Robust Feature Tracking

T. Tommasini, A. Fusiello,and V. Roberto	E. Trucco
Dipartimento di Matematica e Informatica Università di Udine - Italy	Department of Computing and Electrical Engineering Heriot-Watt University - United Kingdom

Overview

This work addresses robust feature tracking . We extend the well-known Shi-Tomasi-Kanade tracker by introducing an automatic scheme for rejecting spurious features. We employ a simple and efficient outlier rejection rule, called X84. Experiments with real and synthetic images confirm that our algorithm makes good features to track better.

Method

Extraction.

Good features are located in the first frame of sequence by examining the minimum eigenvalue of each 2 by 2 autocorrelation matrix.

Registration.

Features are tracked by minimizing the difference between the two corrispondig windows in two consecutive frames, using a Newton-Raphson method.

Robust Rejection.

The residuals between the first and the current frame is taken as the measure of the feature's quality. We use an automatic scheme for rejecting spurious features, employing a simple, efficient, model-free outlier rejection rule, called X84.

Results

sequence (MPEG, I-frames only)	size	parametrs	source
Artichoke	480x512x99	rtrack -monitor	CMU-VASC image database
Hotel	480x512x59	rtrack -monitor	CMU-VASC image database
Stairs	512x768x60	rtrack -monitor	courtesy of F. Isgro, CEE - Heriot-Watt University (UK)
Platform	256x256x20	rtrack -monitor -w 9 -t 0.3	SOFA test sequences
Hyball	170x256x31	rtrack -monitor -w 9	courtesy of C. Plakas, OSL - Heriot-Watt University (UK)
Scale	255x384x21	rtrack -monitor -w 13	courtesy of C. Plakas, OSL - Heriot-Watt University (UK)
Smallrock	170x256x25	rtrack -monitor -w 9	courtesy of Chris Smith IMBC, Crete (GR)
Box	240x320x25	rtrack -monitor -w 9 -t 0.08	courtesy of A. Benedetti, Vision Group, CALTECH (CA)

Reference paper

T.Tommasini, A.Fusiello, E.Trucco and V.Roberto, Making Good features Track Better, in "1998 IEEE Conference on Computer Vision and Pattern Recognition" , S. Barbara (CA).

Code available

An implementation in C of the Robust Tracker is available from our web site.