Target Recognition And Tracking Based On Lidar Point Cloud Data

In recent years,with the continuous development of computer vision technology,research on the recognition and tracking of moving objects has become one of the hot topics.It also has a wide range of application scenarios and requirements in video surveillance,autonomous driving of autonomous vehicles,and medical diagnosis and treatment.At the same time,the field of deep learning is also developing rapidly,and target recognition based on deep learning is the current research and development trend.With the advancement of laser scanning technology,point cloud images obtained by laser radar scanning provide rich visual and geometric information of 3D objects,so the study of target recognition and tracking methods based on point cloud data has greater significance.This paper mainly studies the core issues of moving object recognition and tracking,which are divided into target recognition methods,target tracking algorithms and their specific implementations.Firstly,analyze the advantages and disadvantages of the traditional target recognition algorithm based on point cloud data.Since the real-time recognition effect of moving objects is not obvious,a point cloud target recognition method based on convolutional neural network with better recognition effect and higher accuracy is adopted..For the tracking part,in order to solve the problem that the effect of target recognition is not obvious and the tracking failure caused by multiple targets in the complex background of daily roads,the common tracking methods are analyzed,and the traditional tracking algorithms are improved to achieve The goal of improving tracking accuracy.The basic principle of the Camshift algorithm is explained in detail,and a Kalman filter algorithm is added to the tracking process to process the obtained target data to form a motion track,predict the next position of the target,and solve the tracking problem.Finally,the target tracking is combined with the self-stabilizing gimbal,and the relevant parameters of the target are input into the gimbal to achieve the effect of tracking as the target object moves.The experimental results show that after verification under the complex background of daily travel roads,the improved algorithm in this paper can effectively and accurately extract moving targets that need to be tracked from continuous frames of point cloud images,and it has high robustness.The problems of more target recognition and higher real-time requirements have been effectively solved.The results show that the accuracy of prediction and tracking is high,and the method is feasible and effective.