Model-based motion capture for crash test video analysis

Abstract

In this work, we propose a model-based approach for estimating the 3D position and orientation of a dummy's head for crash test video analysis. Instead of relying on photogrammetric markers which provide only sparse 3D measurements, features present in the texture of the object's surface are used for tracking. In order to handle also small and partially occluded objects, the concepts of region-based and patch-based matching are combined for pose estimation. For a qualitative and quantitative evaluation, the proposed method is applied to two multi-view crash test videos captured by high-speed cameras. © 2008 Springer-Verlag Berlin Heidelberg.