Our gait-recognition method recovers static body and stride parameters of subjects as they walk. Our technique does not directly analyze the dynamic gait patterns, but uses the action of walking to extract relative body parameters. An ad hoc cross-condition mapping method, which allows for the identification of a walking subject viewed under conditions that are different than those at which their initial data were recorded, is used to map between viewing conditions.
The first set of static body parameters our technique measures, as a person walks, are four distances: the vertical distance between the head and foot (d1), the distance between the head and pelvis (d2), the distance between the foot and pelvis (d3), and the distance between the left foot and right foot (d4). The second set of parameters, a subset of the first and less discriminating, but less sensitive to error introduced by variation in viewing conditions, are d1 and d3. These distances are measured only at the maximal separation point of the feet during the double-support phase of the gait cycle and are concatenated to form a four-dimensional walk vector w = [d1, d2, d3, d4] and a two-dimensional walk vector s = [d1, d3] for each subject.