Human activity recoginition with Deep learning

Final project for Nile university CIT-651 Machine learning class - Fall 19 Experiments are performed on Pascal VOC 2012, Kaggle statefarm distracted driver and UCI Har dataset.

UCI HAR Dataset

Fully Connected network

Trained on Raw data features 561x1.

1D CNN

Trained with time stamp readings 9x128.

VOC 2012 Dataset (Indoor, Outdoor, ... )

Trained with 512x512 images with/without bounding box cropping with random augmentation.

Statefarm dataset (Inside vehicle)

Trained with 480x640 images with random augmentation.

References:

Template structure and config scripts from Template link