Research On Automatic Identification System Of Driving Behavior Based On GPS And IMU

With the rapid development of the automotive industry and the increase in car ownership,issues such as traffic safety,energy conservation,and environmental protection are particularly interesting.Among them,traffic accidents are usually caused by drivers,vehicles,road conditions and weather,and the dangerous driving behavior of drivers is the main cause of traffic accidents.The main work of this paper is to design a reasonable recognition model,analyze and recognize the driving behavior,and use this as the core to develop an automatic recognition system applied to the back-end of the server to reduce the occurrence of traffic safety accidents.First,two driving behavior recognition models based on a single sensor are designed.By analyzing the change law of GPS and IMU data in the public UAH-Drive Set data set,it is proved that GPS and IMU can effectively reflect the actual running trajectory of driving behavior.Based on this conclusion,this paper designed two recognition models: 1)GPS-based recognition model: first extract the time-domain features of GPS data,and then use a classification algorithm to classify;2)IMU-based recognition model: first use BH-SNE(Barnes-Hut stochastic neighbor embedding)algorithm to reduce the dimension of IMU data,and then use RBF network to classify.The simulation results of MATLAB show that both models are feasible,and the accuracy are 76.1% and 76.8% respectively.Then two driving behavior recognition models based on multi-sensor fusion are designed.To solve the problem of low accuracy of the recognition model based on a single sensor,this paper combines two single sensor models through a feature-level fusion algorithm,and designs a feature-level fusion model: a recognition model based on BH-SNE.The simulation results show that,compared with the single sensor model,the recognition effect of the model is improved,and the accuracy has reached 89.7%.However,due to the computational complexity of the model,its application in actual identification systems has great limitations.In order to further improve the practicality of the model,this paper replaces the BH-SNE algorithm in the fusion model with a time-frequency domain feature extraction algorithm,and then proposes an improved model: a recognition model based on GPS and IMU time-frequency domain features.The time-frequency domain features of the improved model include dimensional features,dimensionless features and frequency domain features.The simulation results show that the accuracy of the improved model is higher than the traditional BP neural network,support vector machine and the new deep forest algorithm,and the accuracy is 84.5%.Finally,the improved model is re-implemented in Python,and combine QT,multi-threading and signal slot technology to complete a driving behavior automatic recognition system with a visual interface.The system has functions such as category display,video playback and waveform drawing.The test results of the system show that the system can effectively recognize driving behaviors under the premise of meeting real-time performance,and has the advantages of low cost and low calculation complexity.