Frontal Gait Recognition from Incomplete RGB-D Streams Using Gait Cycle Analysis

Gait as a significant biometric feature in human identification is drawing a wide attention nowadays. In many real-life surveillance zones such as banks, airports and corridors, gait recognition is often restricted from the front view. There are situations where a complete gait cycle is not always available due to frame drop caused by devices and the limitation in space of such areas, while most of the existing methods require at least one complete gait cycle. A novel method is proposed to achieve a high recognition rate in such application scenarios, based on dividing a gait cycle into several phases using Constrained Fuzzy C-Means method and converging feature information of a stream into one feature descriptor using gait cycle analysis. Experimental results demonstrate the high performance of our method comparing to other existing ones.