Integrated View Calibration and Tracking
The goal of this project is to take video from a moving camera and to do view calibration - to compute camera motion and create an environment model - in parallel with doing tracking. The benefits are two-fold. Firstly for application use, it is possible to position tracking results within the context of the actual environment model. Secondly in terms of algorithms, the computed camera motion is expected to support and extend tracking capabilities.
Background & Objective: MERL has advanced tracking technologies but currently has no way to automatically capture an environment model such that the motion of the tracked objects can be shown within the context of the environment. When an environment model is used, it is typically a rudimentary manually-created model, maybe just the ground plane, and is used only with a fixed camera. The objective of this work is to show that view calibration in conjunction with tracking can significantly extend system robustness and can enable new applications.
Technical Discussion: Initial work will treat view calibration and tracking as independent system components. View calibration will involve detection of point and line features from video, matching between frames, computation of camera motion, and computation of 3D features for constructing an environment model. Tracking will be performed on the same images, and a subsequent stage of processing will position the tracking results within the environment model. Initial application will be sports-field analysis, but pedestrian detection and determining pedestrian motion (towards the kerb, parallel to the kerb etc) is also a significant application of interest.
Future Direction: The longer-term goal is to further integrate the view calibration and tracking components, and to demonstrate that they are mutually supporting.
Contacts:
Paul Beardsley
Fatih Porikli
Technology Area: Imaging
Modification Date: March 21, 2008
