A Method Of Pedestrian And Vehicle Detection Based On Machine Learning

With the continuous progress of society,people have a greater demand for the safety and functionality of vehicles.The discussion on autonomous driving is increasing,more and more researchers are investing in related research.The development of artificial intelligence has brought new vitality to the research of autonomous driving.It is believed that in the near future,the real autonomous driving will be realized and popularized.Among the tasks related to autonomous driving,the detection of pedestrians and vehicles is a very important task,which is of great significance for realizing vehicle control and ensuring the safety of autonomous vehicles.Therefore,at the beginning of the article,the development of autonomous driving and target detection is reviewed,as well as the current research progress.Next,this article starts with the research of target detection algorithms,analyzes the current advanced target detection algorithm structure and detection process,and summarizes the advantages and disadvantages of various algorithms.Especially for the YOLOv3 target detection algorithm,including the detection process of the YOLOv3 algorithm,network structure and loss function design.Then there is a chapter about optimization and improvement on the basis of the YOLOv3 algorithm,aiming to further improve the autonomous driving environment The accuracy and speed of target detection.In the theoretical study part,this article first explains the structure and operation flow of the neural network.Secondly,two types of target detection algorithms based on neural network models in target detection are introduced and analyzed,and their respective advantages and disadvantages are summarized.The detection mechanism of the yolov3 algorithm is also explained in detail.The detection accuracy and speed of target detection algorithms have always been the main metrics of algorithm performance.How to obtain algorithms with higher accuracy and faster speed has always been the goal of researchers.In the algorithm improvement part,this paper proposes a light network structure,at the same time,a more complete CIOU loss function is used in the design of the loss function to detect the position of the frame.Experiments have proved that the above improvements are of great help to the speed and accuracy of the target detection algorithm.In the final analysis of the experimental results,the KITTI data set and the PASCAL VOC data set are selected to compare and evaluate the performance of the improved algorithm.And binocular stereo camera is also used to obtain sample data in actual driving,based on the school’s surrounding environment,the obtained samples are also used to evaluate the improved algorithm performance.The experimental results prove that the improved algorithm has improved speed and accuracy,whether it is based on the evaluation of the data set or in the detection task of the actual scene.These experimental results prove the effectiveness of improved method proposed in this article,which provides a reference for the research of target detection algorithms.