A fuzzy inferencing system for gait recognition

This paper proposes a method for the identification of individuals from their gait using fuzzy logic. Gait signature is first extracted in the form of a spatiotemporal representation called Gait Energy Image (GEI). Since the dimension of GEI is very high, we use fuzzy principal component analysis (FPCA) for dimension reduction. Unlike traditional PCA, it helps to get rid of the problems of outliers and missing data. Classification of test pattern is done based on two aspects, namely, proximity towards each class and the count of training samples corresponding to each class that are close to the test sample. The impreciseness and subjective nature of these two terms motivate us to use fuzzy logic which is capable of capturing and modeling the uncertainty inherent to the problem domain. We introduce a new gait recognition method based on a fuzzy inferencing system. Feature vectors resulting from the FPCA transformation are used by the fuzzy inferencing system to compute the class assignment. Comprehensive experiments carried out on CMU's Mobo dataset show promising classification rates. The performance has been compared with other approaches and is found to attain improved result.