Research On Vehicle Detection And Collision Warning Method Based On Deep Learning

Vehicle detection and collision warning technology is one of the key technologies in the field of intelligent vehicle assisted driving.This technology can not only facilitate people’s travel,but also reduce the probability of traffic accidents.Therefore,it has important research significance.The driving environment of unstructured road is complex,the traditional vehicle detection method based on vision has the problem of insufficient collision warning information,which makes it difficult to achieve accurate and reasonable collision warning judgment in unstructured road scenarios.Therefore,in view of the driving characteristics of unstructured roads,this thesis further studies the technology of vehicle detection and collision warning by using the method of deep learning.Firstly,the FBS_yolo2 vehicle detection algorithm is proposed based on visual angle of vehicle,and the vehicle detection model is constructed based on FBS_yolo2 algorithm.By analyzing the application of the main deep learning target detection algorithm in vehicle detection,YOLOv2 algorithm is determined to be most suitable for vehicle detection under the condition of vehicle camera.According to the driving characteristics of unstructured road,a vehicle classification method based on visual angle of vehicle is defined,the FBS_yolo2 vehicle detection algorithm is proposed based on YOLOv2 framework,and the vehicle detection model is constructed based on FBS_yolo2 algorithm.Compared with the traditional vehicle detection model,this model has two functions of vehicle detection and vehicle visual angle recognition.This model is more difficult,and the obtained collision warning information is more sufficient,which is more conducive to the research of the subsequent collision warning method.Secondly,the vehicle collision warning method based on information fusion is proposed in unstructured road scenarios.According to the vehicle speed,the sub-model of dangerous driving area is established,and the vehicle information obtained from the sub-model of dangerous driving area and the sub-model of vehicle detection is fused to construct the vehicle collision warning model.The vehicle collision warning strategy for unstructured road is designed.Through comprehensive analysis of the visual angle information and location information of the detected vehicle,the corresponding warning level is obtained,the corresponding warning operation is carried out,and finally a new method of vehicle collision warning is formed.Finally,an experiment is designed to verify the superiority of the vehicle collision warning method,and a software is developed to show the effect of vehicle detection and collision warning.The experimental results show that the vehicle collision warning method based on information fusion takes 25 ms per frame on the computer and 30 ms per frame on the NVIDIA Jetson TX2 development board,which meets the real-time requirements.The accuracy rate of collision warning is more than 94%,which can detect the vehicle,recognize the visual angle of vehicle and complete the collision warning judgment accurately.The software developed on the computer plays an important role in guiding the product development of research results.The vehicle detection and collision warning method based on deep learning solves the problem of insufficient collision warning information in unstructured road environment.It can detect vehicle,recognize its visual angle accurately on unstructured road and make reasonable collision warning judgment.It expands the application scope of vehicle detection and collision warning technology,and has good theoretical significance and important practical value.