Hybrid Algorithm for Multi People Counting and Tracking for Smart Surveillance
Mahwish Pervaiz, Ahmad Jalal, Kibum Kim
Abstract
Reliable people counting and tracking is active research topic in visual surveillance. In this work, a novel approach has been proposed for estimating people and tracking their location in sequence of video frames. Initially, we used Gaussian filter and background removal techniques to preprocess the image. After preprocessing, skin verification and body point detection have been introduced for human verification. For people counting, centroid of silhouettes and jacquard similarity index are developed to track moving objects in video frames. Experimental results on Pets 2009 dataset demonstrate that proposed system give boost of 8% accuracy in terms of tracking accuracy and counting rate as compared to known state-of-the-art methods. This system should be applicable to count and track people in medium density crowd environment.