Research On Anti-occlusion Algorithm Of Tracking Object Based On ELM And Mean Shift

Due to the rapid development of science and technology, especially the computer technology, machine vision has become one of the most challenging hot topics in the current scientific research. Moving object tracking is an important branch of machine vision, and it is widely used in many fields, such as human computer interaction, national defense, military, intelligent transportation and life security. Target tracking, which uses image recognition technology to automatically identify and locate the target from the image according to the characteristic of the target, which is distinguished from the background. In the actual target tracking process, due to the diversity of the target feature, the uncertainty of the speed of motion and the complexity of the tracking background and other factors, often can not accurately track the target, leading to the failure of tracking. Especially target tracking scenes are becoming more and more complicate, tracking technology of the simple background has been unable to meet the practical needs, which has constrained promotion of the application and popularity of technology tracking greatly. Therefore, one of the main difficulties of target tracking is how to improve the stability and real-time performance of the target tracking system.By theory, we can improve the performance of tracking system from software and hardware, such as the processor with higher performance, or the target tracking algorithm.In fact, because of the bottleneck of core technology and the problem of cost, improve tracking performance of the system through hardware innovation faced many restrictions, it's difficult to market and apply it in many areas in a short period of time. Therefore, improve stability and real-time of the target tracking through the improvement and optimization of the software algorithm has an important significance.In the actual process of target tracking, the limit of practical of algorithm we encountered is the shelter. With the change of occluded part, characteristic changes in different degree, which lead to inconvenience to the target tracking. In addition, when the target speed is too high, tracking deviation may occur because of the complexity of the algorithm.In this paper, we mainly study the anti occlusion problem in the target tracking algorithm. First, introduce the tracking principle and tracking method of traditional algorithm, analysis the advantages and defects of Mean Shift target tracking algorithm in depth, draw forth the improved Mean Shift algorithm, and make the simulation and analyze the result.The method of location prediction using neural network is proposed on based on the study of the existing target tracking algorithm. This paper chooses the ELM which have simple structure and powerful function to make position prediction, a new Mean Shift tracking algorithm base on ELM is proposed.The algorithm according to the target position information of the past three moments, using ELM predicts the possible position instead of point position as the mean shift iteration starting point, and iterate in the neighborhood, and get target true position at last. The experiment result show that compared with the existing algorithms, the algorithm proposed in this paper reduces computation time, and can locate the target location in accurately in the shelter conditions, it also improve the real-time and stability of tracking system.