Title: Real-Time Fall Detection for Elderly Care Using YOLO Approach
Author list: Henry Huang1, Angela Zhang2, Jerry Guo3, James Huang1, David Guo4,5
Detailed Affiliations
1Louis D. Brandeis High School, San Antonio, TX USA, 2Basis San Antonio - Shavano Campus, San Antonio, TX, USA, 3Texas A&M University, College of Engineering, College Station, TX USA, 4Department of Information Systems and Cyber Security, University of Texas at San Antonio, San Antonio, TX, USA, 5United Services Automobile Association, San Antonio, TX USA.
Presenter’s email address [
hw719780henry@gmail.com]:
Presenter’s phone number:
Request NSF travel award (DOMESTIC student/postdoc only): Yes ________ No
N___
Have you submitted a paper in addition to your abstract? Yes _______ No
NAbstract body (less than 400 words)
Older adults who are lonely or socially isolated are at significant risk if they experience a fall without immediate assistance, underscoring the necessity for proactive measures. Implementing edge surveillance cameras with fall detection and instant alarm capabilities can significantly mitigate potential harm. The goal of this project focuses on developing an image classification model using the advanced YOLO v8x-cls architecture. The established model, which comprises 5.6 million parameters, is trained to distinguish between annotated binary sample images showing falls (positive) and those without falls (negative). A total of 458 images were collected for training, including 258 images of falling figures and 193 images as negative controls. For testing, 106 images were used, consisting of 70 images with falling figures and 36 without. All images were processed with resize and rotation and saved with the same resolution. With a top-1 accuracy of 95.12% and a perfect top-5 accuracy of 100%, the model demonstrates strong performance in both static images and videos. Additionally, integrating a YOLO v8 pose model further enhances classification accuracy by estimating key points of human figures. This combined approach enables efficient real-time event detection by processing video batches. The project code and pictures for fall detection can be accessed at
https://github.com/s698667/falldowndetectionKeywords: Edge Surveillance, Fall Detection, YOLO
