Study of multiscale detection in near distance image for numbers of people in elevator car
Xu Bin, Jing Mei Yang, Feng Guang Xu, Mangmang Guo · 2015
It is necessary to have a video surveillance of the quantity of people in construction elevator to take down the emergence of casualty like death in group.Considering the relatively tiny spacing in the car and complex background, the results of existing people counting algorithm are not satisfactory.This paper proposes a machine learning recognition method of multiscale ROI image processing for the working conditions of object occlusion and uncertain image size.Multiscale image change according to region is to improve accuracy to object recognition.Then overlapped processing method is used in border area targets.The results of experiment show that the algorithm for object recognition is better in the condition of object occlusion and uncertainimagesize, and greatly reduces runtime and improves accuracy for the condition of complex background in elevator.