Stereo Vision-Based Fall Prediction and Detection using Human Pose Estimation on the AMD Kria K26 SOM
Researchers have developed a vision-based system for predicting and detecting falls in elderly individuals, utilizing human pose estimation on an AMD Kria K26 System-on-Module (SOM). The system captures RGB and depth data, processes it on the edge device to estimate joint keypoints, and then classifies fall activity using a CNN, all while discarding RGB frames to preserve privacy. The multi-threaded pipeline achieved a throughput of 4.5 FPS and demonstrated the feasibility of a cloud-independent, privacy-preserving solution for elderly monitoring. AI
IMPACT Enables privacy-preserving, real-time fall detection for elderly care on low-power edge devices.