Abstract
To improve the accuracy and stability, a doublealgorithm fall detection system is designed. OpenPose and threshold determination are used as algorithm #1 to detect falls, while two neural networks splicing are used independently as algorithm #2 for cross-check. Compared to OpenPose or 3DCNN model alone, the result shows higher F1 and more accurate and pragmatically valuable. A real-time fall detection system is designed based on the study of algorithm #1 and #2. The history data analysis module is added to store and analyze the keyframes of falls, the images of which would be desensitized to protect users’ privacy. A valuable reference model is hereby provided for future products in the potential market.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
