Research On Detection Of Distracted Driving Behavior Based On Deep Learning

More than 90% of all traffic accidents are caused by drivers,and distracted driving is the main part.Therefore,in order to reduce the occurrence of traffic accidents,it is of great significance to monitor and alarm the distracted state of drivers.At present,among the existing detection methods of distracted driving behavior,the most promising and reliable one is based on machine vision.However,from the actual effect of the existing detection methods,we can see that there are still many problems,such as low detection accuracy,poor real-time performance and greatly affected by the environment.Aiming at the problems of detection accuracy and speed,this thesis studies the detection of driver distraction behavior based on deep learning.A new detection method is proposed,and a real-time detection system of distracted driving behavior is built,which can detect the two distracted behaviors of drivers: calling and smoking.Considering the complex and changeable driving environment,this thesis firstly preprocesses the collected cab image to obtain a higher resolution image,which mainly carries out the operations of driving region clipping,normalization,histogram equalization and so on.Then,the improved multi-task Convolutional Neural Network(MTCNN)algorithm is used to detect the driver’s face to obtain the face frame and five feature points.According to the results of face detection,this thesis discriminates the distracted behaviors,which include calling and smoking.The detection method of phone calling behavior adopts the "and" operation between handheld phone behavior and speech behavior,among which the former is discriminated based on Mobile Net V3 network,and the latter is based on the difference of mouth width to height ratio vibration.Smoking behavior was detected by using lightweight convolutional neural network S-CNN.Finally,the final distracted driving behavior detection result is obtained by fusing all detection contents.In order to verify the detection effect of the algorithm in the actual driving environment,this thesis designs a distracted driving behavior detection system based on CEVA DSP.The results show that the detection accuracy of the algorithm in this thesis is 93.3%and 92.4% respectively for drivers’ calling behavior and smoking behavior,and the average detection time for a single frame is about 290 ms.Therefore,it can give the driver timely and accurate distraction warning in the actual driving process,which has practical application value.