An Edge-Device Based Fast Fall Detection Using Spatio-temporal Optical Flow Model
(12/2020-04/2021)
- Designed a CNN model that integrated RGB and optical flow to extract spatio-temporal features from the object detected before.
- Developed a tensor-compressed LSTM architecture to process the fused feature and detect falls in real-time on edge devices.
- Achieved accuracy of 96.23% and 99.37% on Multicam and URFD, with speed of 83.3 FPS and a storage reduction of 210.9x
- Authored and published a paper in 43rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society.
