Gait recognition using low spatial and temporal resolution videos

Due to limited transmission bandwidth and storage capacity, the gait sequences captured by a Closed Circuit Television (CCTV) camera are recorded at low frame-rates with low spatial resolution, which significantly affect the performance of a gait recognition method. Thus, this paper proposes a gait recognition method which uses dual-tree complex wavelet transform for spatial resolution enhancement of the input low resolution silhouettes. To achieve invariance to low frame-rate gait sequences, the method determines seven key frames from a gait sequence based on silhouette matching by weighted Krawtchouk moments to form a Gait Keyframe Image (GKI). The subjects are identified based on a similarity score obtained by Procrustes image distance between the gallery and probe GKIs. Experimental analysis on OU-ISIR gait dataset D demonstrates the efficacy of the method.