Development And Research Of Driving Behavior Detection System Based On Pressure Cushion

The frequent occurrence of traffic accidents not only jeopardizes the safety of life and property of drivers and other pedestrians,but also seriously affects the stability of the society.Drivers' personal factors account for more than 70% of all traffic accidents and are the main cause of accidents.Therefore,the importance of conducting research on personal driving behavior is self-evident.Because of the complexity of the driver's behavior and the limitations of the driving space,the current driving behavior detection is difficult,the testing content is single,the testing equipment is expensive and has obvious defects,also the installation space is limited,and the practical promotion value is not high.In order to solve the above problems,this thesis builds a smart Seat Cushion Pressure Collection System based on car seats,extracts pressure changes and relevant characteristics of driving behavior,develops and designs a variety of project detection functions,and combines machine learning algorithms to detect driving behaviors.The classification system was designed,and the system was verified through real-vehicle driving and driving simulator experiments.Finally,a driving behavior detection system with high detection accuracy,comprehensive detection function,and low non-intrusive driving behavior was realized.The specific research content and results of the paper are summarized as follows:(1)A multifunctional driving behavior detection system has been designed.Through the research and analysis of driving habits in daily life,a variety of driving recognition functions are realized,including personal identification,correct seating before driving,inappropriate driving behavior during driving,and five kinds of single driving operations.The brakes in the driving operations are further differentiated to identify sudden braking and normal braking,enabling detection of driver behavior from multiple angles.(2)The driving behavior detection system was designed and constructed.From the perspective of improving driver's driving safety,the overall design requirements and major research ideas are clearly defined,and the overall architecture and workflow of the system are introduced.The cushion pressure collection system is described fromthe three levels of the perception layer,the transmission layer and the application layer,completing the construction of the cushion prototype and the choice of communication methods.Combining Matlab and Anaconda,the software design of driving behavior detection system is based on the assurance of signal accuracy.Real-time acquisition and storage of pressure signals is realized.Finally,the alarm function is realized through the mobile phone APP and bracelet vibration.(3)Using a variety of machine learning algorithms to achieve the algorithm design of driving behavior detection system.First,the overall framework of the driving behavior detection algorithm is defined,which mainly includes the preprocessing algorithm,feature extraction,classification algorithm and the final driving behavior detection accuracy.The preprocessing algorithm mainly includes the design of FIR filter and the sliding time window function to remove interference signals such as car bumps;the driver's corresponding action changes are analyzed for different recognition functions,and SSV,RSV,NAS,C(X,Y),?,?,RMS,STD,SVC characteristic parameters are extracted;based on seven kinds of machine learning algorithms,the establishment of a driving behavior classifier was completed,and three performance evaluation parameters Precision,Accuracy,and F-measure of the classifier were calculated by using a 10-fold cross validation method.The effectiveness of the classifier was verified.(4)Experiments are designed to verify and evaluate several key issues and classifier performance.The experimental design includes driving simulator data acquisition experiments,real-vehicle driving data acquisition experiments,personal identification,proper seating,improper driving behavior,and single driving operation experiments;the evaluation experiment includes the evaluation of the effects of the two kinds of sensor deployment solution and the real-vehicle driving experiment evaluates the pre-processing effect;the verification experiment includes the verification of the relationship between the eigenvalue and the driving behavior and the accuracy calculation of the classifier.The experimental results show that the system achieves accuracy of 97%,99%,100%,92%,and 76% respectively in my identification function,correct seating recognition,improper driving behavior identification,driving operation identification,and emergency braking recognition.The accuracy reached 84%,basically reaching the expected effect of the experiment.