Title: A deep regression convolutional neural network using whole image-based inferencing for dynamic visual crowd estimation

Authors: Shen Khang Teoh; Vooi Voon Yap; Humaira Nisar

Addresses: Department of Computer and Communication Technology, Faculty of Information and Communication Technology, Universiti Tunku Abdul Rahman, Malaysia ' Department of Electronic Engineering, Faculty of Engineering and Green Technology, Universiti Tunku Abdul Rahman, Malaysia ' Department of Electronic Engineering, Faculty of Engineering and Green Technology, Universiti Tunku Abdul Rahman, Malaysia

Abstract: As intelligent surveillance system applications become ubiquitous, automated crowd counting solutions must be made continually faster and accurate. This paper presents an improved convolutional neural network (CNN) architecture for accurate visual crowd counting in crowd images. Multi-column convolutional neural network (MCNN) is widely used in previous works to predict the density map for visual crowd counting. However, this method has limitations in predicting a quality density map. Instead, the proposed model is architected using powerful CNN layers, dense layers, and one regressor node with whole image-based inference. Therefore, it is less computationally intensive and inference speed can be increased. Tested on the mall dataset, the proposed model achieved 2.01 mean absolute error and 8.53 mean square error. Moreover, benchmarking on different CNN architectures has been conducted. The proposed model shows promising counting accuracy and reasonable inference speed against the existing state-of-art approaches.

Keywords: visual crowd counting; convolutional neural network; CNN; whole image-based inference; edge embedded platform; multi-column convolutional neural network; MCNN.

DOI: 10.1504/IJBIDM.2023.127306

International Journal of Business Intelligence and Data Mining, 2023 Vol.22 No.1/2, pp.100 - 114

Received: 14 Oct 2021
Accepted: 18 Dec 2021
Published online: 30 Nov 2022 *