Human Identification And Gender Recognition From Boxing

We describe an approach of human identification and gender classification based on boxing action. A period detection approach based on time-involved-cutting-plane is first applied and then a boxing sequence of a period is represented by an averaged silhouette. A Nearest Neighbor classifier based on Euclidian distance is used for human identification. The experiments were carried out on the KTH boxing dataset on which the accuracy can reach 80% or higher. After dimensionality reduction by PCA, a SVM is used for gender classification. The experimental results on a dataset containing 20 males and 20 females demonstrate that by applying the proposed algorithm the gender recognition can reach the accuracy of 80% or higher. We also present a numerical analysis of the contributions of different human components. Experimental results show that the head has a positive impact on system performance with the basis of the arm while the buttocks and the leg have not.