Developing a Tracker for Crowded and Semi-crowded Scenes

Supervised by Prof.Mohammadzade, Deep Learning Lab, Sharif University of Technology

First, fine tuning YOLO for CCTV videos in order to have a more accurate detection of peoplepresence in videos.

Second, developing the tracker’s accuracy and speed using aspect-ratioand box-size joint distribution. A real-time accurate multi-object tracking system that operates with minimal computational overhead. The proposed tracking-by-detection method follows the online real-time tracking approach established in prior literature. Introducing a novel cost function called the Bounding Box Similarity Index, our job reduces the dependency on the Kalman Filter, resulting in decreased computational requirements

Third, using the developed tracker for video synopsis.

The results will soon be published as a paper(Expected in early January)