Multi-Image Crowd Density Estimation using Multi Column Deep Neural Network

Abstract

Crowd density estimation is a challenging problem for security applications which is a regression task consisting of feature extraction and regression steps. In this paper, multicolumn deep neural networks (MDNN) is proposed for crowd density estimation. Regression task is performed using the feature vector obtained from MDNN. 2000 images selected from the UCSD pedestrian dataset are used in the experiments. In the images, the region of interest (ROI) is filtered and the remaining parts are removed. In order to avoid distortions due to camera position, perspective normalization is applied as a pre-processing step which yields considerable performance improvement. © 2020 IEEE.