Research On Forward Vehicle Ranging System Based On Deep Learning And Monocular Vision

With the rapid development of economy,vehicle driving safety has become a safety issue of great importance today.Therefore,this paper proposes a vehicle ranging and early warning system based on monocular and vision technology.Firstly,the system uses the improved algorithm based on yolov5 s to identify the vehicles that need to be detected and imaged,and then uses the ranging algorithm for the target based on this recognition to achieve the ranging effect of the target.Finally,the ranging system is transplanted on the development board to test the ranging accuracy.The main contents of this paper also include :(1)Aiming at the problems of model training gradient disappearing,small object target recognition difficulty and poor recognition accuracy caused by boundary box regression function in yolov5 s algorithm.Based on the yolov5 s algorithm,this paper uses the ELU activation function to replace the original activation function of the algorithm.Then add the attention mechanism module to the backbone network of yolov5 s algorithm to strengthen the feature extraction of small and medium-sized objects.Finally,ciou is used＿Loss function replaces the original regression function of yolov5 s,which improves the convergence rate and measurement accuracy of the loss function.In this paper,relevant experiments are carried out on the self-built data set.The experimental results show that compared with the previous algorithm,the modified algorithm map is improved by 3.1%,the convergence rate is improved by 0.8%,and the loss is reduced by 2.5%.(2)Aiming at the problems of large error and low robustness of the traditional geometric vehicle ranging model,based on the traditional model,a vehicle feature point ranging model is proposed in this paper.The model uses a monocular camera to extract the key points based on the image and generates the spatial information of the vehicle for ranging.Finally,the QT test platform is built to realize the debugging of the algorithm and the visualization of ranging results.The results show that compared with the traditional ranging model,the improved algorithm improves the ranging accuracy by 3%.With the increase of the test distance,the error of the ranging algorithm is stable at 4%,which can fully meet the actual needs.(3)This paper uses NVIDIA jet son Xavier NX development board as the carrier and completes the transplantation of the improved vehicle detection algorithm and vehicle distance detection system.At the same time,fully considering the needs of anticollision early warning,the system designs the vehicle distance alarm system.To test the ranging ability of the vehicle target and the measurement accuracy of the front vehicle distance in the system proposed in the paper,different test scenarios relative to the prohibited state and motion state are also set.The experimental results show that the embedded development board of the system proposed in this paper has good vehicle recognition ability and ranging ability.Considering the above factors,the front vehicle ranging system proposed in this paper can meet the expected requirements in recognition accuracy and early warning effect,and has practical significance and application value.