论文部分内容阅读
Scale variation significantly affects the result during human tracking, especially in crowded scenes.This paper proposes a novel scale adaptive algorithm which is mainly based on the original Kanada-Lucas-Tomasi(KLT)-based features.We apply KLT algorithm for feature detection in target to get the size information.This original feature is not sensitive to environment so it is suitable to solve the scale change problem in crowded senses.Depending on the stable scale feature, our method can obviously improve tracking performance and help handle drift and occlusion issues.Experimental results on famous datasets prove that our method has a good accuracy and robustness on scale change, occlusion and drift.