Research On The Detection Algorithm Of Fatigue Driving And Dangerous Driving Behavior Based On Deep Video

At present,China is a large automobile manufacturing country,and its car ownership ranks second in the world.The increasing number of vehicles not only brings convenience to travel,but also brings a series of problems.Among them,frequent traffic accidents have become the most important problem.The number of deaths caused by annual traffic accidents has increased year by year.According to the investigation report of the National Highway Traffic Safety Administration,in the accidents,the fatigue caused by fatigue driving Traffic accidents account for a large proportion.Therefore,fatigue driving has attracted the attention of the state and the government,and an effective and real-time fatigue detection system is urgently needed.This article analyzes and compares the existing fatigue detection and dangerous driving behavior recognition algorithms at home and abroad,combines system requirements and the latest technological achievements,and solves the difficult points that traditional color images cannot accurately detect face recognition,fatigue driving,and dangerous driving behavior when driving at night.A method for fatigue driving detection and dangerous driving behavior warning based on computer vision and Intel Realsense depth camera is proposed.This paper uses the LBP features of the face,and uses the cascade classifier of the AdaBoost algorithm to implement feature detection for face detection.In the face area,a combination of random forest and global linear regression algorithm is used to train the model and detect the 68 feature point of the face.Using feature point to detect the area of eye and mouth area;the closed or opened state of the eyes and mouth is determined by calculating the aspect ratios of the eye and mouth,ratios is calculating by the feature points,and using improved fatigue parameters based on PERCLOS to detect the driver's fatigue status.In this paper,infrared and depth images are used to detect drivers' dangerous driving behaviors,such as facing up,smoking,and making phone calls.Use the cascade classifier based on LBP features for side face recognition,calculate the time to continuously detect the side face in non-orthographic state,and detect the driver's focus state;obtain the depth of the face from the depth image,and use morphological processing and depth threshold segmentation methods,Remove the noise to retain the head depth image,detect the calling behavior according to the changes in the shape and area of the connected area of the face;detect the smoking behavior based on the changes in the mouth edge and depth,and make a real-time reminder when the above dangerous driving behavior occurs to ensure the driver Driving safety.The experimental results show that,compared with separate fatigue detection,combining fatigue detection and dangerous driving behavior detection can more effectively prevent drivers from fatigue driving and traffic accidents caused by dangerous driving behavior,and improve driver driving safety.