BiYO(5) - A Pioneering Approach for Vision Based Fall Detection Using Bidirectional Feature Pyramid Networks and YOLOv5

View at publisher Summarise with VIRAT
Authors: Priya S, K. Amshakala
Cited by: 0
References: 28

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Fall detection systems play a crucial role in monitoring the safety of the people all around. Severe falls may lead to serious physical injuries and if timely assistance is not provided it may even lead to death. Object detection techniques help to precisely identify and locate the falls. YOLO (You Only Look Once) is one such single pass object detection algorithm which works fine when the distance of the person is at the optimal range from the camera. YOLO failed to recognize the falls in situations like poor resolution images, crowded environments and if the person falls far away from the camera and appear very smaller in size. The proposed work aims to handle these setbacks in real time. The proposed methodology incorporates Bi-directional Feature Pyramid Networks (BiFPN) in the YOLO V5 architecture to extract the comprehensive movements at diverse scales. Adaptive Spatial Feature Fusion (ASFF) fuses the relevant features from the distinct layers of the neural networks to capture more intricate patterns even in the low resolution images. Experimental result analysis proved that the model was more robust as it was able to perform well in the crowded environments, in the environments with low resolution and also if the size of the individual is very smaller in the input frame. Additional Cross Stage Partial Networking (CSP) layers and Coordinate Attention (CA) mechanism are also introduced in the YOLO backbone to enhance the feature extraction process.

Fields of Study & Subjects

Advanced Neural Network Applications Context-Aware Activity Recognition Systems Human Pose and Action Recognition

Publication details

Journal / source
International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
Pages
1–8
Publication date
8 January 2026
Language
EN