CNN+Bi-directional-LSTM; Video classification; Four categories（Normal; Smoking; Using mobile; Off seat）
code -> optic.py[-d The directory of dataset]: This script USES the optical flow to process dataset images.
vgg_lstm.py: Model definition.
log_IBRD.py: Log format definition.
dataset -> img -> Train-Image & Test-Image
h5 -> The weights of the VGG model and the Bi-LSTM model.
npy -> The feature maps of data.
The dataset you can download on Driver Behavior Dataset
Package details:
Python 2.7
h5py 2.7.0
Keras 1.2.2
numpy 1.13.1
opencv 2.4.11
Theano 0.8.2