An intelligent safety alarm system and method for seat belt fastening monitoring

A multi-sensor system with semi-supervised learning models enhances seat belt monitoring accuracy by analyzing insertion, tension, and rewinding states, reducing false and missed alarms and ensuring safety.

CN118810674BActive Publication Date: 2024-12-06HONGSI AUTO PARTS (CHANGZHOU) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411296410.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-18
Publication Date
2024-12-06
Estimated Expiration
2044-09-18

AI Technical Summary

Technical Problem

Existing seat belt monitoring systems fail to accurately distinguish between actual and fake fastening behaviors, leading to high false alarm and missed alarm rates, especially in complex driving environments, thus compromising safety.

Method used

A multi-sensor system integrated with a semi-supervised learning model using a K nearest neighbor and support vector machine model to analyze seat belt insertion, tension, pull-out position, and rewinding status, providing comprehensive data fusion and intelligent alarm triggering.

Benefits of technology

The system significantly reduces false and missed alarms by accurately identifying fake fastening behaviors, ensuring high reliability and safety through real-time monitoring and adaptive adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118810674B_ABST
    Figure CN118810674B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent safety alarm system and method for monitoring false fastening of seat belts, S1, installing multiple sensors at the vehicle seat, the seat belt pull-out end and the reel position; S2, obtaining the data detected by all sensors; S3, transmitting all the sensor detection data of S2 to the intelligent control system, and processing the sensor detection data using a semi-supervised integrated model; S4, in the intelligent control system, using the semi-supervised integrated model to perform fusion analysis on the received sensor detection data to determine whether the passenger wears the seat belt correctly and whether there is false fastening behavior; S5, reminding the driver or passenger to correct the wearing behavior; S6, recording the abnormal detection situation in the system storage unit for subsequent analysis. The present invention improves the recognition accuracy of false fastening behavior, effectively reduces the system and driver misjudgment caused by artificial false fastening, avoids unnecessary harm to passengers caused by accidents, and ensures the high reliability of the system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring, and in particular to an intelligent safety alarm system and method for monitoring seat belt fastening. Background Art

[0002] With the development of the automobile industry and the improvement of people's safety awareness, seat belts, as an important passive safety device, have become an indispensable part of vehicles. The existing seat belt system effectively protects the lives of occupants in the vehicle through physical restraint when the vehicle collides. However, as drivers and passengers gradually adapt to the seat belt system, some new problems also arise. For example, in order to circumvent the vehicle's safety alarm system, some passengers choose to only insert the seat belt into the buckle without actually wearing it, or bypass the system detection in other ways. This "fake fastening" behavior seriously weakens the protective function of the seat belt and may cause serious personal injury in the event of a vehicle accident.

[0003] Traditional seat belt alarm systems mainly determine whether passengers have fastened their seat belts by detecting whether the seat belt buckle is inserted. However, simple detection methods cannot effectively distinguish whether the seat belt is actually worn by the passenger, which leads to frequent false alarms or missed alarms of the system. For traditional systems that only detect the buckle status, as long as the seat belt is inserted into the buckle, the system considers that the seat belt has been correctly used. The above judgment method ignores the actual wearing condition of the seat belt on the passenger's body. For example, the passenger can choose to insert the seat belt into the buckle and then put the seat belt across the back of the seat, or pull it out to a certain length and then directly rewind it. The above operations cannot be accurately identified by the traditional detection system. This detection defect causes the seat belt to fail to provide the proper protection when an accident occurs, increasing the risk of passenger injury.

[0004] In response to the above problems, there are some improvement measures in the prior art, such as adding seat belt tension sensors and seat pressure sensors to improve the ability to identify the actual wearing conditions of seat belts. However, the above technical solutions still have certain limitations, such as unreasonable sensor layout, insufficient data processing capabilities, and the inability to adaptively adjust in complex environments, resulting in the system still failing to achieve the desired effect in practical applications. In addition, traditional detection systems lack intelligent data processing and analysis methods and are unable to combine data from multiple sensors for comprehensive analysis, resulting in low detection accuracy for false-belt behaviors in some cases and the inability to achieve adaptive adjustments to complex driving environments.

[0005] In summary, the existing seat belt monitoring technology has obvious shortcomings. It cannot effectively identify the behavior of pretending to wear a seat belt, and exhibits a high false alarm rate and missed alarm rate in complex driving environments, which reduces the reliability and practicality of the system. Therefore, there is an urgent need for an intelligent safety alarm system for pretending to wear a seat belt to effectively solve the shortcomings of the existing technology and improve the overall safety of the vehicle. Summary of the invention

[0006] An object of the present invention is to provide an intelligent safety alarm system and method for seat belt fake fastening monitoring, which improves the recognition accuracy of fake fastening behavior and effectively reduces false alarms and missed alarms caused by single-point failures of sensors or data anomalies, thereby ensuring the high reliability of the system.

[0007] An intelligent safety alarm method for seat belt fastening monitoring according to an embodiment of the present invention comprises the following steps:

[0008] S1, installing multiple sensors at the vehicle seat, the seat belt pull-out end and the retractor position, including a first sensor for detecting the seat belt buckle state, a second sensor for detecting the seat belt tension, a third sensor for detecting the seat belt pull-out position and a fourth sensor for detecting the seat belt rewinding position;

[0009] S2, obtaining seat belt extension position data detected by the third sensor, obtaining seat belt buckle state data detected by the first sensor, obtaining seat belt rewinding position data detected by the fourth sensor, and obtaining seat belt tension data detected by the second sensor;

[0010] S3, transmit all sensor detection data of S2 to the intelligent control system, and process the sensor detection data using a semi-supervised integration model based on Scikit-learn, in which a semi-supervised K nearest neighbor model and a semi-supervised support vector machine model are integrated;

[0011] S4. In the intelligent control system, a semi-supervised integrated model is used to perform fusion analysis on the received sensor detection data to determine whether the passenger wears the seat belt correctly and whether there is any false fastening behavior, including determining whether the seat belt passes through the correct pull-out position, whether it is correctly inserted into the socket, and whether it is correctly rewound after insertion;

[0012] S5. When it is determined that the passenger is not wearing the seat belt correctly or is pretending to wear the seat belt, the intelligent control system triggers the safety alarm system and sends an alarm signal through the vehicle audio, video or vibration device to remind the driver or passenger to correct the wearing behavior;

[0013] S6. Record the abnormality detection situation in the system storage unit for subsequent analysis.

[0014] Optionally, the S2 step includes:

[0015] S21, install a first sensor at the bottom of the vehicle seat to detect the state of the seat belt buckle, the first sensor detects the operation of inserting the seat belt into the socket, and the first sensor outputs a signal of ,in Indicates the seat belt insertion status signal. When the seat belt is inserted , when not inserted ;

[0016] S22, install a second sensor in the middle or near the seat belt to detect the tension state of the seat belt, the second sensor generates a signal by measuring the tension of the seat belt, and the second sensor outputs a signal of ,in Indicates the seat belt tension signal. When the seat belt is under stress , when no force is applied ;

[0017] S23, a third sensor is installed at the seat belt pulling end position, which is used to detect the pulling position of the seat belt. The third sensor identifies whether the seat belt is pulled out from the correct position. The third sensor outputs a signal of ,in Indicates the seat belt extension position signal. When the seat belt is correctly extended , when not pulled out correctly ;

[0018] S24, install a fourth sensor at the seat belt retractor position to detect the rewinding position of the seat belt. The fourth sensor monitors whether the seat belt is correctly rewound after insertion. The fourth sensor outputs a signal of ,in Indicates the seat belt rewinding status signal. When the seat belt is retracted correctly , when not rewound correctly .

[0019] Optionally, the S3 step includes:

[0020] S31: The seat belt insertion state signal output by the first sensor , the seat belt tension signal output by the second sensor , the seat belt extension position signal output by the third sensor And the seat belt rewinding status signal output by the fourth sensor Transmitted to the central processing unit of the intelligent control system via the vehicle's internal communication bus;

[0021] S32, the intelligent control system pre-processes the received signal data, including signal normalization processing and noise filtering processing, to obtain a normalized data set ,in , , , They are respectively the normalized seat belt insertion state signal, seat belt tension signal, seat belt pull-out position signal and seat belt rewind state signal;

[0022] S33, using the semi-supervised ensemble model based on Scikit-learn to process the preprocessed data set;

[0023] S34. In the semi-supervised ensemble model, the semi-supervised K nearest neighbor model is used to preliminarily classify the data set, and the preliminary classification results are obtained by calculating the distance between the samples to be classified and the samples of known categories. ;

[0024] S35. Use the semi-supervised support vector machine model to optimize the preliminary classification results, construct the optimal classification hyperplane, further classify the samples, and obtain the final classification results. .

[0025] Optionally, the step S34 includes:

[0026] S341, inputting the preprocessed data set into a semi-supervised K nearest neighbor model;

[0027] S342, calculate the samples to be classified With labeled samples And unlabeled samples The weighted Euclidean distance between them is calculated by combining the different weights of different sensor signals on the classification results, and the weighted formula is adopted:

[0028] ;

[0029] ;

[0030] in, represents the mth eigenvalue of the sample x to be classified, and the corresponding features include the normalized seat belt insertion state signal, seat belt tension signal, seat belt pull-out position signal and seat belt rewinding state signal; Indicates labeled samples The mth eigenvalue of ; Represents unlabeled samples The mth eigenvalue of ; Is with characteristics The relevant weight coefficient reflects the relative importance of the feature in the classification process; It is the high-order weight coefficient of the feature, which is used to deal with the nonlinear relationship between high-order features;

[0031] S343, calculate the preliminary classification result of the sample to be classified x according to the categories of the nearest k labeled samples and the potentially labeled unlabeled samples :

[0032] ;

[0033] in, represents the preliminary classification result of the sample x to be classified, u represents the number of unlabeled samples, is the weight associated with the i-th labeled sample, is the category of the i-th labeled sample (0 or 1), is the weight associated with the jth unlabeled sample, is the unlabeled sample obtained through model self-learning The latent class probability, represents the category probability of unlabeled samples obtained through self-learning, is the distance between the sample to be classified and the unlabeled sample calculated in S342, is a parameter used to control distance attenuation.

[0034] Optionally, the step S35 includes:

[0035] S351, the preliminary classification results Together with the preprocessed data set, it is input into the semi-supervised support vector machine model, where As the initial label of the semi-supervised support vector machine model, it helps the semi-supervised support vector machine model determine the initial classification boundary;

[0036] S352. In the semi-supervised support vector machine model, by constructing a classification hyperplane, different categories in the sample space are divided to maximize the interval between samples, and the optimization objective function is:

[0037] ;

[0038] in, is the normal vector of the classification hyperplane, b is the bias, is the slack variable for a specific sample, represents the slack variable of the entire sample set, and C is the penalty parameter; The weight coefficient for adjusting the consistency between the preliminary classification results and the final classification results is used to balance the maximization of the classification interval, the penalty for misclassified samples, and the consistency with the preliminary classification results;

[0039] S353. Construct Lagrangian function to solve optimization problem:

[0040] ;

[0041] in, is the Lagrange multiplier, represents the Lagrange multiplier vector of all sample points, is the actual category label of the sample, is the feature vector of the sample;

[0042] S354, by solving the Lagrangian function, combined with the preliminary classification results , get the optimized classification hyperplane, and use the hyperplane to finally classify the input samples to get the final classification result of the samples :

[0043] ;

[0044] in, is the final classification result of sample x, The function is used to determine the classification result. The output is 1 if the sample is normal, and the output is 0 if the sample is abnormal.

[0045] Optionally, the S4 step includes:

[0046] S41. The final classification result after processing by the semi-supervised ensemble model Fusion analysis is performed with the pre-processed sensor detection data to determine the use status of the seat belt:

[0047] ;

[0048] in, is the comprehensive evaluation function after fusion analysis; is the fusion coefficient, which is used to adjust the weight ratio of sensor data and the final classification result; and are the linear weight and nonlinear weight coefficient of sensor data, respectively, reflecting the influence of each sensor signal in the overall judgment. and Represent linear and nonlinear mapping functions, corresponding to the preprocessed sensor signal data , The final classification result The weight coefficient of the factor is used to adjust its role in the overall evaluation;

[0049] S42: For the detection results of different sensors, the following judgments are made respectively:

[0050] The normalized seat belt insertion status signal Analyze to determine whether the seat belt is correctly inserted into the socket:

[0051] ;

[0052] in, and is the calibration factor, and It is the logarithmic probability of the insertion state signal and the Sigmoid function transformation result, reflecting the correctness of the seat belt insertion action;

[0053] The normalized seat belt tension signal Analyze to determine whether the seat belt is under stress:

[0054] ;

[0055] in, and is the calibration factor, and The square root and hyperbolic tangent function transformation results of the tension signal are used to evaluate the stress condition of the seat belt;

[0056] The normalized seat belt pull-out position signal Analyze to determine whether the seat belt passes through the correct pull-out position:

[0057] ;

[0058] in, and is the calibration factor, and They are the exponential and Sigmoid function transformation results of the pull-out position signal, respectively, to evaluate the correctness of the pull-out position;

[0059] The normalized seat belt rewinding status signal Analyze to determine whether the seat belt is retracted correctly:

[0060] ;

[0061] in, and is the calibration factor, and are the square and logarithmic transformation results of the rewinding state signal, respectively, to evaluate the rewinding action of the seat belt;

[0062] S43, each judgment result and the final classification result Combined with the above, a comprehensive judgment model is formed to comprehensively score the correctness of passengers wearing seat belts:

[0063] ;

[0064] Among them, S is the comprehensive score, , , , are the weight coefficients of insertion state, tension, pull-out position and rewinding state, respectively, to adjust the proportion of each score in the overall score. Evaluating functions for fusion analysis The weight coefficient of

[0065] S44: Determine whether the passenger wears the seat belt correctly based on the comprehensive score S and set a threshold :

[0066] ;

[0067] in, Passengers wear seat belts correctly. Indicates false wearing behavior or incorrect wearing.

[0068] An intelligent safety alarm system for seat belt false fastening monitoring, comprising:

[0069] Sensor modules, respectively used to detect the insertion state, tension state, pull-out position and rewinding state of the seat belt, each sensor module including one or more sensors;

[0070] The data processing module is connected to the sensor module, receives the output signal of the sensor, pre-processes the output signal, including signal normalization processing and noise filtering, and outputs the pre-processed signal data set;

[0071] The semi-supervised learning integrated model module, including the semi-supervised K nearest neighbor model and the semi-supervised support vector machine model, receives the output signal data set of the data processing module and processes it through the following steps:

[0072] The semi-supervised K nearest neighbor model performs preliminary classification on the signal data set and outputs preliminary classification results;

[0073] The semi-supervised support vector machine model optimizes the preliminary classification results, constructs the classification hyperplane, and outputs the final classification results;

[0074] The fusion analysis module receives the final classification results of the semi-supervised learning integrated model module and the pre-processed sensor detection data, generates a comprehensive score through fusion analysis, and determines whether the passenger wears the seat belt correctly;

[0075] The alarm module compares the comprehensive score generated by the fusion analysis module with the set threshold value. When the comprehensive score is less than the set threshold value, the alarm device is triggered to emit an audible, visual or vibration alarm to alert passengers or drivers.

[0076] The storage module is used to record and store anomaly detection results and alarm triggering information for subsequent analysis and reference.

[0077] The beneficial effects of the present invention are:

[0078] The present invention realizes accurate recognition of fake-fastening behavior of seat belts by introducing multi-sensor fusion technology and intelligent algorithms based on semi-supervised learning. Compared with the traditional system that only relies on the simple judgment of the seat belt insertion status, the present invention can comprehensively monitor the insertion, tension, pull-out position and rewinding status of the seat belt through the comprehensive data of the first sensor, the second sensor, the third sensor and the fourth sensor, which not only improves the recognition accuracy of fake-fastening behavior, but also effectively reduces false alarms and missed alarms caused by single-point failures of sensors or data anomalies, thereby ensuring the high reliability of the system.

[0079] The present invention adopts an integrated model based on semi-supervised learning, including a semi-supervised K nearest neighbor model and a semi-supervised support vector machine model. The combined model can not only process the data of labeled samples and unlabeled samples, effectively improving the accuracy of classification, but also can use the information of unlabeled samples to train the model when the labeled data is insufficient, thereby improving the robustness of the model. The semi-supervised K nearest neighbor model is used for preliminary classification, and then combined with the optimization processing of the semi-supervised support vector machine model, the optimal classification hyperplane can be constructed, the accurate classification of complex samples can be achieved, and the situation that the passenger does not wear the seat belt correctly can be effectively identified, thereby triggering the alarm system.

[0080] The present invention further enhances the intelligence of the system through a fusion analysis model, which fuses the classification results processed by the semi-supervised learning model with the sensor detection data to form a comprehensive scoring mechanism. By setting thresholds, it can conduct a quantitative assessment of the passengers' seat belt wearing conditions in real-time monitoring and determine whether an alarm needs to be issued based on the comprehensive score. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0082] Figure 1 A flow chart of an intelligent safety alarm system and method for seat belt fastening monitoring proposed by the present invention;

[0083] Figure 2The present invention provides a schematic diagram of the structure of a semi-supervised learning integrated model in an intelligent safety alarm system and method for seat belt fastening monitoring. DETAILED DESCRIPTION

[0084] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0085] refer to Figure 1-Figure 2 , an intelligent safety alarm method for seat belt false fastening monitoring, comprising the following steps:

[0086] S1, installing multiple sensors at the vehicle seat, the seat belt pull-out end and the retractor position, including a first sensor for detecting the seat belt buckle state, a second sensor for detecting the seat belt tension, a third sensor for detecting the seat belt pull-out position and a fourth sensor for detecting the seat belt rewinding position;

[0087] S2, obtaining seat belt extension position data detected by the third sensor, obtaining seat belt buckle state data detected by the first sensor, obtaining seat belt rewinding position data detected by the fourth sensor, and obtaining seat belt tension data detected by the second sensor;

[0088] S3, transmit all sensor detection data of S2 to the intelligent control system, and process the sensor detection data using a semi-supervised integration model based on Scikit-learn, in which a semi-supervised K nearest neighbor model and a semi-supervised support vector machine model are integrated;

[0089] S4. In the intelligent control system, a semi-supervised integrated model is used to perform fusion analysis on the received sensor detection data to determine whether the passenger wears the seat belt correctly and whether there is any false fastening behavior, including determining whether the seat belt passes through the correct pull-out position, whether it is correctly inserted into the socket, and whether it is correctly rewound after insertion;

[0090] S5. When it is determined that the passenger is not wearing the seat belt correctly or is pretending to wear the seat belt, the intelligent control system triggers the safety alarm system and sends an alarm signal through the vehicle audio, video or vibration device to remind the driver or passenger to correct the wearing behavior;

[0091] S6. Record the abnormality detection situation in the system storage unit for subsequent analysis.

[0092] In this embodiment, step S2 includes:

[0093] S21, install a first sensor at the bottom of the vehicle seat to detect the state of the seat belt buckle, the first sensor detects the operation of inserting the seat belt into the socket, and the first sensor outputs a signal of ,in Indicates the seat belt insertion status signal. When the seat belt is inserted , when not inserted ;

[0094] S22. Install a second sensor in the middle or near the seat belt to detect the tension state of the seat belt. The second sensor generates a signal by measuring the tension of the seat belt. The output signal of the second sensor is ,in Indicates the seat belt tension signal. When the seat belt is under stress , when no force is applied ;

[0095] S23, a third sensor is installed at the seat belt pulling end position, which is used to detect the pulling position of the seat belt. The third sensor identifies whether the seat belt is pulled out from the correct position. The third sensor outputs a signal of ,in Indicates the seat belt extension position signal. When the seat belt is correctly extended , when not pulled out correctly ;

[0096] S24, install a fourth sensor at the seat belt retractor position to detect the rewinding position of the seat belt. The fourth sensor monitors whether the seat belt is correctly rewound after insertion. The fourth sensor outputs a signal of ,in Indicates the seat belt rewinding status signal. When the seat belt is retracted correctly , when the roll is not correctly .

[0097] In this implementation, step S3 includes:

[0098] S31: The seat belt insertion state signal output by the first sensor , the seat belt tension signal output by the second sensor , the seat belt extension position signal output by the third sensor And the seat belt rewinding status signal output by the fourth sensor Transmitted to the central processing unit of the intelligent control system via the vehicle's internal communication bus;

[0099] S32, the intelligent control system pre-processes the received signal data, including signal normalization processing and noise filtering processing, to obtain a normalized data set ,in , , , They are respectively the normalized seat belt insertion state signal, seat belt tension signal, seat belt pull-out position signal and seat belt rewind state signal;

[0100] S33, using the semi-supervised ensemble model based on Scikit-learn to process the preprocessed data set;

[0101] S34. In the semi-supervised ensemble model, the semi-supervised K nearest neighbor model is used to preliminarily classify the data set, and the preliminary classification results are obtained by calculating the distance between the samples to be classified and the samples of known categories. ;

[0102] S35. Use the semi-supervised support vector machine model to optimize the preliminary classification results, construct the optimal classification hyperplane, further classify the samples, and obtain the final classification results. .

[0103] In this embodiment, step S34 includes:

[0104] S341, inputting the preprocessed data set into a semi-supervised K nearest neighbor model;

[0105] S342, calculate the samples to be classified With labeled samples And unlabeled samples The weighted Euclidean distance between them is calculated by combining the different weights of different sensor signals on the classification results, and the weighted formula is adopted:

[0106] ;

[0107] ;

[0108] in, represents the mth eigenvalue of the sample x to be classified, and the corresponding features include the normalized seat belt insertion state signal, seat belt tension signal, seat belt pull-out position signal and seat belt rewinding state signal; Indicates labeled samples The mth eigenvalue of ; Represents unlabeled samples The mth eigenvalue of ; Is with characteristics The relevant weight coefficient reflects the relative importance of the feature in the classification process; It is the high-order weight coefficient of the feature, which is used to deal with the nonlinear relationship between high-order features;

[0109] S343, calculate the preliminary classification result of the sample to be classified x according to the categories of the nearest k labeled samples and the potentially labeled unlabeled samples :

[0110] ;

[0111] in, represents the preliminary classification result of the sample x to be classified, u represents the number of unlabeled samples, is the weight associated with the i-th labeled sample, is the category of the i-th labeled sample (0 or 1), is the weight associated with the jth unlabeled sample, is the unlabeled sample obtained through model self-learning The latent class probability, represents the category probability of unlabeled samples obtained through self-learning, is the distance between the sample to be classified and the unlabeled sample calculated in S342, is a parameter used to control distance attenuation.

[0112] In this implementation, step S35 includes:

[0113] S351, the preliminary classification results Together with the preprocessed data set, it is input into the semi-supervised support vector machine model, where As the initial label of the semi-supervised support vector machine model, it helps the semi-supervised support vector machine model determine the initial classification boundary;

[0114] S352. In the semi-supervised support vector machine model, by constructing a classification hyperplane, different categories in the sample space are divided to maximize the interval between samples, and the optimization objective function is:

[0115] ;

[0116] in, is the normal vector of the classification hyperplane, b is the bias, is the slack variable for a specific sample, represents the slack variable of the entire sample set, and C is the penalty parameter; The weight coefficient for adjusting the consistency between the preliminary classification results and the final classification results is used to balance the maximization of the classification interval, the penalty for misclassified samples, and the consistency with the preliminary classification results;

[0117] S353. Construct Lagrangian function to solve optimization problem:

[0118] ;

[0119] in, is the Lagrange multiplier, represents the Lagrange multiplier vector of all sample points, is the actual category label of the sample, is the feature vector of the sample;

[0120] S354, by solving the Lagrangian function, combined with the preliminary classification results , get the optimized classification hyperplane, and use the hyperplane to finally classify the input samples to get the final classification result of the samples :

[0121] ;

[0122] in, is the final classification result of sample x, The function is used to determine the classification result. The output is 1 if the sample is normal, and the output is 0 if the sample is abnormal.

[0123] In this embodiment, step S4 includes:

[0124] S41. The final classification result after processing by the semi-supervised ensemble model Fusion analysis is performed with the pre-processed sensor detection data to determine the use status of the seat belt:

[0125] ;

[0126] in, is the comprehensive evaluation function after fusion analysis; is the fusion coefficient, which is used to adjust the weight ratio of sensor data and the final classification result; and are the linear weight and nonlinear weight coefficient of sensor data, respectively, reflecting the influence of each sensor signal in the overall judgment. and Represent linear and nonlinear mapping functions, corresponding to the preprocessed sensor signal data , The final classification result The weight coefficient of the factor is used to adjust its role in the overall evaluation;

[0127] S42: For the detection results of different sensors, the following judgments are made respectively:

[0128] The normalized seat belt insertion status signal Analyze to determine whether the seat belt is correctly inserted into the socket:

[0129] ;

[0130] in, and is the calibration factor, and It is the logarithmic probability of the insertion state signal and the Sigmoid function transformation result, reflecting the correctness of the seat belt insertion action;

[0131] The normalized seat belt tension signal Analyze to determine whether the seat belt is under stress:

[0132] ;

[0133] in, and is the calibration factor, and The square root and hyperbolic tangent function transformation results of the tension signal are used to evaluate the stress condition of the seat belt;

[0134] The normalized seat belt pull-out position signal Analyze to determine whether the seat belt passes through the correct pull-out position:

[0135] ;

[0136] in, and is the calibration factor, and They are the exponential and Sigmoid function transformation results of the pull-out position signal, respectively, to evaluate the correctness of the pull-out position;

[0137] The normalized seat belt rewinding status signal Analyze to determine whether the seat belt is retracted correctly:

[0138] ;

[0139] in, and is the calibration factor, and are the square and logarithmic transformation results of the rewinding state signal, respectively, to evaluate the rewinding action of the seat belt;

[0140] S43, each judgment result and the final classification result Combined with the above, a comprehensive judgment model is formed to comprehensively score the correctness of passengers wearing seat belts:

[0141] ;

[0142] Among them, S is the comprehensive score, , , , are the weight coefficients of insertion state, tension, pull-out position and rewinding state, respectively, to adjust the proportion of each score in the overall score. Evaluating functions for fusion analysis The weight coefficient of

[0143] S44: Determine whether the passenger wears the seat belt correctly based on the comprehensive score S and set a threshold :

[0144] ;

[0145] in, Passengers wear seat belts correctly. Indicates false wearing behavior or incorrect wearing.

[0146] An intelligent safety alarm system for seat belt false fastening monitoring, comprising:

[0147] Sensor modules, respectively used to detect the insertion state, tension state, pull-out position and rewinding state of the seat belt, each sensor module including one or more sensors;

[0148] The data processing module is connected to the sensor module, receives the output signal of the sensor, pre-processes the output signal, including signal normalization processing and noise filtering, and outputs the pre-processed signal data set;

[0149] The semi-supervised learning integrated model module, including the semi-supervised K nearest neighbor model and the semi-supervised support vector machine model, receives the output signal data set of the data processing module and processes it through the following steps:

[0150] The semi-supervised K nearest neighbor model performs preliminary classification on the signal data set and outputs preliminary classification results;

[0151] The semi-supervised support vector machine model optimizes the preliminary classification results, constructs the classification hyperplane, and outputs the final classification results;

[0152] The fusion analysis module receives the final classification results of the semi-supervised learning integrated model module and the pre-processed sensor detection data, generates a comprehensive score through fusion analysis, and determines whether the passenger wears the seat belt correctly;

[0153] The alarm module compares the comprehensive score generated by the fusion analysis module with the set threshold value. When the comprehensive score is less than the set threshold value, the alarm device is triggered to emit an audible, visual or vibration alarm to alert passengers or drivers.

[0154] The storage module is used to record and store anomaly detection results and alarm triggering information for subsequent analysis and reference.

[0155] Example 1: Example 1 was tested in City A. At 10:30 am on May 15, 2024, the testing team conducted a field test of the intelligent safety alarm system on the Fifth Ring Expressway in City A. The test vehicle was a family sedan equipped with the intelligent safety alarm system of the present invention. Passengers simulated a variety of seat belt wearing and fake fastening behaviors to verify the system's detection capabilities and alarm accuracy.

[0156] In the first scenario of the test, after the vehicle started, passenger A inserted the seat belt into the buckle, but did not fasten the seat belt around his body. Instead, he wrapped the seat belt around the seat back. At 10:32, the system detected that the seat belt insertion status signal output by the first sensor was 1, indicating that the seat belt had been inserted into the buckle, but at the same time, the system detected that the seat belt tension signal output by the second sensor was 0, indicating that the seat belt was not under stress, the pull-out position signal of the third sensor was 0, indicating that the seat belt was not pulled out from the correct position, and the rewinding status signal of the fourth sensor was 0, indicating an abnormal rewinding.

[0157] The system input the data into the semi-supervised learning integrated model, and the preliminary classification result was 0, indicating an abnormality. Subsequently, the semi-supervised support vector machine model optimized the preliminary classification result, and the final classification result was also 0, confirming that passenger A did not wear the seat belt correctly. The system immediately issued an audible and visual alarm to remind passenger A to put on the seat belt again. During this process, the system generated a detailed detection report, recording the output status of each sensor, the system's comprehensive score, and the alarm trigger time of 10:33.

[0158] In the second test scenario, at 10:45, passenger B deliberately pulled the seat belt out for a long distance, rewound it, and inserted it into the buckle. The operation simulated the situation of false wearing of the seat belt. The system first detected that the buckle signal of the first sensor was 1, indicating that the seat belt was inserted; the signal detected by the tension sensor was 0, indicating that the seat belt was not under force; the pull-out position sensor signal was 1, indicating that the seat belt was pulled out in the correct position, but the rewind sensor signal was 0, indicating that the seat belt was not rewound correctly. After the system analyzed the data, the preliminary classification result and the final classification result were both 0. The system determined that passenger B had falsely fastened the seat belt and triggered the sound and light alarm at 10:46.

[0159] In the third test scenario, the vehicle was driving on a complex mountain road at 11:05. Passenger C wore the seat belt correctly, and the system detected that the output signals of each sensor were normal: the buckle state signal was 1, the tension signal was positive, and the pull-out position and rewind state signals were both 1, indicating that the seat belt was used correctly. After the semi-supervised ensemble model processed the data, the preliminary classification result and the final classification result were both 1, indicating normal. The system did not issue an alarm and recorded the normal use of the seat belt.

[0160] To verify the effectiveness and superiority of the system, the test team conducted a comparative experiment. On the afternoon of May 15, 2024, the same test was conducted using a traditional seat belt alarm system. In 10 simulations of false-fastening behaviors, the traditional system failed to detect any abnormalities because it only relied on the buckle signal, and no alarm was triggered in all tests. However, the intelligent safety alarm system of the present invention successfully identified all false-fastening behaviors under the same test conditions, and the accuracy of triggering the alarm was 100%. In addition, in 20 normal wearing tests, neither the traditional system nor the system of the present invention generated false alarms, but the system of the present invention showed stronger adaptability in a variety of complex scenarios.

[0161] To further analyze the performance of the system, the testing team conducted statistics on the response time of the intelligent safety alarm system of the present invention. In 10 false system behavior detections, the average response time of the system from data acquisition to alarm triggering was 1.2 seconds, the shortest time was 0.9 seconds, and the longest time was 1.5 seconds. The system showed extremely high real-time performance. In 20 normal wearing tests, the system did not generate any false alarms, showing high precision and high reliability.

[0162] Through detailed field tests and data analysis, the present invention has demonstrated excellent detection capabilities and efficient real-time performance, and can accurately identify false alarms and issue alarms in a timely manner. Test results show that the present invention has significant advantages in complex driving environments, which not only improves the safety of passengers, but also significantly reduces the possibility of false alarms and missed alarms, showing extremely high practical value.

[0163] The present invention adopts an integrated model based on semi-supervised learning, including a semi-supervised K nearest neighbor model and a semi-supervised support vector machine model. The combined model can not only process the data of labeled samples and unlabeled samples, effectively improving the accuracy of classification, but also can use the information of unlabeled samples to train the model when the labeled data is insufficient, thereby improving the robustness of the model. The semi-supervised K nearest neighbor model is used for preliminary classification, and then combined with the optimization processing of the semi-supervised support vector machine model, the optimal classification hyperplane can be constructed, the accurate classification of complex samples can be achieved, and the situation that the passenger does not wear the seat belt correctly can be effectively identified, thereby triggering the alarm system.

[0164] The present invention further enhances the intelligence of the system through a fusion analysis model, which fuses the classification results processed by the semi-supervised learning model with the sensor detection data to form a comprehensive scoring mechanism. By setting thresholds, it can conduct a quantitative assessment of the passengers' seat belt wearing conditions in real-time monitoring and determine whether an alarm needs to be issued based on the comprehensive score.

[0165] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent safety alarm method for seat belt false fastening monitoring, It is characterized in that The steps include: S1, installing multiple sensors at the vehicle seat, the seat belt pull-out end and the retractor position, including a first sensor for detecting the seat belt buckle state, a second sensor for detecting the seat belt tension, a third sensor for detecting the seat belt pull-out position and a fourth sensor for detecting the seat belt rewinding position; S2, obtaining seat belt extension position data detected by the third sensor, obtaining seat belt buckle state data detected by the first sensor, obtaining seat belt rewinding position data detected by the fourth sensor, and obtaining seat belt tension data detected by the second sensor; S3, transmit all sensor detection data of S2 to the intelligent control system, and process the sensor detection data using a semi-supervised integration model based on Scikit-learn, in which a semi-supervised K nearest neighbor model and a semi-supervised support vector machine model are integrated; S4. In the intelligent control system, a semi-supervised integrated model is used to perform fusion analysis on the received sensor detection data to determine whether the passenger wears the seat belt correctly and whether there is any false fastening behavior, including determining whether the seat belt passes through the correct pull-out position, whether it is correctly inserted into the socket, and whether it is correctly rewound after insertion; S5. When it is determined that the passenger is not wearing the seat belt correctly or is pretending to wear the seat belt, the intelligent control system triggers the safety alarm system and sends an alarm signal through the vehicle audio, video or vibration device to remind the driver or passenger to correct the wearing behavior; S6. Record the abnormal detection situation in the system storage unit for subsequent analysis; The S2 step includes: S21, install a first sensor at the bottom of the vehicle seat to detect the state of the seat belt buckle, the first sensor detects the operation of inserting the seat belt into the socket, and the first sensor outputs a signal of ,in Indicates the seat belt insertion status signal. When the seat belt is inserted , when not inserted ; S22, install a second sensor in the middle or near the seat belt to detect the tension state of the seat belt, the second sensor generates a signal by measuring the tension of the seat belt, and the second sensor outputs a signal of ,in Indicates the seat belt tension signal. When the seat belt is under stress , when no force is applied ; S23, a third sensor is installed at the seat belt pulling end position, which is used to detect the pulling position of the seat belt. The third sensor identifies whether the seat belt is pulled out from the correct position. The third sensor outputs a signal of ,in Indicates the seat belt extension position signal. When the seat belt is correctly extended , when not pulled out correctly 0; S24, install a fourth sensor at the seat belt retractor position to detect the rewinding position of the seat belt. The fourth sensor monitors whether the seat belt is correctly rewound after insertion. The fourth sensor outputs a signal of ,in Indicates the seat belt rewinding status signal. When the seat belt is retracted correctly , when the roll is not correctly ; The S3 step includes: S31: The seat belt insertion state signal output by the first sensor , the seat belt tension signal output by the second sensor , the seat belt pull-out position signal output by the third sensor And the seat belt rewinding status signal output by the fourth sensor Transmitted to the central processing unit of the intelligent control system via the vehicle's internal communication bus; S32, the intelligent control system pre-processes the received signal data, including signal normalization processing and noise filtering processing, to obtain a normalized data set ,in , , , They are respectively the normalized seat belt insertion state signal, seat belt tension signal, seat belt pull-out position signal and seat belt rewind state signal; S33, using the semi-supervised ensemble model based on Scikit-learn to process the preprocessed data set; S34. In the semi-supervised ensemble model, the semi-supervised K nearest neighbor model is used to preliminarily classify the data set, and the preliminary classification results are obtained by calculating the distance between the samples to be classified and the samples of known categories. ; S35. Use the semi-supervised support vector machine model to optimize the preliminary classification results, construct the optimal classification hyperplane, further classify the samples, and obtain the final classification results. ; The step S34 includes: S341, inputting the preprocessed data set into a semi-supervised K nearest neighbor model; S342, calculate the samples to be classified With labeled samples And unlabeled samples The weighted Euclidean distance between them is calculated by combining the different weights of different sensor signals on the classification results, and the weighted formula is adopted: ; ; in, Represents the sample to be classified The mth eigenvalue of , the corresponding features include the normalized seat belt insertion state signal, the seat belt tension signal, the seat belt pull-out position signal and the seat belt rewinding state signal; Indicates labeled samples The mth eigenvalue of ; Represents unlabeled samples The mth eigenvalue of ; Is with characteristics The relevant weight coefficient reflects the relative importance of the feature in the classification process; It is the high-order weight coefficient of the feature, which is used to deal with the nonlinear relationship between high-order features; S343, calculate the samples to be classified according to the categories of the nearest k labeled samples and the potentially labeled unlabeled samples The preliminary classification results : ; in, Represents the sample to be classified The preliminary classification results of represents the number of unlabeled samples, is the weight associated with the i-th labeled sample, is the category of the i-th labeled sample (0 or 1), is the weight associated with the jth unlabeled sample, is the unlabeled sample obtained through model self-learning The latent class probability, represents the category probability of unlabeled samples obtained through self-learning, is the distance between the sample to be classified and the unlabeled sample calculated in S342, is a parameter used to control distance attenuation; The step S35 includes: S351, the preliminary classification results Together with the preprocessed data set, it is input into the semi-supervised support vector machine model, where As the initial label of the semi-supervised support vector machine model, it helps the semi-supervised support vector machine model determine the initial classification boundary; S352. In the semi-supervised support vector machine model, by constructing a classification hyperplane, different categories in the sample space are divided to maximize the interval between samples, and the optimization objective function is: ; in, is the normal vector of the classification hyperplane, b is the bias, is the slack variable for a specific sample, represents the slack variable of the entire sample set, and C is the penalty parameter; The weight coefficient for adjusting the consistency between the preliminary classification result and the final classification result is used to balance the maximization of the classification interval, the penalty of misclassified samples, and the consistency with the preliminary classification result; S353. Construct Lagrangian function to solve optimization problem: ; in, is the Lagrange multiplier, represents the Lagrange multiplier vector of all sample points, is the actual category label of the sample, is the feature vector of the sample; S354, by solving the Lagrangian function, combined with the preliminary classification results , get the optimized classification hyperplane, and use the hyperplane to finally classify the input samples to get the final classification result of the samples : ; in, For sample The final classification result is The function is used to determine the classification result. The output is 1 if the sample is normal, and the output is 0 if the sample is abnormal. The S4 step includes: S41. The final classification result after processing by the semi-supervised ensemble model Fusion analysis is performed with the pre-processed sensor detection data to determine the use status of the seat belt: ; in, is the comprehensive evaluation function after fusion analysis; is the fusion coefficient, which is used to adjust the weight ratio of sensor data and the final classification result; and are the linear weight and nonlinear weight coefficient of sensor data, respectively, reflecting the influence of each sensor signal in the overall judgment. and Represent linear and nonlinear mapping functions, corresponding to the preprocessed sensor signal data , The final classification result The weight coefficient of the factor is used to adjust its role in the overall evaluation; S42: For the detection results of different sensors, the following judgments are made respectively: The normalized seat belt insertion status signal Analyze to determine whether the seat belt is correctly inserted into the socket: ; in, and is the calibration factor, and It is the logarithmic probability of the insertion state signal and the Sigmoid function transformation result, reflecting the correctness of the seat belt insertion action; The normalized seat belt tension signal Analyze to determine whether the seat belt is under stress: ; in, and is the calibration factor, and The square root and hyperbolic tangent function transformation results of the tension signal are used to evaluate the stress condition of the seat belt; The normalized seat belt pull-out position signal Analyze to determine whether the seat belt passes through the correct pull-out position: ; in, and is the calibration factor, and They are the exponential and Sigmoid function transformation results of the pull-out position signal, respectively, to evaluate the correctness of the pull-out position; The normalized seat belt rewinding status signal Analyze to determine whether the seat belt is retracted correctly: ; in, and is the calibration factor, and are the square and logarithmic transformation results of the rewinding state signal, respectively, to evaluate the rewinding action of the seat belt; S43, each judgment result and the final classification result Combined with the above, a comprehensive judgment model is formed to comprehensively score the correctness of passengers wearing seat belts: ; Among them, S is the comprehensive score, , , , are the weight coefficients of insertion state, tension, pull-out position and rewinding state, respectively, to adjust the proportion of each score in the overall score. Evaluating functions for fusion analysis The weight coefficient of S44: Determine whether the passenger wears the seat belt correctly based on the comprehensive score S and set a threshold : ; in, Passengers wear seat belts correctly. 0 means false tie behavior or incorrect wearing.

2. An intelligent safety alarm system for seat belt false fastening monitoring, comprising the method as claimed in claim 1, It is characterized in that include: Sensor modules, respectively used to detect the insertion state, tension state, pull-out position and rewinding state of the seat belt, each sensor module including one or more sensors; The data processing module is connected to the sensor module, receives the output signal of the sensor, pre-processes the output signal, including signal normalization processing and noise filtering, and outputs the pre-processed signal data set; The semi-supervised learning integrated model module, including the semi-supervised K nearest neighbor model and the semi-supervised support vector machine model, receives the output signal data set of the data processing module and processes it through the following steps: The semi-supervised K nearest neighbor model performs preliminary classification on the signal data set and outputs preliminary classification results; The semi-supervised support vector machine model optimizes the preliminary classification results, constructs the classification hyperplane, and outputs the final classification results; The fusion analysis module receives the final classification results of the semi-supervised learning integrated model module and the pre-processed sensor detection data, generates a comprehensive score through fusion analysis, and determines whether the passenger wears the seat belt correctly; The alarm module compares the comprehensive score generated by the fusion analysis module with the set threshold value. When the comprehensive score is less than the set threshold value, the alarm device is triggered to emit an audible, visual or vibration alarm to alert passengers or drivers. The storage module is used to record and store anomaly detection results and alarm triggering information for subsequent analysis and reference.