MLP based Pedestrian Counting

Kang-Min Kim, Gyu-Jin Kim, Gyu-Jin Jang, Young-Nam Kim, Moon-Hyun Kim

Abstract: In this paper, we presented an efficient pedestrian counting method for estimating the total number of crowed pedestrians in the video. For this, it detect interest points of the foreground area and the principal components of features - the number of foreground pixels, the number of edge pixels, and textures. These interest points and the principal components of features as an input value of multi-layer neural network that has been presented in this paper, it was learned in the back propagation algorithm utilizing sigmoid function.