Research And Implementation Of Assistant Driving Safety Warning System Under Next Generation Network

The number of domestic cars is increasing,and the subsequent vehicle safety accidents are becoming more and more frequent.Advanced Vehicle Assistant Driving System(ADAS)is a technology that has developed rapidly in the field of vehicle applications in recent years and can be effectively used to avoid driving risks.Due to the complexity of the domestic driving road environment,the real-time detection and anti-collision warning of various types of target objects existing on the driving road are the key points to be solved.In addition,it is also a potential threat to the driver's dangerous behavior under abnormal conditions,so the detection of the driver's safe driving behavior is also an indispensable part of safe driving.Among them,the assisted driving system based on machine vision has always been the focus of current research.Through the use of deep learning related algorithms to real-time detection of driving,pedestrians,non-motor vehicles in the road environment to remind drivers of collision prevention,and for drivers Whether there are bad driving behaviors such as:fatigue driving,distracted driving,etc.Real-time detection and timely warning to ensure the safety of the driving environment.This article takes safety assisted driving as the starting point,and the main research work is as follows:Firstly,through the analysis of the forward collision warning function in ADAS,we have a deep understanding of the current mainstream deep learning target detection algorithms,quantify the target detection related algorithms and optimize the learning hyperparameters,and make Special roads data sets based on public data sets.And trained to obtain a special road target recognition model,which improves certain mAP accuracy.It also analyzes the related principles and algorithm about distancc measurement based on the machine vision.Based on the results of target detection,it performs stereo matching and distance detection and proposes an anti-collision warning model.Then analyze the basic principles and implementation of the driver's dangerous driving behavior detection function,train a deep learning face detection model based on the public data set,and combine the face key point detection algorithm to write a three-stage dangerous driving behavior based on face key points judging algorithms,judging dangerous driving behaviors,giving different degrees of dangerous driving warnings to different degrees of dangerous driving behaviors,and verifying with public data sets in real driving environments,the detection accuracy of dangerous driving behavior recognition can reach more than 95%.Finally,the deep learning algorithm is quantitatively transplanted to the embedded development platform terminal,with multi-threaded programming,the operation speed of the deep learning algorithm in the video is accelerated to 30fps to achieve real-time detection effect,and the application test of the vehicle networking cloud platform under IPv6 network is carried out to achieve the function of security monitoring.This paper aims at the research of collision warning system and dangerous driving behavior detection system based on machine vision,trains a special target detection model based on deep learning,and designs a set of collision warning and prevention terminal system suitable for assistant driving on embedded arm64 low-power processor after quantitative transplantation,and completes the network under IPv6 network combined with the cloud data platform of the Internet of vehicles Network communication,after testing in the actual environment,has completed the verification of the auxiliary driving function under the Internet of vehicles system,which can meet the general requirements of the auxiliary driving anti-collision function.