Robust tracking based on particle filter supported by SVR

In this paper, we propose the use of the particle filter supported by support vector regression (SVR) in order to track people under constraints and unstructured environment such as light variation and shadow. These hard constraints provide errors in the tracking. For this, we propose a robust algorithm for tracking based on particle filter which seems to be useful since it has the ability of tracking with robustness. However, this algorithm needs to calculate the probability density function (PDF) on which we propose to use the SVR to robustly estimate this density of probability. So as to show the performance of this new approach, we have tested the proposed method in our own dataset comprising different scenarios such as walking and running actions in hard constraints. The obtained results allowed us to validate the performance and the robustness of the proposed framework based on tracking by particle filter supported by support vector regression (SVR).