Indoor and Outdoor Crowd Density Level Estimation with Video Analysis through Machine Learning Models
Mahira Arefin, Md. Anwar Hussen Wadud, Anichur Rahman
TL;DR
The paper addresses automatic indoor and outdoor crowd density level estimation from video by building a Python-based pipeline that uses a Caffe-based MobileNetSSD detector to identify and track people, count individuals, and categorize crowd density with FPS metrics. It leverages a Kaggle crowd-counting dataset for evaluation and reports high accuracy (over 97%), emphasizing ease of use and cost-effectiveness. The approach combines object detection, tracking, counting, and density labeling (normal/medium/high) with graphical data representations to support real-time crowd management. The work highlights practical surveillance applications and outlines plans to expand models, datasets, and speed for faster, more accurate density estimation.
Abstract
Crowd density level estimation is an essential aspect of crowd safety since it helps to identify areas of probable overcrowding and required conditions. Nowadays, AI systems can help in various sectors. Here for safety purposes or many for public service crowd detection, tracking or estimating crowd level is essential. So we decided to build an AI project to fulfil the purpose. This project can detect crowds from images, videos, or webcams. From these images, videos, or webcams, this system can detect, track and identify humans. This system also can estimate the crowd level. Though this project is simple, it is very effective, user-friendly, and less costly. Also, we trained our system with a dataset. So our system also can predict the crowd. Though the AI system is not a hundred percent accurate, this project is more than 97 percent accurate. We also represent the dataset in a graphical way.