Fast Human Detection Using LDA via L1-Norm

Abstract

Fast detection is of vital importance in human detection sometimes. Considering the high dimensions of the features widely used in human detection, it will severely slow the detection speed. Therefore, in this paper, we try to find a way by using Linear Discriminant Analysis(LDA) via L1-norm regularization to solve this problem. It reduces the dimension of feature from 3780 to 150 before classification, and gets a more fast speed then SVM and LDA, while keeps a competitive accuracy.