Multi-target Algorithm And Point-tracking Algorithm Based On Vision

Target tracking is an influential computer vision task.Because of its academic potential and practical application value,it has attracted the attention of academic and commercial aspects,and has extended many related subtasks.In general,the challenges faced by this task are similar: such as lighting,brightness changes,and occlusion issues.Different subtasks also have differences,such as the correlation problem of multi-target tracking tasks,and the pixel-level precision tracking problem of point tracking.This article proposes a series of methods to effectively solve the problems faced by two sub-tasks in the field of target tracking: multi-target tracking and point tracking.The main work of this paper is as follows:(1)Aiming at the problem of noise detection and occlusion in multi-target tracking,this paper proposes a long and short-term memory model to increase the time information in the tracking process.In the update of the memory model,in addition to calculating the similarity between each target and the template,a memory model that simulates the human memory curve is also used to update the template.At the same time,a new scoring function is proposed to select better candidate detection results and existing trajectories.Spatio-temporal information confidence is an effective processing method,which can effectively reduce missed detection,false detection and occlusion.On the MOT16 data set,this paper verifies the feasibility and effectiveness of the method through experiments.(2)Aiming at the problem that the traditional method is not accurate when tracking specific points,the tracking will be offset,and it cannot meet the requirements of adding dynamic icons to the video.This paper proposes a method based on deep neural network to extract key points.,Combining feature matching to obtain the homography matrix,and then transforming,effectively solve the problem of tracking specific points.Through experiments on the "Malanshan" audio and video data set,and comparison with various methods,the effectiveness of the method in this paper is verified.The work in this paper has a good effect on solving the noise detection and occlusion problems in the multi-target tracking task.The proposed point tracking method also has greater advantages than other methods.