Research On Driver’s Face Detection And Tracking Based On Convolution Neural Network

With the development of road traffic and the increase of vehicle population,road traffic safety has been an acute problem,in particular the high incidence of drowsiness driving,the contradiction of human,vehicle and road has been a pressing issues.Driver’s fatigue monitor system is using the visual information technology such as detecting the driver’s face and tracking,when the system detected driver fatigue driving,remind the driver to rest,avoiding to fatigue driving and dangerous driving.And The driver face detection is an important component of the system.This paper analysis and summary the existing driver’s face detection technology,mainly explores driver’s face detection based on deep convolution neural network,and studying tracking based kernel correlation filters and proses multi convolution feature connected detection algorithm and kernel correlation filters tracking algorithm that used re-detection mechanicsEfficient subimages of detecting generating is the key of detection algorithm,in order to improve the detection precision,using the multi-scale sliding window of generating the subimages will make time wasting to detecting a driver’s face.This paper used the region proposal network to generate the subimages that decreased the number of subimages.In order to learn more feature from training samples,design a skip layer connection network to extract face feature.Meantime,using the L2 Norm way to normalize the weight before skip-layers connecting,this can make detection better.Designing an effective way to tracking the face based on kernel-correlation tracking algorithm.The input is the left-top coordinates,width and height of face,the output is the predicted position of face in a new frame.For driver’s face are susceptible to the effect of illumination,occultation,deformation in the tracking process,we proposed a face re-detection method based online learning.Establishing training dataset online,and training a random fern classifier,re-detection face position when the confidence of tracking predicted position lower the pre-defined threshold.