Pedestrian tracking in video sequences: A particle filtering approach

In this work we study the methods for pedestrian tracking in video sequences and indicate various applications of these methods ranging from surveillance systems to aiding the visually impaired persons. First, we define the general problem of object tracking that comprises the tasks of object detection, identifying the flow of object location in consecutive video images and finally analysis of the tracked trajectory data. We review the well known object tracking techniques i.e. the Mean-Shift and the CAMSHIFT algorithm and discuss their properties. Then we introduce the computational technique known as particle filtering (PF) and explain how we have applied it to the tasks of pedestrian tracking. We compare the PF approach against the Mean-Shift and the CAMSHIFT algorithms in terms of tracking robustness and the required computational demand. We conclude, that on the tested video sequences, the PF tracker outperforms the Mean-Shift and by a small margin the CAMSHIFT algorithm. The PF tracker requires more computational power, however, its tracking performance can be flexibly adjusted to the application requirements.