Motion Compensated Background Filtering and Tracking

The main focus of my work is in detecting and tracking moving objects in sequences where the camera is moving. This means the background is non-static and as such a standard background filtering algorithm will not suffice. One solution would be the use of a model based system but this would require prior knowledge of the objects that are to be tracked in order to separate them from the background.

In order to overcome this problem a system of motion compensated background filtering is used. The principle is to maintain a statistical model of the background by modeling the average difference between the current frame and the previous frame which is warped to overcome the global motion

In order for this approach to be successful the background model needs to be robust to:

• Deficiencies in the motion model used
• Innacurate estimation of the global motion
• Variations in the scene conditions

• Background Motion Estimation
• Background Filtering
• Object Mask Filtering and Object Detection
• Object Tracking

Personal | Links | Group | School | University Last updated by Michael Knowles on 20th October 2004 Maintained by Michael Knowles