Model-Based Gait Enrolment in Real-World Imagery

We present a model-based approach to gait extraction that is capable of reliable operation on real-world imagery. Hierarchies of shape and motion are employed to yield relatively modest computational demands, avoiding the high-dimensional search spaces associated with complex models. Anatomical data is used to generate shape models consistent with normal human body proportions. Mean gait data is used to create prototype gait motion models, which are adapted to fit individual subjects. Accuracy is evaluated on subjects filmed from a fronto-parallel view in controlled laboratory conditions, for which some gait parameters are known. We further show that comparable performance is attained in outdoor conditions. As such, we describe a new approach to enrolment for gait recognition technologies, allowing reliable subject gait extraction in real-world imagery.