Font Size: a A A

Design Of Driver Behavior Analysis System Based On Computer Vision

Posted on:2020-11-13Degree:MasterType:Thesis
Country:ChinaCandidate:H S XuFull Text:PDF
GTID:2432330590985575Subject:Control engineering
Abstract/Summary:PDF Full Text Request
A large part of traffic accidents are caused by the driver's dangerous driving behavior,including fatigue,phone calls,smoking and long-term deviation from the line of sight.This paper uses computer vision technology to design a real-time analysis system of driver behavior that can be operated on the general vehicle-mounted platform to supervise such dangerous driving behavior.This paper introduces the development and application of computer vision technology in the field of driver behavior analysis.The image data of the driver's seat area is acquired by the camera,and the driver's dangerous driving behavior is detected in real time through digital image processing technology.The main research contents of this paper are as follows:(1)We have used the MTCNN(Multi-task Cascaded Convolutional Networks)algorithm based on deep learning to achieve driver face detection and facial feature point location.By training the face detection model for the cab,the robustness of face detection to the environment is improved,so that we can obtain the exact position of the driver's face in the image.The NCNN library is used to call the detection model of the detection algorithm on the vehicle platform,and good detection results are achieved.(2)By segmenting the detected face images with feature points,the region images such as eyes,mouth and ears are obtained.We use the combination of HOG(Histogram of Oriented Gradient)features and SVM(Support Vector Machine)to model and classify the driver's dangerous driving behavior,and then realize the detection and analysis of the above various dangerous driving behaviors.At the same time,a detection algorithm for judging the lens occlusion is designed.(3)We test the driver behavior analysis system in the actual environment.Finally,an analysis and detection system with high detection rate,low false detection rate,fast running speed and less occupation of hardware resources is realized to meet the project requirements.In summary,this paper uses the MTCNN face detection and feature point localization algorithm based on deep learning to realize the detection of the driver's face.Then the HOG feature and SVM are combined to analyze the driver's behavior.In the end,we realized a high-performance driver behavior analysis system that meets the needs of practical applications.
Keywords/Search Tags:Computer vision technology, Behavior analysis, Convolutional neural network, Machine learning, Vehicle-mounted platform
PDF Full Text Request
Related items