Recognition-Based Motion Capture and the HumanEva II Test Data

Quantitative comparison of algorithms for human mo- tion capture have been hindered by the lack of standard benchmarks. The development of the HumanEva I & II test sets provides an opportunity to assess the state of the art by evaluating existing methods on the new standardized test videos. This paper presents a comprehensive evaluation of a monocular recognition-based pose recovery algorithm on the HumanEva II clips. The results show that the method achieves a mean relative error of around 10-12 cm per joint.