Research Of Object Tracking Method Based On ORB Features And Particle Filter

From ancient times to the modern, security personnel who is actually the mostimportant fact in scene monitoring. But man is a destabilizing factor, people willoversight something. So this is a major flaw of traditional surveillance. As the timesprogress, intelligent monitoring has become the essential aspect in the monitoringfield. It owed to progress of technology and camera. One camera equipment whichcan warn people if there is some dangerous thing happened automatically. That, notonly can solve human waste, but also for social stability is to play an important role.However, the target tracking, as the basic of intelligent monitoring, which directlydetermines the quality of the development trend of the future of intelligent monitoring.In the monitoring, to the interested target, for accurate, real-time tracking, follow-upanalysis and identification of target behavior plays a decisive role. And target trackingis an iterative process, the lastest one results direct impact on the accuracy of thefollow-up results. Therefore, improving the accuracy of target tracking algorithm, isthe main goal of this study. To date, in the domestic and overseas,a lot of scholarshave proposed many first-class algorithm about target tracking algorithm. They wereused in different ares.The algorithm is roughly divided into three categories, namely:point tracking, nuclear track and silhouettes tracing. Point tracking’s main idea is:first,it try to find the target centroid location, in the each frame. Then, this algorithmcomputer the distance between the continuous frame of the case and centroid toaccomplish the tracking. Nuclear track, with the core to represent the appearance andshape of the target, while the nucleus is usually oval or rectangular box, to determinethe location of nucleus and extract the target feature, and finally through thedisplacement to calculate the nuclear target tracking. In this method——Silhouettetracking,the target is represented by the model, not the simple geometry, throughobject region.Such as: the edge of target, the contour. Then, predict the target area inthe each frame to accomplish the target tracking. And now,a lot of the popular targettracking methods, they used the global features to represent the object model. Simpleis the best advantage to use the global features. While the weakness of that is aboutocclusion. It cann’t solve it well. The information of the target is not enough torepresent the object, due to the concusion. And it will due to lose the target.In this paper, we mainly use the second type of target tracking methods, while using local features to represent the interested target. Because, when deal with theocclusion, the local feature has advantage over the global features. So we chose thefar out ORB features, to represent the trait of the target. Then we will achieve targettracking tasks in the framework of the particle.The Particle Filter is one of the most popular algorithm, in tracking field.Because it is nonlinear and non-Gaussian.The unite of ORB and the PF is used fortarget tracking, not only can solve the occlusion problems, but also has a wonderfulperformance in the aspects of real-time. Due to the ORB features with rotationalinvariance, and faster. Of course, under different scenarios, the demand for targettracking tasks is not the same. In the case of occlusion in serious cases, we can usemulti-track approaches to solving the problems. Firstly, we can determines the centerof the target sub-module. We must to find another algorithm to resolve the problem ofthe all keep out. It is to track the object, through the predict. We will predict theposition of the target, use the previous frame. If there is a part of the block which wecan achieve the target track by the above methods. In order to achieve the goal oftracking tasks better, we can also embed semi-automatic target tracking, people wasinvolved in tracking the target. Human-computer interaction will improve theaccuracy of tracking task greatly.And avoid the human’s shortcomings effectively.Thus it will improve the overall effect of target tracking.