Abstract
It has been observed that the gait motion of the human body is stable across individuals for the lower body but extremely variable for the upper body. This regularity of the lower body has made it attractive for biometric approaches to identity recognition that can capitalize on systematic deviations to a regular structure. The motion of the upper body on the other hand has been termed attitudinal as it is not necessary for locomotion but seems to vary with the manner of the walker. We wished to contrast these two sources of information in the recognition of identity from point light walkers. To do this we started with a library of 29 point light walkers and created three types of displays: full body, upper body and lower body displays. Each point-light display consisted of a side view of a single gait cycle. Next, for each display condition we provided observers with a sequential presentation of two walkers for each of the possible pairwise combinations of the 29 individuals and had them give a dissimilarity rating for the pair. We thus obtained an average dissimilarity matrix for each of the three display conditions. These matrices were analyzed using techniques of hierarchical data clustering that provided us with an indication of which walkers were grouped together and which individual walkers were prototypical for each cluster. These results showed that all three display conditions shared some prototypes, but that the prototypes for the whole-body and upper-body clusters were nearly identical. These results suggest that the motion of the upper-body predominates in the recognition of identity by human observers. However, theoretical analyses are ongoing to determine to what extent human observers' recognition of identity from the lower body can be accounted for by biometric theories of identity recognition.
EPSRCAttitudinal and Biometric Contributions to the Recognition of Identity from Point-light Walkers