140 likes | 264 Views
This project aims to track objects in videos using a simple and efficient silhouette tracking technique. The input is a video, and the output is a marked video with the tracked object outlined in red. Color histograms are used for stable object representation, especially in varying illumination conditions. The approach involves object representation, initial contour placement, color histogram generation, noise removal, contour evolution, and stochastic search for best contour selection. Kalman filtering is applied for prediction error correction. Results show fast processing with good object tracking even with skipped frames or partial occlusion.
E N D
Object Tracking Group M Sujith Thomas (Y7111037) Surya Prakash (Y7111063)
Problem Definition • To track an object using Silhouette Tracking. • Input will be a video • Output will again be a video with the tracked object being marked by a red contour.
Motivation • Developing a simple and fast technique to track an object across frames in the video. • Color histogram could be a stable object representation that works well for varying illumination conditions.
Literature Referred • Alper Yilmaz, Omar Javed, M. Shah, “Object tracking: A survey,” ACM Computing Surveys (CSUR) , 38(4),2006. • Michael J. Swain,Dana H. Ballard,“Color Indexing”, International Journal of Computer Vision, 1991. • David Forsyth and Jean Ponce, Computer Vision: A modern Approach, Prentice Hall India 2004
Our Approach • Object Representation • Initial Contour in first frame • Color Histogram generation • Noise removal from the color Histogram
Our Approach Contd.. • Contour Evolution • State of the contour defined by the position of its centroid, velocity and acceleration • Next state is predicted based on the current state • [Stochastic Search] Best contour selected by searching the neighborhood stochastically based on Sigma value of the errors. • [Kalman Filtering] After the new position is found, error in prediction is obtained and the modeling parameters are updated.
Result-3 Ref: The VIVID Tracking Evaluation Web Site, CMU http://www.vividevaluation.ri.cmu.edu/main.html
Time Taken in Processing • For Result-3 • Processing time = 0.23 Seconds/ Frame (Excluding disk input-output time)
Conclusions • Representing objects using a color histogram gives good results when we require fast tracking of objects. This is because it is computationally inexpensive. • Method presented can track the object even when a few frames are skipped. • This approach can take care of partial occlusion of object. • Approach fares well only when object color is atleast slightly different from that of the background.