ABSTRACT:
Falls are one of the leading causes of serious injuries and health complications among elderly individuals, especially those living alone or without continuous supervision. Immediate medical assistance after a fall plays a crucial role in reducing the severity of injuries and preventing life-threatening situations. Traditional fall detection methods often rely on wearable sensors or manual monitoring, which can be uncomfortable, unreliable, or impractical for long-term use. To address these challenges, this project presents a vision-based Fall Detection System for Elderly using Artificial Intelligence, implemented on a Raspberry Pi platform.
The proposed system utilizes a Raspberry Pi 4 equipped with a CSI camera module to continuously monitor human activity in real time. Human posture information is extracted using MediaPipe Pose Estimation, which detects and tracks 33 body landmarks and converts visual data into numerical skeletal features. These features are used to represent body orientation and movement patterns without storing raw video data, thereby ensuring privacy and reducing computational overhead.
A machine learning–based Random Forest classifier, trained offline using pose landmark datasets containing fall and non-fall activities, is employed to classify human postures. The trained model is saved as a lightweight .pkl file and deployed on the Raspberry Pi for real-time inference. To minimize false alarms, the system incorporates geometric validation based on body orientation and temporal analysis, ensuring that a fall is confirmed only when abnormal posture persists for a predefined duration.
Upon detecting and confirming a fall event, the system automatically sends an SMS alert using the Twilio Cloud Messaging API to caregivers or family members, enabling immediate response and assistance. The system operates entirely on the Raspberry Pi without requiring external servers or high-end hardware, making it cost-effective, portable, and suitable for home-based elderly care.
Experimental results demonstrate that the proposed system achieves reliable fall detection accuracy while maintaining real-time performance on resource-constrained hardware. The integration of computer vision, machine learning, and IoT communication makes the system a practical and scalable solution for enhancing elderly safety and independent living.

INTRODUCTION:
The global increase in the elderly population has raised significant concerns regarding their safety, health, and independent living. Among various health risks faced by older adults, falls represent one of the most serious and common causes of injury, hospitalization, and even mortality. According to medical studies, a large percentage of elderly individuals experience at least one fall each year, and many of these incidents occur when the individual is alone and unable to seek immediate assistance. Delayed medical response after a fall can lead to severe complications such as fractures, internal injuries, long-term disability, and psychological trauma.
Traditional fall detection solutions primarily rely on wearable devices, such as accelerometers or panic buttons, which require continuous user compliance. However, elderly individuals may forget to wear these devices, remove them due to discomfort, or may be unable to activate emergency buttons after a fall. In addition, wearable sensors may generate false alarms during routine activities and often require frequent maintenance and battery replacement. These limitations highlight the need for a non-intrusive, automated, and reliable fall detection system.
Recent advancements in computer vision and artificial intelligence have enabled the development of vision-based monitoring systems capable of understanding human posture and activity without physical contact. Vision-based systems provide the advantage of continuous monitoring while eliminating the need for wearable equipment. When combined with efficient machine learning algorithms and embedded computing platforms, such systems can operate in real time with high reliability and low cost.
This project presents a Fall Detection System for Elderly using Artificial Intelligence, implemented on a Raspberry Pi platform. The system uses a Raspberry Pi CSI camera to capture live video streams and applies MediaPipe Pose Estimation to extract human skeletal landmarks. These landmarks represent key body joints and are used to model posture and movement patterns. Instead of processing raw images, the system works with numerical pose data, which reduces computational complexity and improves processing speed.
A Random Forest machine learning model is trained using pose landmark datasets containing both fall and non-fall activities. The trained model is then deployed on the Raspberry Pi as a lightweight serialized file for real-time inference. To further enhance reliability, the system incorporates geometric posture analysis and temporal validation, ensuring that a fall is confirmed only when abnormal posture persists for a defined duration. This approach significantly reduces false detections caused by sudden movements or brief postural changes.
Once a fall event is detected and confirmed, the system immediately sends an SMS alert using the Twilio Cloud Messaging API to caregivers or family members. This enables rapid response and timely medical assistance, thereby improving the safety and quality of life of elderly individuals. The proposed system is cost-effective, privacy-conscious, and suitable for home-based deployment, making it a practical solution for smart healthcare and assisted living environments.