Driver’s Abnormal Behavior Recognition And Early-Warning Based On Computational Vision

The rapid development of motor vehicle industry not only provides convenience for people’s life,but also brings huge potential traffic hazards.Frequent traffic accidents seriously endanger people’s property safety and life safety.Abnormal driving behavior is one of the main causes of traffic accidents.Therefore,in order to ensure driving safety and avoid traffic accidents,this paper develops a set of high accuracy and real-time driver abnormal behavior recognition and early-warning system based on Computational Vision,which can intelligently monitor and evaluate driver driving status and timely warn abnormal behavior.The abnormal driving behavior identified by the system includes fatigue driving behavior,smoking behavior and drinking behavior based on blinking frequency and yawning behavior.The function modules of the system mainly include target detection module,target classification module,abnormal behavior recognition and warning module.The main research contents are as follows:(1)Target detection moduleThe network model theory and network structure related to MobileNet-SSD are introduced in detail.By analyzing the network feature map to locate the redundant network layer in the MobileNet-SSD model,the MobileNet-SSD model is compressed and optimized by pruning and quantifying strategies,and the SU-MobileNet-SSD model is obtained.Through experiments,the superiority of SU-Mobile Net-SSD algorithm is proved from two aspects of accuracy and speed,and the detection of face,water cup and eye is completed.(2)Target classification moduleCombining with the theoretical knowledge of convolution neural network model,this paper builds the eye state classification network model.In order to enhance the feature extraction ability and robustness of the network,optimization strategies such as data augmentation,ReLU activation function,local response normalization and Dropout are adopted,and the training strategy of "emergency termination" is used to accelerate the convergence speed of the network and effectively avoid over-fitting.Through experiments,the advantages of the eye state classification model designed in this paper are proved from the convergence speed,accuracy and speed of the network model.(3)Abnormal Behavior Recognition and Early-Warning ModuleThe realization of abnormal behavior recognition and early-warning module is based on high precision and fast target detection algorithm and target classification algorithm.Among them,the recognition of fatigue driving behavior based on eye movement feature is realized by using the variant PERCLOS algorithm and the number of blinking frames at a time.The behavior decision-maker is designed by analyzing the action characteristics of mouth,cigarette,water cup and the position relationship with face.The method realizes the recognition of yawning,smoking and drinking behavior.Finally,according to the degree of urgency of driver’s abnormal behavior,the abnormal behavior grade model is established,and the effectiveness and practicability of the system are proved by experiments.