Vote mapping based improved human tracking for intelligent surveillance systems

摘要

Human tracking is a challenging task and a significant part of the design of an intelligent surveillance system. Though the existing tracking techniques accomplished the reasonable outcomes in terms of accuracy and robustness, there is a scope for improving the tracking performance. In this paper, vote mapping of patched confidence methodology is used with the consecutively increasing number of patches. The system aims to provide robustness to occlusion and global scene changes by utilising the number of patches from the bounding box of an image. An individual patch is tracked by kernelised correlation filter and applied to the vote mapping methodology. The consecutively increasing number of patch approach and vote mapping provide robustness to the occlusion in real time tracking scenarios. The qualitative and quantitative analysis reveals the superiority of the proposed vote mapping-based tracker over the existing kernel-based trackers.