Introduction

Problem

Approach

Results

Discussion

Code

Discussion

The results show that the tracker works but is not perfect. Some situations cause it to lose the subject's head and then failing to recover from it. It seems that the color module is not pulling the ellipse to the head as strongly as it should. The gradient module also seems to get easily distracted with the background.

Some recommendations to improve the results of the tracker are using a Gaussian filter on the image before calculating the gradients. A dynamic window size can be implemented to help deal with fast head movement. The optimal value can be computed from the head's velocity using previous frames.

The goals were attained but not with 100 percent success. Implementing the recommendations will make the tracker more robust and suitable for its use in the Human Activities Recognition project.

References

1. Stan Birchfield. Elliptical Head Tracking Using Intensity Gradients and Color Histograms. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 232-237, June 1998.
2. Anant Madabhushi and J. K. Aggarwal. A Bayesian Approach to Human Activity Recognition. IEEE Workshop on Visual Surveillance Systems, CVPR, 1999.
3. Eamonn Keogh. Exact Indexing of Dynamic Time Warping. 28^th International Conference on Very Large Data Bases, Hong Kong, pages 406-417, 2002.