Real-Time Tracking Demo

Real-Time Object Tracking from a Moving Video Camera

A software approach on a PC

Yoav Rosenberg and Michael Werman

Real-time tracking of moving objects from a moving camera has the following stages:

Stage 1: Computing the global motion. Affine motion can be computed in real-time.
Stage 2: Finding points whose motion is inconsistent with the global motion.
No image warping is necessary, adding to the computational efficiency.
Stage 3: Clustering inconsistent points into moving objects.

Initial article on this work is Representing Local Motion as a Probability Distribution Matrix and Object Tracking, Y. Rosenberg and M. Werman, DARPA IUW'97.

Following are Mpeg sequences that demonstrate the results. It should be noted that everything is performed in real time on a PC using only software!

Example 1: Computing Global Motion is demonstrated by stabilization. Note that Stabilization is performed only to demonstrate this stage, and is not necessary for tracking. Click on an Image to view the MPEG Video

Original Video Stabilized Video

Example 2: Tracking moving objects from a moving camera.
Click on an Image to view the MPEG Video

Marking Tracked Objects

Back To Hebrew University Computer Vision Lab