Visual Crowd Counting System Using Deep Learning

• Main Author: Mohd Wafi Nazrul Adam, Mohd Ridhwan Oxley Adam
• Format: Final Year Project / Dissertation / Thesis
• Published: 2021
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utar.edu.my/4286/1/17ACB05930_FYP2.pdf; http://eprints.utar.edu.my/4286/

This project is about developing a visual crowd counting system using deep learning. The entirety of this project will only be using Python for both the back-end and the front-end development. The goal of this project is to develop a working system that could take in images and estimate the number of crowds in those images as well as display it’s estimated density map and a graph of predicted count against its ground truth as well as its accuracy in Mean Absolute Error (MAE) and Mean Squared Error (MSE). The back-end will be using a neural network model based on the Single-Image Crowd Counting via Multi-Column Convolutional Neural Network (Zhang, et al., 2016) and is developed through the PyTorch framework, an open-source machine learning library. The model will be trained using the Mall Dataset and the Adam optimization algorithm. The trained model has an accuracy of 2.45 in MAE and 9.72 in MSE when tested using the Test portion of the dataset. The front-end is developed from scratch using the PyQT5 toolkit and QtDesigner.