The Video Pedestrian Tracking Algorithm Based On TLD Frame

The pedestrian tracking based on videos is the foundation and key of the machine vision,has deep basic effect on many technical applications,with important research significance and application value.Nowadays,this task has been in-depth study,and made a lot of research results.Current pedestrian tracking algorithms have always shown their respective advantages in specific environments,while we even have no algorithm focus on the long term pedestrian tracking.Since,we put a new pedestrian tracking algorithm forward,based on the frame of TLD algorithm,with the base of fully understanding about the TLD algorithm frame.The specific research work done includes five aspects as follows:(1)Research about pedestrian tracking basis theoretical.Carefully studied the principle of object tracking,special pay attention to the study on characteristics of pedestrian tracking,and carefully studied and compared classic pedestrian tracking algorithms.Read a lot of literatures about pedestrian tracking method,classify them and find the points of requirement that different kinds of methods focused on,their own advantages and disadvantages.(2)Research about the TLD algorithm.Understood the operation flow of TLD,reappeared the algorithm tracking process by experiments and verified its tracking results.Deeply studied the TLD algorithm frame,principle of actions and theoretical basis.At the same time,analyzed the process and details of the implement,so that we can thorough understand hole algorithm.(3)Applied the TLD algorithm frame to the pedestrian tracking.Collected problems during the algorithm application,adjust related parameters according to the pedestrian tracking characteristic.At the same time,analyzed the reason why those problems occurred after TLD algorithm study,and determined the direction to solve and basic idea.(4)Improving the TLD algorithm.For Satisfying pedestrian tracking demands,analyzed and evaluated every modules in the algorithm,and took the HOG features and SVM classifier into our algorithm,so the detection module got stronger to detect pedestrians and make better effects.Finally,we got a new algorithm more suitable to pedestrian tracking.(5)Experimental verification.The software experiment platform VS2010 and Matlab R2014 a and auxiliary software are used to verify the algorithm,and the comparison and quantification analysis of the experimental results are carried out to prove the feasibility and effectiveness of the algorithm.The innovation of the research is that :(1)The “Detection-Tracking-Learning” algorithm frame has been took into pedestrian tracking field and it provides a new idea to solve pedestrian tracking problems.(2)At the same time,we have improved original algorithm targeted,with the principle of keeping the main algorithm frame complete,to improve the pedestrian tracking consequent.The shortcomings of the study are that the algorithm complexity makes the operation speed slowly and not ideal,so we need continue to improve the algorithm frame,optimize the calculation especially the process about lots of candidate rectangles,reduce the time complexity for faster process.On this side,we need more study and experiments.