Analysis and evaluation of human tracking methods from video

Abstract

Human tracking has always been an interest, due to its importance in many areas, from security, surveillance, defence, as well as for mobile robotics applications like human following. Recently, a large number of tracking algorithms have been proposed, and a lot of them have been compared for extensive evaluation, however there are still limited to certain methods or situations only. Here, analysis towards tracking algorithms - Kalman Filter (KF), Mean-Shift Filter (MSF) and Partial Least Square (PLS) - are done to evaluate their performances. Two situations of human walking and running in outdoor surrounding were chosen, and the computational time over tracking in several parameters of colour effect, human orientation and movement were analysed and evaluated. The effect of these parameters were also analysed while the selected algorithms perform the tracking. Based on the analysis, it is shown that MSF and KF perform well in tracking human in the designated conditions. Apart from that, it can also be concluded that the selected parameters do give impacts towards the performance of these algorithms in tracking human, which can act as a guide in developing a more robust algorithm in the future.