Identification of people from gait has become a challenge problem in computer vision. However, the conditions under which the problem is ``solvable'' are not understood or characterized. In this context, we introduce the Human ID Gait Challenge Problem, which consists of a large data set (about 1.2 TB of data related to 1870 sequences from 122 subjects spanning 5 covariates), a set of experiments of increasing difficulty, a baseline algorithm, and its performance on the challenge problem. The baseline algorithm involves silhouette estimation by background subtraction and similarity computation by temporal correlation of the silhouettes. The instructions to get the gait data, the source code of the baseline algorithm, and scripts used to run the challenge experiment are available at this site. This infrastructure should facilitate the understanding of the strengths and weaknesses of different gait recognition algorithms by allowing identical experiments across identical data sets. The scientific motivation behind the design of the challenge problem is that, as a research community, over time, we can address the following questions:
Last revised: January 15, 2004