Automatic counterflow detection in subway corridors by image processing

The purpose of our study is to provide an automatic tool for the detection of abnormal individual or crowd motion in subway corridors, in particular counterflows in one way corridors. First, we developed a motion estimation method that takes into account two difficulties: real time constraint and non-rigid moving objects (the pedestrians). We chose three techniques, for this purpose. Each belongs to a specific class of methods. These are block matching (matching technique), optical flow (differential technique) and Gabor filtering (frequential technique). We then attempted to improve both the speed and performance of each algorithm. As a result, we produced a very efficient optical flow technique. The obtained motion vectors were then filtered and used for the construction of motion trajectories. The trajectories were finally used to detect counterflows.