Crowd counting using mask region convolutional neural network

• Main Author: Lee, Theresa Ying
• Other Authors: Kai-Kuang Ma
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/149684
• Institution: Nanyang Technological University

Crowd counting aims to provide an estimate of the number of objects (not limited to people), in both sparse and congested environments. The purpose of this is to establish a smart population analysis system which will lead to a smart city. It will be beneficial and can be implemented to increase public safety [1-5], improve congestion monitoring and traffic management [6, 7] and aid in disaster management. As such, there are a variety of state-of-the art techniques that can be used to for crowd counting, starting off with counting by detection, regression, density to deep learning techniques based on Convolutional Neural Networks (CNNs) such as scale-aware models and context-aware models. With diverse applications, crowd counting is applicable from commercial to military purposes and thus, has been deeply studied. In this paper, I have proposed to use Mask Region Convolutional Neural Networks (RCNN) for crowd counting. It contains a simple framework that is easy to train while only adding a small overhead compared to Faster RCNN. In 2016, Mask RCNN produced outstanding results in three segments of the COCO challenge: instance segmentation, bounding-box object detection and person keypoint detection. [8] By applying this framework to the ShanghaiTech Dataset A, a comparison with four other crowd counting techniques will be included.