|
Object
Tracking Using an Enhanced Adaptive Background CAMSHIFT Algorithm
This is my master's
project with Professor Mircea Nicolescu at Computer Vision Lab, in Department of Computer Science &
Engineering, University of Nevada at Reno.
Abstract
After
decades of rapid development, visual
tracking has become an important issue in computer vision. For applications such as robot vision and wide area surveillance, robust and flexible tracking algorithms are highly desirable with a
requirement of minimal training and
computational resources, as well as adaptation to moving cameras. As a fast and simple
method of tracking, the "Continuously Adaptive Mean Shift" (CAMSHIFT) algorithm developed
in Intel OpenCV library was designed to handle
precise tracking of facial location on a non-moving camera. However, this
method often fails in the
presence of camera motion. The goal of
the project is to extend the CAMSHIFT face-tracking algorithm in order to
allow a non-stationary camera and real video sequences containing non-rigid
objects.
The proposed algorithm, the Enhanced
Adaptive Background CAMSHIFT (EABCshift),
addresses those difficulties by including a flexible background representation which
can be continuously re-learned with minimal additional computational expense. The system also extends the original CAMSHIFT
by incorporating hue and saturation information instead of just the hue
value of the object, as well as a weighted color histogram. The data has
been tested against the CAMSHIFT algorithm, and the performance of the proposed
algorithm is shown to be better than the original one in a wide range of
difficult scenarios.
|