Feature Selection Using Genetic Algorithms for Human Gait Recognition.

摘要

Many research studies have demonstrated that gait can serve as a useful biometric feature for human identification at a distance. Here we manifest the importance of feature selection in gait recognition systems. Feature selection is an important factor which impacts the classification accuracy. This goal is achieved by discarding irrelevant and redundant information which affects both the classifier's performance and system's efficiency. Traditional gait recognition systems have mostly been evaluated without considering the most relevant features. In this study, we are going to investigate the use of Genetic Algorithm (GA) for selecting an optimal subset of features for a model-free gait recognition approach without degrading the classification accuracy. First, features are extracted using Kernel Principal Component Analysis (KPCA) on four spatio-temporal projections of silhouettes. Then, GAs are applied to choose a subset of Eigen-vectors that represent a subject's identity. Our experimental results, conducted on Georgia Tech (GT) database, indicate considerable performance improvements.