Research On Detection Method Of Vehicle Dangerous Behavior Based On Vehicle Motion

In this study,fatigue driving,distracted driving and illegal driving are defined as dangerous vehicle behavior,which bring collision risk and abnormal vehicle motion.Dangerous vehicle behavior is one of the major contributing factors for traffic accidents.Driver Assistance Systems is considered as an effective method to prevent traffic accidents and improve vehicle safety.Thus,developing detection method for dangerous vehicle behavior is very important for driver assistance systems.Detection method for dangerous vehicle behavior is studied based on vehicle motion in this study.Firstly,vehicle information acquisition system is designed.Internal and external information acquisition systems are developed based on the characteristic of vehicle motion under different kinds of dangerous vehicle behavior.And the experiments of dangerous vehicle behavior are carried out to collect vehicle motion features.Secondly,a feature selection algorithm is proposed to select the vehicle motion information.Based on information theory,the characteristics of relevance and redundancy among features are analyzed.Thus,a novel feature selection algorithm is proposed.Mathematical meaning of the proposed algorithm is proved with information theory.And redundancy-complementariness dispersion is defined to improve the proposed algorithm.To illustrate the effectiveness of the proposed algorithm,experiments are applied with four frequently used classifiers on ten datasets.The results verify the superiority of the proposed method compared with seven representative feature selection algorithms.Based on the proposed feature selection algorithm,the method for detecting of distracted,fatigue and illegal driving is then studied.Thirdly,the common features of vehicle motion under distracted and fatigue driving are extracted.And the detection method is developed based on the common features.The proposed feature selection algorithm is used to select the significant contribution of the vehicle motion features for detecting distracted and fatigue driving.Thus,Wilcoxon rank sum test is employed to analyze the common features of vehicle motion under distracted and fatigue driving.In order to identify a time window for effectively extracting the common features,a double-window method is used.Then,detection method for dangerous vehicle behavior is studied based on support vector machine and particle swarm optimization algorithm.Fourthly,detection method of illegal driving is proposed via sparse reconstruction.In this study,the Least-squares Cubic Spline Curves Approximation(LCSCA)is used to normalize vehicle motion features in order to keep the same dimension of the vehicle motion features.The-minimization problem of sparse reconstruction is relaxed to the minimization problem(0 < p < 1).A hybrid algorithm Orthogonal Matching Pursuit-quasi-Newton(OMPN)is proposed to effectively find the sparse solution.Then,a Sparse Reconstruction and Similarity based Trajectory Classifier(SSM)is developed to detect illegal driving.Finally,detection method of illegal driving is improved by a hybrid kernel function.A polynomial kernel function and a gauss kernel function are used to establish a hybrid kernel function,which is satisfied with Mercer's theorem.Vehicle motion features are transferred to higher space with hybrid kernel function.A Hybrid Kernel Orthogonal Matching Pursuit(HKOMP)is developed to effectively find the sparse solution.Then,a Hybrid Kernel Similarity Sparsity Model(HKSSM)is proposed to detect illegal driving.The results of this study can effectively promote the development of driver assistance systems and improve vehicle safety.