Building a Multiple Object Tracking System with Occlusion Handling in Surveillance Videos

Building a Multiple Object Tracking System with Occlusion Handling in Surveillance Videos

Raed Almomani (Wayne State University, USA) and Ming Dong (Wayne State University, USA)
Copyright: © 2013 |Pages: 13
DOI: 10.4018/978-1-4666-3994-2.ch053
OnDemand PDF Download:
$37.50

Abstract

Video tracking systems are increasingly used day in and day out in various applications such as surveillance, security, monitoring, and robotic vision. In this chapter, the authors propose a novel multiple objects tracking system in video sequences that deals with occlusion issues. The proposed system is composed of two components: An improved KLT tracker, and a Kalman filter. The improved KLT tracker uses the basic KLT tracker and an appearance model to track objects from one frame to another and deal with partial occlusion. In partial occlusion, the appearance model (e.g., a RGB color histogram) is used to determine an object’s KLT features, and the authors use these features for accurate and robust tracking. In full occlusion, a Kalman filter is used to predict the object’s new location and connect the trajectory parts. The system is evaluated on different videos and compared with a common tracking system.
Chapter Preview
Top

Background

Generally, Object tracking can be divided into three major categories (Yilmaz, Li, & Shah, 2004): Correspondence-based object tracking, transformation-based object tracking and contour-based object tracking.

Correspondence-Based Object Tracking

Tracking is performed by collecting the object information during tracking and using this information to predict and verify the object new location. The researchers used different object previous information such as object state (velocity and acceleration) and regional information (color, texture, area and shape) (McKenna, Raja, & Gong, 1999; Zhang, & Freedman, 2005). Different filtering techniques are suggested to model object information such as Kalman filtering (Stauffer, & Grimson, 2000) and particle filtering (Rittscher, Kato, Joga, & Blake, 2000).

Transformation-Based Object Tracking

Tracking is performed by using the object information in consecutive frames to estimate the motion of the object. The most common transformation based trackers are ''template matching'' (Lipton, Fujiyoshi, & Patil, 1998), ''Mean shift'' (Comaniciu, Ramesh, & Meer, 2003) and KLT (Kanade-Lucas-Tomasi) (Shi, & Tomasi, 1994). Template matching tracker is implemented by searching the whole image for a similar template. The template matching tracker is challenged by the heavy computation required and the sensitivity to illumination variation. Mean shift, on the other hand, tracks the distribution of an object in real time with robust tracking performance. However, Mean shift tracker does not deal with deformable objects and the appearance variations decreases the tracker performance (Allili, & Ziou, 2008). KLT tracker finds features that are optimal for tracking first, and then computes the translation of these features and the quality of each tracked path (Zhou, Yuan, & Shi, 2008).

Complete Chapter List

Search this Book:
Reset