We have enhanced the capabilities of human pose estimation using the openpifpaf library. Our augmentation includes support for multi-camera and multi-person tracking. Additionally, we have incorporated a long short-term memory (LSTM) neural network into the system to predict two classes: "Fall" or "No Fall". To accomplish this, we extract five temporal and spatial features from the detected poses. These features are then fed into an LSTM classifier for further processing and classification.
pip install -r requirements.txtThe UP-Fall Detection was used to train the LSTM model.
Here are some of the desired outputs by the fall detector
