Abstract

In this paper a Computer vision based Tracking and Counting System is proposed which can deal
with object occlusion effectively. First an Adaptive Background Estimator which uses pixel
averaging technique is employed to obtain a background image and is updated periodically to
reduce the effect of illumination and environmental changes. Foreground extractor with a
combination of Background Subtraction and Frame differencing is used to obtain a foreground
mask. Then using a blob analyzer, object properties like area, centroid etc. is found. A tracker is
employed to track the characters using its centroid and size. Counting is done based on the position
of centroid and object tagging. Occlusion effect can be minimized by considering the area of an
object to detect merged states. The main Objective is to obtain a low computational system which
can be possibly used in real time tracking.