

[0094] Integrate the dynamic information of obstacles within a period of time into a time series, input the obstacle time series into the preset LSTM model, and predict the future steering angle. Adjust the safety risk score according to the prediction results, and output the adjusted safety risk score R' O .

[0095] Collect data on environmental factors such as weather, lighting, and road conditions. Use the correction function to adjust the risk score based on environmental information and output a further adjusted safety risk score R" O .

[0096] R' O and R” O Through weighted fusion, the overall safety score R is calculated t .

[0097] When the vehicle stops, the overall safety score Rt is calculated in real time, and the current warning level is determined based on the preset threshold set, and a corresponding warning signal is issued.

[0098] After implementation, the effectiveness of the system was verified through field testing, and model parameters and threshold settings were adjusted based on the collected data feedback to optimize system performance and response speed.

[0099] By dynamically collecting obstacle information and scoring it in real time, potential risks can be discovered in a timely manner, and safety risks can be assessed based on the location, speed, acceleration and other data of the obstacle, effectively avoiding accidents. The LSTM model is used to predict the future direction of the obstacle, which enhances the system's foresight of future dynamic changes and helps to take measures in advance to deal with possible risks. Environmental information, such as weather conditions and road conditions, is integrated to adjust the safety score, making the safety assessment more comprehensive and accurate. Through multi-step safety risk assessment and diversified data input, the system can more carefully assess the impact of various factors on safety, enhancing the comprehensiveness and accuracy of the assessment. The K-means clustering algorithm is used to automatically classify safety risks, and the risk level can be automatically divided according to different safety scores, realizing the automation and intelligence of risk management. When the vehicle stops, based on the overall safety score calculated in real time and the preset risk level threshold, the system can instantly determine the current safety level and issue corresponding warning signals to improve the safety of passengers and pedestrians.

[0100] In step S1 of this embodiment,

[0101] The obstacle feature vector includes the position (x t ,y t )、speed(v x,t ,v y,t ) and acceleration (a x,t ,a y,t ); The obstacle dynamic information O t Expressed as: t =[x t ,y t ,v x,t ,v y,t ,a x,t ,a y,t ] T ; Its matrix form is expressed as:

[0102]

[0103] In practical applications, in order to improve the effect of model training, features need to be standardized or normalized. Before calculating the security risk score, outliers need to be identified and processed.

[0104] Evaluation function f O It is expressed as:

[0105] R O =β0+β1x t +β2y t +β3v x,t +β4v y,t +β5a x,t +β6ay,t

[0106] Among them, R O It is a safety risk score based on dynamic information of obstacles;

[0107] β0, β1, β2...,β6 are model parameters;

[0108] The calculation method of the model parameters is:

[0109] The obstacle dynamic information O t Convert to design matrix D, and convert multiple design matrices D and R O Integrate into training data where d i is the feature vector of the i-th sample, R o,i is the corresponding security risk score; construct the cost function J(β), and obtain the optimal model parameters by minimizing the cost function

[0110] The design matrix D is expressed as:

[0111]

[0112] Where N is the number of samples, each row contains the dynamic information of the ith obstacle and a constant term 1, and the feature vector d of each sample i It is expressed as:

[0113] d i =[1,x i,t +y i,t +v x,i,t +v y,t +a x,i,t +a y,i,t ]

[0114] The cost function J(β) is expressed as:

[0115]

[0116] Solve the cost function J(β), and the solution formula is expressed as:

[0117]

[0118] Where D is an N*7 design matrix, each row represents the feature vector of a sample, and Y is the response vector, which is an N*1 vector containing the security risk scores of all samples.

[0119] With accurate parameter estimation, the model is able to predict safety risks more accurately, thereby improving the effectiveness of the early warning system. Linear models provide good interpretability, making the relationship between the various components of the safety risk score and the characteristics of obstacles clearer. Linear regression models are relatively simple and easy to implement and maintain, which is particularly important for real-time systems because they can quickly calculate and update risk scores. Computing resources can be effectively utilized to process large amounts of data in real time, ensuring the efficient operation of the system. Model parameters can be adjusted based on new data, allowing the system to adapt to changing environmental conditions and traffic conditions.

[0120] In step S2 of this embodiment,

[0121] The obstacle dynamic information sequence is represented as X = {O t-n ,...,O t-1 ,O t}; The prediction function of the obstacle turning angle is expressed as:

[0122]

[0123] in is the obstacle turning angle predicted at time t+Δt

[0124] The correction function g is expressed as:

[0125]

[0126] Among them, R O Provide safety risk scores based on dynamic obstacle information;

[0127] is the steering angle at the future time point t+Δt;

[0128] γ is an adjustment parameter used to adjust the sensitivity of the impact of the predicted steering angle on the safety risk score;

[0129] The adjustment parameter γ is calculated by collecting and analyzing the statistical relationship between the predicted steering angle change and the accident rate. The specific method includes: gathering enough historical data, including steering angles and related accident records. The data should be sufficient for effective analysis. Calculate the mean, median, minimum and maximum values ​​of the steering angle, as well as the frequency of accidents. In addition, observe the relationship between the extreme values ​​of the steering angle and the occurrence of accidents. Divide the steering angle into several segments (for example, small steering, medium steering, and large steering), and observe the accident rate in each segment. Set different γ values ​​based on the accident rate of each segment. For example, if large steering is associated with a higher accident rate, then a higher γ value can be tried in this segment. Based on the collected data and segmented observations, set γ to a fixed value, use these preliminary set γ values ​​in actual operation, and adjust the size of γ based on feedback from accidents.

[0130] Among them, the detailed configuration of the LSTM model is:

[0131] 1) Input and Output:

[0132] Input: The input vector will include historical obstacle feature vectors, such as position, velocity, acceleration. If the window size is n, the input shape is [n, 6].

[0133] Output: The output of the model is the predicted steering angle θ, a continuous value.

[0134] 2) Network architecture:

[0135] Number of layers: Two layers of LSTM are used to capture complex dependencies in time series data.

[0136] Number of units: 128 LSTM units per layer.

[0137] Regularization: A Dropout layer is used after each LSTM layer with a dropout rate of 0.5 to reduce overfitting.

[0138] Hyperparameters:

[0139] Learning rate: 0.001.

[0140] Batch size: 64, which is a balance between efficiency and performance.

[0141] Epochs: Adjusted based on the performance of the validation set, usually starting from 50.

[0142] The model training and verification methods are:

[0143] 1) Data preprocessing:

[0144] Data standardization: All input features are standardized (subtract the mean and divide by the standard deviation) to ensure that the model input layer is not affected by input features of different magnitudes.

[0145] Sequence generation: Construct a time series sample of length n from historical data as model input.

[0146] Data partitioning: The dataset is divided into training set (70%), validation set (20%), and test set (10%).

[0147] 2) Cross-validation: Use k-fold cross-validation to evaluate the generalization ability of the model. Select 5-fold cross-validation and set aside a portion of the data for validation each time during training.

[0148] 3) Performance evaluation: The mean square error (MSE) and mean absolute error (MAE) are used as the main performance evaluation indicators. These two indicators can intuitively reflect the size of the prediction error.

[0149] 4) Optimization and Adjustment: Use the Adam optimizer for model training because it can converge faster than the traditional stochastic gradient descent method.

[0150] 5) Monitor the performance of the validation set and use early stopping to prevent overfitting. If the loss on the validation set does not improve for 10 consecutive epochs, stop training.

[0151] Through the above precise and complete scheme, the LSTM model will be able to effectively learn from historical obstacle dynamic information to predict future steering angles, thereby providing accurate safety risk assessment for vehicles in practical applications. This approach not only enhances the model's predictive ability, but also ensures its reliability and stability in new environments through a rigorous training and verification process.

[0152] The correction function g uses a combination of logarithms and exponentials, a form that can handle the nonlinear effect of the predicted steering angle on safety risk. The risk score can be significantly increased when the steering angle is large, reflecting the significant impact that the increase in steering angle may have on safety risk. Safety risk scores usually need to be sensitive to small changes in input variables, especially when these variables are close to critical values. The exponential function provides a nonlinear way to amplify small changes, making the model more sensitive to changes in risk. The logarithmic function is mathematically smooth and can convert the result of the exponential function amplification into a gradual, smooth response. This means that the adjustment of the risk score will not cause drastic changes due to sudden small fluctuations, thereby providing a more stable risk assessment.

[0153] In step S3 of this embodiment, the environmental information is represented by E t=[e 1,t ,e 2,t ,e 3,t ,...,e m,t ] T , where e i,t Representing different environmental factors, the correction function h is expressed as:

[0154] R” O =R O (1+α1*e1+α2*e2+...+α m *e m )

[0155] Among them, R O Provide safety risk scores based on dynamic obstacle information;

[0156] In this embodiment, the key environmental factors affecting the safety risk score include: weather conditions (rain, snow, fog, etc.), road conditions (dry, slippery, icy), traffic density, time period (peak / off-peak), lighting and visibility, etc., where weather data is obtained from the national meteorological service agency. Road conditions are obtained through road monitoring systems or mobile application data. Traffic data is collected using traffic camera systems or traffic flow monitoring equipment.

[0157] Each α i Indicates different environmental factors i The impact strength on the security risk score; by collecting and analyzing environmental factors and security event data, using statistical methods to calculate the impact strength of each environmental factor i The influence coefficient α i . Influence coefficient α i Methods for determining include:

[0158] Environmental data and relevant traffic accident records were collected to create a dataset containing all key variables. Multiple linear regression was used to analyze the relationship between environmental factors and accident data, which included evaluating the influence strength of each factor (i.e., the influence coefficient α i ), simulate the effect of the test correction function through historical data, verify its performance under real-world conditions, monitor environmental data changes in real time, and update the security risk score in real time through software automation.

[0159] By considering the predicted steering angle and environmental factors, the risk score can be made more accurate and better reflect the actual safety situation. The adjustment parameters and influence coefficients are determined by statistical analysis and machine learning algorithms. These methods promote the implementation of data-driven decision-making and help scientifically manage traffic safety. As new data accumulates, the adjustment parameters γ and influence coefficients α can be continuously updated. i , so that the system can be continuously optimized and the efficiency and effectiveness of long-term operation can be improved.

[0160] In step S4 of this embodiment, the overall safety score is expressed as: R t =w1R' O +w2R” O , where w1 and w2 are R' O , R” O The security weight of w1 and w2 is obtained as follows:

[0161] Collect multiple samples, each sample includes a security risk score R' O , R” O , integrate multiple samples into a data set where are the safety risk scores of the i-th sample based on obstacle steering prediction and environmental information, y i is the corresponding safety status indicator;

[0162] By minimizing the objective function, the overall safety score after weighted fusion reflects the actual safety status, where the objective function is expressed as:

[0163]

[0164] Among them, w1, w2 ≥ 0, w1 + w2 = 1;

[0165] The objective function is iterated multiple times according to the gradient algorithm. In each iteration, the weight is updated according to the gradient of the objective function with respect to the weight, which is expressed as:

[0166]

[0167] Where μ is the learning rate, Represents the derivative of the loss function with respect to the weight w1;

[0168] Through iterative updates, the difference between the overall safety score and the actual safety status is minimized, and the weight at this time is the determined weight. Cross-validation methods can also be applied to evaluate the generalization ability of the model, avoid overfitting, and ensure that the model can maintain stable performance on different data subsets.

[0169] Specifically, during iterative updating, weight w1 and weight w2 are initialized to equal values. In this embodiment, each is 0.5 to ensure that the initial state meets the constraint that the sum of the weights is 1;

[0170] After each weight update, the weights are normalized to ensure that they sum to 1. This can be achieved using the following formula:

[0171]

[0172] Among them, the loss function L is the mean square error, expressed as The analytical gradient is calculated as:

[0173] set up And w2=1-w1, gradient can be calculated according to the following formula:

[0174]

[0175] Analytical gradient calculation can avoid the instability and calculation errors of numerical methods.

[0176] Determining weights through a data-driven approach ensures that the selection of weights is based on actual data, thereby improving the accuracy and practicality of the scoring model. The optimized weights ensure that the overall safety score better reflects the actual safety situation, which helps to achieve more effective safety monitoring and management. This method allows the system to adjust weights in different environments and conditions to adapt to possible environmental differences. Through clear weight allocation, the decision-making process of the model is more transparent and easy to explain, which is particularly important for safety-related systems. Normalized weights ensure the stability of the model and avoid extreme weight allocation situations, thereby enhancing the overall stability and reliability of the system.

[0177] In step S5 of this embodiment, the scoring data set is represented as {R 1,t ,R 2,t ,R 3,t ,...,R M,t}, the steps to perform cluster analysis on the scoring data set using the K-means clustering algorithm are:

[0178] Initialization: Randomly select n overall security risk scores as the initial cluster centers {c1, c2, c3, ..., c n};

[0179] Cluster assignment: For each overall security risk score R i,t , calculate its relationship with each cluster center c j The distance R i,t Assign it to the cluster represented by the nearest cluster center, that is, for each overall security risk score R i,t , assigned to cluster {k=argmin||R i,t -c j || 2};

[0180] Update cluster centers: Recalculate the center of each cluster to obtain the average of all overall security risk scores in each cluster, and the new cluster center c′ j The calculation method is:

[0181] Repeat the steps of cluster assignment and updating cluster centers until the cluster centers no longer change significantly;

[0182] After clustering is completed, the midpoint between two adjacent cluster centers is used as the split point to define the threshold of the security risk level and form a set Λ={λ1,λ2,...,λ n}; through the threshold {λ1,λ2,,...,λ n The overall security risk score data is divided into n segments, each segment corresponding to a security risk level.

[0183] By dividing the overall security risk score into different segments, the system can more accurately identify and respond to different levels of security risks. This helps provide decision makers with different levels of early warning and response strategies. Cluster analysis provides a quantitative analysis method that classifies risk scores through mathematical models, reducing the subjectivity of human judgment. With clear security risk levels, relevant personnel and systems can respond more quickly and accurately, especially in emergency situations, and appropriate response measures can be quickly determined. Automatic clustering and threshold definition reduce manual operation and maintenance requirements, making the system more intelligent and automated. Using a data-driven approach to determine security risk levels and corresponding thresholds enhances the scientific nature and accuracy of decision-making.

[0184] In step S6 of the present embodiment, when the vehicle stops and shifts to P gear or the driving speed exceeds 50 and then drops to 0, the overall safety risk score of the effective area for door opening is obtained according to the dynamic information of the obstacle, and it is determined whether to send a restriction signal to the door restriction device according to the result of the overall safety risk score, and a warning message is sent to a mobile phone and an in-vehicle display module, wherein the warning message includes a warning level; the effective area for door opening includes a left area and a right area.

[0185] In this embodiment, the function L(R t ) can be expressed as:

[0186]

[0187] Read the gear position and vehicle speed data by connecting to the vehicle ECU computer. When the vehicle stops and shifts to the P gear, or when the driving speed exceeds 50 and then drops to 0, immediately calculate the overall safety score when the vehicle stops. Use a preset function to determine the current warning level according to the overall safety score. The warning levels include no risk (Rt ≤ λ1), low risk (λ1 < Rt ≤ λ2), medium risk (λ2 < Rt ≤ λ3), and high risk (Rt > λ3); when any side door is opened, if the warning level of that side area is low risk, use the forced voice and video reminder mode of Bluetooth or V2X communication mode. If the warning level of that side area is medium risk, the warning light emits a yellow flash to remind the following vehicle to pay attention to the door opening, send a switching signal to the display module, the touch display screen displays the image information obtained by the camera, and the getting-off personnel are reminded to pay attention to observing the situation behind through voice; if the warning level of that side area is high risk, the warning light emits a red flash to remind the following vehicle to pay attention to the door opening, the touch display screen receives the switching signal and displays the image information obtained by the camera, and sends a restriction signal to the door restriction device to control the opening amplitude of the opened door.

[0188] The vehicle stop triggers the calculation of the safety score and the warning, minimizing the response time and improving the efficiency of handling emergencies. Through the real-time updated safety risk score, the system can more accurately reflect the safety status in the current environment and situation. Set a multi-level warning mechanism, conduct hierarchical management according to the safety risk level, make the warning system more refined, and corresponding measures can be taken for different levels of risks. Through real-time warning and feedback, enhance the safety awareness of drivers and passengers, and promote them to take necessary safety measures. All warnings and responses of the system are based on real-time data and scientific calculations to ensure the accuracy and reliability of decisions.

[0189] Embodiment 2

[0190] As Figure 2-4 shown, the present invention also provides a real-time monitoring method and device for enhancing the safety of getting off the vehicle. Based on the real-time monitoring method for enhancing the safety of getting off the vehicle in the previous embodiment, it includes the following modules:

[0191] Camera module 1: Set in the middle of the rear windshield, used to obtain the image information of the preset area and send the image information to the display module;

[0192] Radar module 3, used to collect the motion information of moving objects in the preset area and send the motion information to the processor module in real time. The motion information includes distance information, speed information, and direction information;

[0193] Wireless connection module 4, used to connect to the mobile phones of the vehicle occupants and transmit warning information and image information to the mobile phones;

[0194] Processor module 6 obtains the vehicle speed data and the dynamic information of obstacles in the preset area. If it is detected that the vehicle stops, the warning level is obtained in real time, and a corresponding warning signal is issued according to the warning level;

[0195] The display module 2 includes a touch screen disposed in the armrest box, and the touch screen displays image information acquired by the camera when receiving a switching signal;

[0196] The door restriction device 5 is used to control the opening amplitude of the opened door according to the restriction signal.

[0197] The camera module, display module, radar module, wireless connection module and door restriction device are all connected to the processor module.

[0198] The present invention adopts multiple early warning methods to comprehensively prevent door collision accidents. First, the dynamic information of obstacles in the preset area is collected by radar, and the motion information is sent to the dynamic module in real time. The processor module obtains the early warning level according to the dynamic information of the obstacle, and sends the corresponding early warning signal according to the early warning level. The corresponding risk level is used to warn the people getting off the car. The control module controls the operation of the warning module and the door restriction module, fully considering the change of the distance between the moving object and the door due to the motion properties of the obstacle, and improving the accuracy of the detection of moving objects near the door. People in the car can observe the rear vehicle through the touch screen and drive when they are sure it is safe.

[0199] In this embodiment, a server module is also included, and the server module includes:

[0200] Obstacle dynamic information scoring unit: used to receive data information of multiple obstacles and convert it into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , by evaluating the function f O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ;

[0201] The first adjustment unit: collects dynamic information of obstacles within a period of time. t , and integrate them into a dynamic obstacle information sequence, which is input into the preset LSTM model. The LSTM model outputs the prediction of the steering angle, and adjusts the safety risk score R calculated from the dynamic obstacle information based on the steering prediction. O , and output the adjusted security risk score R' O ;

[0202] The first adjustment unit: collects environmental information and adjusts the safety risk score R calculated from the dynamic information of obstacles based on the environmental information.O , and output the adjusted security risk score R" O ;

[0203] Safety risk overall scoring unit: the safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ;

[0204] Security risk grading unit: Integrate the generated multiple overall security scores into a scoring data set, and perform cluster analysis on the scoring data set using the K-means clustering algorithm.

[0205] In view of the shortcomings of the prior art, the present invention dynamically collects obstacle information and performs real-time scoring, timely discovers potential risks, and evaluates safety risks based on the location, speed, acceleration and other data of the obstacle, effectively avoiding accidents. The LSTM model is used to predict the future turning of the obstacle, and the safety score is adjusted through prediction, which enhances the system's foresight of future dynamic changes and helps to take measures in advance to deal with possible risks. Environmental information, such as weather conditions and road conditions, is integrated to adjust the safety score, making the safety assessment more comprehensive and accurate. Through multi-step safety risk assessment and diversified data input, the system can more carefully evaluate the impact of various factors on safety, enhancing the comprehensiveness and accuracy of the assessment. The K-means clustering algorithm is used to automatically classify safety risks, and the risk level can be automatically divided according to different safety scores, realizing the automation and intelligence of risk management. When the vehicle stops, according to the overall safety score calculated in real time and the preset risk level threshold, the system can instantly determine the current safety level and issue corresponding warning signals to improve the safety of passengers and pedestrians.

[0206] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A real-time monitoring method for increasing the safety of getting off a vehicle, characterized in that: include: Step S1: Obstacle dynamic information safety assessment Receive data information of multiple obstacles and convert them into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , the obstacle dynamic information O t is a matrix formed by multiple obstacle feature vectors. Each row in the matrix represents a feature vector of an obstacle. O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ; Step S2: Score adjustment based on obstacle turn prediction Collect dynamic information of obstacles over a period of time t , and integrate them into a dynamic information sequence of obstacles, input the dynamic information sequence of obstacles into the preset LSTM model, and the LSTM model outputs the prediction of the steering angle The safety risk score calculated from the obstacle dynamic information is adjusted by the correction function g R O , and output the adjusted security risk score R' O , the correction function g is based on the steering prediction Adjust the security risk score R O ; The obstacle dynamic information sequence is represented as X = {O t-n ,...,O t-1 ,O t }; The prediction function of the obstacle turning angle is expressed as: in is the obstacle turning angle predicted at time t+Δt; The correction function g is expressed as: Among them, R O Provide safety risk scores based on dynamic obstacle information; is the steering angle at the future time point t+Δt; γ is an adjustment parameter used to adjust the sensitivity of the impact of the predicted steering angle on the safety risk score; The adjustment parameter γ is calculated by collecting and analyzing the statistical relationship between the predicted steering angle change and the accident rate; Step S3: Score adjustment based on environmental information Collect environmental data and convert it into environmental information t , to represent the environmental factors at time t, and use the correction function h to adjust the safety risk score R calculated from the obstacle dynamic information O , and output the adjusted security risk score R" O , the correction function h is based on the environmental information E t Adjust the security risk score R O ; Step S4: Overall safety risk assessment The safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ; Step S5: Security risk classification The generated multiple overall safety scores are integrated into a scoring data set. The K-means clustering algorithm is used to perform cluster analysis on the scoring data set. The number of clusters n is selected as the number of levels predetermined by the early warning system. The n cluster centers are found through the clustering algorithm. The overall safety score is divided into n segments through the n cluster centers. That is, the threshold set is determined according to the cluster center or cluster boundary. Each threshold λ i Indicates the boundaries that distinguish different security risk levels; Step S6: Security risk assessment Obtain the vehicle speed data and dynamic information of obstacles in the preset area. If the vehicle stops, the overall safety score R is calculated in real time. t , and by presetting the function L(R t ) determines the current warning level and issues a corresponding warning signal according to the warning level, where the function L(R t ) By comparing R t and the elements in the threshold set Λ to determine the warning level.

2. The real-time monitoring method for increasing the safety of getting off a vehicle according to claim 1, characterized in that: In step S1, The obstacle feature vector includes the position (x t ,y t )、speed(v x,t ,v y,t ) and acceleration (a x,t ,a y,t ); The obstacle dynamic information O t It is expressed as: Evaluation function f O It is expressed as: R O =β0+β1x t +β2y t +β3v x,t +β4v y,t +β5a x,t +β6a y,t Among them, R O It is a safety risk score based on dynamic information of obstacles; β0, β1, β2...,β6 are model parameters; The calculation method of the model parameters is: The obstacle dynamic information O t Convert to design matrix D, and convert multiple design matrices D and R O Integrate into training data where d i is the feature vector of the i-th sample, R o,i is the corresponding security risk score; construct the cost function J(β), and obtain the optimal model parameters by minimizing the cost function 3. The real-time monitoring method for increasing the safety of getting off a vehicle according to claim 1, characterized in that: In step S3, the environmental information is represented as where e i,t Representing different environmental factors, the correction function h is expressed as: R" O =R O (1+α1*e1+α2*e2+...+α m *e m ); Among them, R O Score the safety risk based on the dynamic information of obstacles; each α i Indicates different environmental factors i The intensity of the impact on the security risk score; By collecting and analyzing environmental factors and security incident data, statistical methods are used to calculate each environmental factor e i The influence coefficient α i .

4. The real-time monitoring method for increasing the safety of getting off a vehicle according to claim 1, characterized in that: In step S4, the overall safety score is expressed as: R t =w1R' O +w2R” O , where w1 and w2 are R' O , R” O The security weight of w1 and w2 is obtained as follows: Collect multiple samples, each sample includes a security risk score R' O , R” O , integrate multiple samples into a data set where are the safety risk scores of the i-th sample based on obstacle steering prediction and environmental information, y i is the corresponding safety status indicator; By minimizing the objective function, the overall safety score after weighted fusion reflects the actual safety status, where the objective function is expressed as: Among them, w1, w2 ≥ 0, w1 + w2 = 1; The objective function is iterated multiple times according to the gradient algorithm. In each iteration, the weight is updated according to the gradient of the objective function with respect to the weight, which is expressed as: Where μ is the learning rate, Represents the derivative of the loss function with respect to the weight w1; Through iterative updates, the weight is determined until the difference between the overall security score and the actual security status is minimized.

5. The real-time monitoring method for increasing the safety of getting off a vehicle according to claim 1, characterized in that: In step S5, the scoring dataset is represented as {R 1,t ,R 2,t ,R 3,t ,...,R M,t }, the steps to perform cluster analysis on the scoring data set using the K-means clustering algorithm are: Initialization: Randomly select n overall security risk scores as the initial cluster centers {c1, c2, c3, ..., c n }; Cluster assignment: For each overall security risk score R i,t , calculate its relationship with each cluster center c j The distance R i,t Assign it to the cluster represented by the nearest cluster center, that is, for each overall security risk score R i,t , assigned to cluster {k = argmin||R i,t -c j ∣∣ 2 }; Update cluster centers: Recalculate the center of each cluster to obtain the average of all overall security risk scores in each cluster, and the new cluster center c′ j The calculation method is: Repeat the steps of cluster assignment and updating cluster centers until the cluster centers no longer change significantly; After clustering is completed, the midpoint between two adjacent cluster centers is used as the split point to define the threshold of the security risk level and form a set Λ={λ1,λ2,...,λ n }; through the threshold {λ1,λ2,...,λ n The overall security risk score data is divided into n segments, each segment corresponding to a security risk level.

6. The real-time monitoring method for increasing the safety of getting off a vehicle according to claim 1, characterized in that: In step S6, when the vehicle stops and shifts to P gear or the driving speed exceeds 50 and then drops to 0, the overall safety risk score of the effective area for door opening is obtained according to the dynamic information of the obstacle, and a determination is made whether to send a restriction signal to the door restriction device according to the result of the overall safety risk score, and a warning message is sent to a mobile phone and an in-vehicle display module, wherein the warning message includes a warning level; the effective area for door opening includes a left area and a right area.

7. A real-time monitoring device for increasing the safety of getting off a vehicle, based on the real-time monitoring method for increasing the safety of getting off a vehicle according to any one of claims 1 to 6, characterized in that: Includes the following modules: A camera module (1) is arranged in the middle of the rear windshield and is used to obtain image information of a preset area and send the image information to a display module; The display module (2) comprises a touch screen arranged in the armrest box, and the touch screen displays image information acquired by the camera when receiving a switching signal; A radar module (3) is used to collect motion information of moving objects in a preset area and send the motion information to the processor module in real time, wherein the motion information includes distance information, speed information and direction information; A wireless connection module (4), used to connect to a mobile phone of a person in the vehicle and transmit warning information and image information to the mobile phone; The processor module (6) obtains the vehicle speed data and the dynamic information of obstacles in the preset area. If it is detected that the vehicle stops, the warning level is obtained in real time, and a corresponding warning signal is issued according to the warning level, and whether to send a door restriction signal; The vehicle door limiting device (5) is used to control the opening amplitude of the vehicle door to be opened according to the limiting signal.

8. The real-time monitoring device for increasing the safety of getting off a vehicle according to claim 7, characterized in that: Also included is a server module, the server module comprising: Obstacle dynamic information scoring unit: used to collect data information of multiple obstacles and convert them into obstacle feature vectors, and integrate multiple obstacle feature vectors to form obstacle dynamic information. t , by evaluating the function f O Process the dynamic information of obstacles and output the security risk score R based on the dynamic information O ; The first adjustment unit: collects dynamic information of obstacles within a period of time. t , and integrate them into a dynamic obstacle information sequence, which is input into the preset LSTM model. The LSTM model outputs the prediction of the steering angle, and adjusts the safety risk score R calculated from the dynamic obstacle information based on the steering prediction. O , and output the adjusted security risk score R' O ; The first adjustment unit: collects environmental information and adjusts the safety risk score R calculated from the dynamic information of obstacles based on the environmental information. O , and output the adjusted security risk score R" O ; Safety risk overall scoring unit: the safety score R' O and R” O The overall safety score R is calculated by combining the preset weights with the weighted fusion model t ; Security risk grading unit: Integrate the generated multiple overall security scores into a scoring data set, and perform cluster analysis on the scoring data set using the K-means clustering algorithm.

Images