Research On Recognition Methods Of Driving Postures Based On Images

With the rapid development of economy,the number of motor vehicles increases rapidly.The conflict between traffic conditions and traffic flow becomes increasingly prominent.Traffic accidents occur frequently and cause heavy casualties and property losses.Researchers generally hold the view that more than 80% of road traffic accidents are caused by improper driving behavior.Thus it's particularly important to monitor and identify driver's driving behavior and correct or intervene risk behavior drivers are implementing.In view of this,this paper studies driver's real-time detection and recognition methods of driving behavior based on images,in order to enhance the accident prevention ability of vehicle active safety system and driving safety alerting system.Firstly,the detection method of Driving Posture Area(DPA)based on Deformable Part Model(DPM)was studied.Comparison and analysis of detection methods of DPA based on HOG feature and SVM classifier,Haar feature and Adaboost classifier,and LBP feature and Adaboost classifier were made.The experimental results showed that the method based on DPM was superior to the other three methods,and precision reached 97.5%.Secondly,this paper studied the driving posture recognition method based on Global DPM Features.The recognition methods of driving posture based on skin color feature and Random Forrest classifier,HOG feature and SVM classifier,Gabor feature and SVM classifier,LBP feature and SVM classifier were compared and analyzed.The experimental results showed that the real-time performance and precision of the proposed method were better than the other four methods,and its detection precision reached 92.5%.Finally,this paper proposed a driving posture recognition method based on Local DPM Fusion Feature.The method included the definition of Driving Posture Core Area(DPCA),DPCA detection,DPM scoring model building,Local DPM Fusion Feature extraction,and classification of driving posture using the RBF kernel Support Vector Machine(SVM).Based on Kaggle image sets and SEU image sets,comparative experiments were carried out.The experimental results showed that driving posture recognition based on the DPCA was better than that based on DPA,and its recognition precision reached 94.38%.