Close to real-time reliable feature tracking of moving vehicles in a highway.(Link)
Note: For better visualisation, the tracks appear after the vehicle has moved about 10 pixels so that shorter tracks are filtered out.
As part of automated surveillance of crowded areas, the problem of counting the number of people is complicated by various factors - articulated motion of people changes their appearance, highly crowded environment occludes people, perspective effects changes the scale of people, and low resolution of CCTV videos limits the information content. We extend motion-based algorithms and evaluate appearance-based algorithms for counting the number of people in the context of a highly crowded marketplace. The work was documented as my masters thesis.