Target Tracking And 3D Reconstruction Based On Video Stream

Human beings perceive the world through their eyes,so computer vision has always been a hot topic of people's research,and it has been used to serve the good life of human beings.Among them,3D reconstruction technology based on feature matching is the most cutting-edge research direction in computer vision,which is also widely used in traffic monitoring,medical imaging,address survey,collection exhibition and other fields.It generates matching descriptors of images by feature extraction algorithm,uses the correlation of descriptors between different images for correlation matching,and finally completes two-dimensional to three-dimensional image reconstruction by calculating camera related parameters.All technologies related to computer vision are inseparable from feature matching,and the quality of feature matching results will determine the results of computer vision technology.Starting from feature matching,this paper studies the algorithm of target tracking and 3D reconstruction as follows:1)In order to solve the problem that the traditional feature matching algorithm can not match the image perfectly because of the multi-scale,noise,light intensity and rotation of different images,an improved sift-brisk feature extraction algorithm is proposed.By constructing the dog pyramid,the algorithm can find feature points in multiscale and generate binary descriptors in the form of constructing tangent circle.Compared with the traditional feature extraction algorithm,this algorithm has strong robustness under the interference of similar scene,rotation,scale and so on.2)In order to solve the problem of large computation and high time complexity,a group matching algorithm is proposed.The algorithm starts from the similarity of feature descriptors,and finds the most similar group of feature descriptors in the form of grouping.After experimental simulation,under the condition of a picture,the matching time of the algorithm is twice as long as that of the traditional violence matching algorithm.3)This paper analyzes the correlation tracking method and optical flow method in the specific target tracking algorithm,and from the KLT corner tracking algorithm in the optical flow method,presents the process of using the improved sift-brisk feature extraction algorithm to complete the specific target tracking.4)The algorithm of sparse 3D point cloud reconstruction is analyzed.The process of coordinate transformation and relevant parameters in camera calibration is given.The simulation of sparse 3D point cloud reconstruction is completed by experiment.5)In order to solve the problem that the threshold information of dense growth is less in dense point cloud reconstruction algorithm,an improved region growth algorithm is proposed.The algorithm makes 3D dense reconstruction more precise by giving multidimensional information of growth threshold.After experimental simulation,the improved region growing algorithm has a better reconstruction effect in the reconstruction of 3D dense point cloud.