Identification Methord Of Vehicle Driver Manual Distraction Based On Machine Vision

With the continuous improvement of car ownership in our country,the incongruous development between road space and traffic demand has contributed to the worsening of traffic environment and increasingly severe traffic safety,which has seriously threatened drivers’ life safety and property safety.According to relevant research results,manual distracted driving behavior of car drivers is still one of the main reasons for frequent traffic accidents.Therefore,it is of great significance to accurately identify car drivers’ distracted driving behaviors to reduce the incidence of traffic accidents.The existing methods of identifying car drivers’ manual distraction did not fully consider the information of the car driver’s body joints.Based on this,by making full use the data information of the car drivers’ body joints,an identification method of the car drivers’ distracted driving behaviors based on machine vision was proposed in this paper.The research contents of this study are mainly as follows:(1)Twenty experimental drivers were recruited for real vehicle driving experiments and virtual driving experiments,causing that experimental data were collected.Then,the experimental data were manually labeled.Finally,the data set of distracted driving behavior was constructed.(2)The Lightweight Open Pose network was trained and was used to extract the car driver’s body joints,so that the location information of the driver’s body joints could be extracted and saved in virtual driving environment and real vehicle driving environment.Then,several characteristic parameters such as the car driver’s limb angle and Euclidean distance between joints were extracted in order to construct the car driver manual distraction identification model by using multi-feature parameters.(3)The data were preprocessed by moving average filter(MAF)and factor analysis method.By using particle swarm optimization algorithm(PSO),the smoothing factor of probabilistic neural network(PNN)was iteratively optimized,the optimal smoothing factor was obtained,resulting in that the car driver manual distraction identification model of PSO-PNN was constructed.In order to verify the superiority of the constructed PSO-PNN model,it was compared with the original PNN and the BP neural network optimized by PSO.Finally,the experimental results show that the accuracy of PNN identification model optimized by PSO increases by8.29% to reach 91.5%.This study demonstrated the feasibility of machine vision identification of distracted driving behavior of car drivers by using the characteristic information of the body joints of car drivers,and provided support for the application of distracted detection and identification of car drivers from a certain angle and to a certain extent.