People Counting System with C-Deep Feature in Dense Crowd Views

摘要

People counting in a crowded scene is an urgent and vital task of monitoring the surveillance systems. Accrual guesses of a dense crowd views are effected from a different illuminations and inter-class variations, so comes to be a complicated issue and still remains as an active research area. To tackle this fact, this paper establishes an effective people counting framework in dense crowd views that automatically appraisals the accurate number of people. In this paper, a new intuition Color Deep system which utilizes based on the color-based feature and convolutional neural network (CNN)-based feature is proposed for detecting and estimating the people numbers. Unlike the other, this paper proposes C-Deep feature by contributing the color transformation matrix and segmentation. Firstly, the color transformation matrix is introduced and then C-Deep features is calculated by using the Deep CNN with color feature matrix to handle the occlusion, inter-class variations and density levels. Calculation experiments on the challenging public crowd counting dataset achieve the lowest miss rate than state-of-the-art results. This shows the effectiveness of the proposed framework.