Long range Gait recognition in range-gated viewing video by moment based descriptors

In complicated environment of low visibility, long range Gait receives increasing attention as a biometric in surveillance. Given the requirement to process sequences obtained by range-gated imaging system, computational demands are high, motivating the development of basic approaches. Based on these baseline approaches, we use a family of moment descriptors to describe a single shape by using a masking circle, and analyze gait with specificity and speed using moments to acquire invariant properties. Results show that the Hu1 moment performs consistently better than other moments at discriminating between subjects. And by constraining the database to specific starting foot and direction of walk we are able to dramatically increase the recognition rate over 80%.