Video Analytics for Real‑Time Fall Detection & Activity Recognition

Overview

I lead the development of a real‑time fall detection and daily‑activity recognition system to improve safety and quality of life for independently living older adults.

What I built

• Designed a real‑time video analytics pipeline for fall detection with low latency and high recall.
• Created a custom dataset of falls and daily activities; used pose estimation + ML to boost detection accuracy.
• Deployed edge inference on NVIDIA Jetson Orin Nano for on‑device processing and immediate cloud‑based alerts to caregivers.

Tech & Methods

• PyTorch · TensorRT · OpenCV · Pose Estimation · HAR models · ONNX · Jetson Orin Nano
• MLOps for dataset versioning, evaluation, and model iteration

Impact

• Enables immediate alerts with privacy‑preserving, on‑device processing.
• Architecture designed for 24/7 monitoring and scalable caregiver dashboards.