An Effective reduction of Gait Recognition Time by using Gender Classification

Abstract

This work investigates the possibility of utilizing gender classification in gait recognition to reduce search space time as gender classification reduces the total number of search subjects from database. In this paper, gender classification is performed by utilizing Sparse spatiotemporal features along with most effective features, more informative less effective features and shape features. The spatiotemporal interest points are detected by improved Harris corner detector which is simulated annealing optimized Harris corner detector. The spatiotemporal interest points among the video frames are treated as the central point around which a cuboid of pixel values are extracted and processed to form final descriptor. The most effective features and more informative less effective features are extracted by the cofactor entropy while the shape features are extracted using angular radial transform and FFT and the feature fusion is performed using BAT as in our previous works. Then the Sparse Multi-kernel Support Vector Machine (SM-SVM) is used for subject classification. Experimental results show that the proposed method provides efficient gender classification and thus reduces the gait recognition time.