Camera Placement Considering Occlusion for 
Robust Motion Capture

Xing Chen, James Davis

Abstract

In multi-camera tracking systems, camera placement can have a significant impact on the overall performance. In feature-based motion capture systems, degradation can come from two major sources, low image resolution and target occlusion. In order to achieve better tracking and  automate the camera placement process, a quantitative metric to evaluate the quality of multi-camera configurations is needed. We propose a quality metric that estimates the error caused by both image resolution and occlusion.. It includes a probabilistic occlusion model that reflects the dynamic self-occlusion of the target. Using this metric, we show the impact of occlusion on optimal camera pose by analyzing several camera configurations. Finally, we show camera placement examples that demonstrate how this metric can be applied toward the automatic design of more accurate and robust tracking systems.

Selected Figures

Left: Two cameras constrained to move on an outer sphere try to cover a inner spherical target space; the angle with which they intersect the sphere center is q. The occlusion and resolution   metrics are minimized at different angles. Center: 3D uncertainty vs. q considering resolution only . Right: 3D uncertainty vs. q considering occlusion only. (q is from 0 to 180 degrees.)

  3D uncertainty  vs. number of cameras. Error due to imager resolution is nearly minimized by only two cameras. Many more cameras are needed to robust insure against occlusion.