Effect of arm positions on the curve spreads: Implications on the database and new application
Abstract
We investigate the effect of restrained arm positions on the recognition rate of the Curve Spreads of persons walking toward the camera. The Curve Spread, which is a matrix of changes of body curvatures, with the rows as time axis and the columns as the spatial axis, is a technique for human gait recognition when correlated and ranked with other Curve Spreads. Using the Curve Spreads of persons, each walking with arms swinging freely on the sides and with arms in other positions (folded in front, both and either side in akimbo), we measured the recognition rate using two kinds of database: one which consist of Curves Spreads of all the aforementioned variations in arm positions and another which consist of Curve Spreads from the gait with arms swinging freely on the sides (baseline gait) only. 98% and 86% first and second rank recognition, respectively, were obtained for the first database and 4% for both first and second rank recognition for the second database. These results posed new questions on the structure and role of the database in gait recognition. Detection of distinct foot dominance for every arm position has implications on other gait researches. The Curve Spreads can also function as a diagram showing the spatial and temporal factors of gait cycles making it possibly applicable to clinical gait analysis.