Research On Vehicle Status Recognition And Early Warning Based On Machine Vision

With the rapid growth of the number of cars,traffic accidents occur frequently.This has made people pay more attention to the safe driving of cars,scholars have done a lot of research on intelligent transportation systems,and got a rapid development.The function of the intelligent transportation system is mainly to control the traffic flow through the road monitoring system and traffic lights,thereby reduces the incidence of traffic accidents.However,the occurrence of traffic accidents is mainly caused by the collision of vehicles and vehicles.Therefore,it is very important to the vehicle to sense the environment in which it is located.Vehicle detection technology is one of the key technologies of vehicle safety technology and is of great significance to vehicle safety.At present,there are many anti-collision products of the market,such as the ADAS system developed by Mobileye and the unmanned system developed by Tesla,but they are all relatively expensive and not diffusely popularized.In this context,this paper proposes an application study based on machine vison of vehicle status recognition and early warning.The video image of the road is collected by using a camera with a lower cost than the radar,and then the video image is detected in real time by the vehicle state recognition model,and the voice alarm is used to remind the driver,thereby achieving the purpose of safe assisted driving.The main contents and results from this study are as follows:1)This paper is based on the Adaboost algorithm in the machine learning method and the YOLOv3 algorithm in the deep learning method.Then,through the collection of on-site traffic vehicle picture,a training data set is constructed.Uses the Adaboost algorithm in the traditional machine learning method to train the vehicle tails detection model and uses the YOLOv3 algorithm in the deep learning method to train the vehicle tail detection model,and evaluate and test the accuracy of the model.2)After the model detects the tail of the vehicle,the taillight area of the tail of the vehicle is positioned.Then,it is judged whether the taillight is turned on by threshold dividing the area where the tail light is located.Therefore,a method of recognizing the driving state of the vehicle is proposed,and a voice alarm system for alerting the driver is designed and developed.3)The vehicle states recognition and early warning model built by YOLOv3 algorithm and the vehicle state recognition and early warning model built by Adaboost algorithm are compared.The results show that the model accuracy,real-time performance and stability of the YOLOv3 algorithm is better than that of the Adaboost algorithm.Moreover,the model accuracy of YOLOv3 algorithm is 84.6% and basically meet the requirements.In summary,the vehicle states recognition method proposed to this paper is feasible.