adrian-soch / fall_detector

Project comparing traditional ML and Deep Learning approcahes to fall detection

Fall-Detector

Preprocessing

1. Download and unzip datasets.
2. Confirm annotations included the start and stop frames at the beginning of the .txt.
3. Excludes dataset without annotations.
4. Run fallDetectorV0\preprocess.py.
   • this creates the MHIs and the contour feature csv files
   • PCA mean and eigenvalues are saved in fallDetectorV0\pcaEig.joblib

Training

SVM

1. Run fallDetectorV0\training.py
   • sklearn SVM object is saved in fallDetectorV0\svm.joblib

CNN

1. Run fallDetectorV0\training.ipynb.
   • Weights are saved to trained_model\fdnet.pt

Online fall detection

SVM

1. Run fallDetectorV0\onlineFallDetector.py

CNN

Not implemented

Note: Code reference for MHI and CNN training: https://medium.com/diving-in-deep/fall-detection-with-pytorch-b4f19be71e80