Accuracy enhancement and false acceptance reduction in multiple pedestrian detection and tracking

This paper discusses the implementation of multiple pedestrian tracking in a pedestrian detection framework. Pedestrian detection and tracking is used in modern day ADAS (Advanced Driver Assistance Systems) for detecting the possibility of collision with a pedestrian by capturing video stream. The ADAS are responsible for generating warning to driver or automatically controlling the vehicle. The multiple pedestrian tracking method we propose uses a weighted average of the velocities of each pedestrian to predict it in the next frame. The method we propose has the ability to handle entry, exit and occlusion cases, which are bound to occur when there are multiple pedestrians moving in and out of the field of view of the camera. This method also uses a Haar-like feature based matching and sampling to enhance the result of tracking.