REAL TIME FALL DETECTION SYSTEM FOR ELDERLY SAFETY USING CNN

Abstract

Abstract- Elderly people who fall have considerable health risks that, if not treated right once, could result in fatalities or major injuries. Recent developments in deep learning methodologies have made it possible to create effective fall detection systems that use input video streams. This research uses deep learning techniques to present a revolutionary smart fall detection system for older people. The suggested system is made up of various essential parts. First, security cameras or wearable technology are used to collect input video data that shows the actions of senior citizens. Next, preprocessing methods are used to lower noise and improve the quality of the video streams. From the preprocessed video frames, feature extraction techniques are used to extract useful information and identify pertinent patterns suggestive of falls. For training and assessment, a large dataset with annotated examples of falls and everyday activities is used. Because convolutional neural networks (CNNs) are good at extracting spatial characteristics from picture input, they are used as the classification model. Using features that are extracted, the CNN model is trained on the dataset to determine which events are falls and which are not falls. In order to identify falls, preprocessed video frames are fed into a trained CNN model, which makes real-time predictions about the probability of a fall. When a fall event is detected, the right warnings or notifications can be sent to emergency services or caregivers, allowing for quick help and intervention. Results from experiments show how well the suggested deep learning-based fall detection system works at correctly identifying fall events while minimizing false alarms. Keywords: Deep learning, Fall detection, Elderly care, Convolutional Neural Networks (CNNs), Video processing, Smart monitoring etc.