Study On Fatigue Driving Detection Algorithm Based On Deep Belief Network

Continue to increase rapidly in recent years,motor vehicles,a large number of traffic accidents happened to countries in the world caused huge casualties and wealth loss.Appeared in the process of driving fatigue factors be induced or lead to one of the important reasons of the traffic accident.Through the driver fatigue detection technology,real-time to evaluate driver's state,judge whether the driver fatigue.Based on this technology,fatigue testing equipment to the driver fatigue early warning,to reduce the number of traffic accidents.Therefore,fatigue driving detection technology research is of great significance to improve the road traffic safety.In this paper,based on the non-contact fatigue test method,this study proposes a framework for recognising driver drowsiness expression by using facial dynamic fusion information and a Deep Belief Network(DBN)to address the aforementioned problem.The main research work of this paper are as follows:(1)In this paper,We propose A framework of driver fatigue expression recognition,and the landmark and texture of face and mouth were used as characteristics of fatigue detection.Using the viola-jones face detection algorithm and the hierarchical linear regression method of face alignment,video and the landmark in the image were extracted from the coarse to the thin.(2)We presented based on the DBN of fatigue test method,the facial texture and landmark extracted from image sequence,using Restricted Boltzmann Machine(RBM)into DBN,the data processing fusion with DBN,in order to improve the driver fatigue detection performance.(3)The experiment verifies the effectiveness of using fusion information(landmark and texture)and DBN.The influence of different parameters on the experiment,such as facial subregion,hidden layer number in DBN and hidden unit number,temporal resolutions,etc.It also compares the effects of using single information and fusion information,video based method and image based method on experimental results.