ami190 / video_motiondetection

Using C3N to do video classification

Video Motion Classification

I have dataset of videos with walking and jogging for video classification, created a 3-D CNN model implemented in Keras .

I divide our approach into two parts :

1. Create video files into tensors and store in npy files:

a) Download dataset

b) Extract files and examine folder structure - each folder contains videos belonging to the category

c) Make data ready by generating a CSV file for training and testing, each CSV file containing path to video and category it belongs to.

d) Convert videos to numpy array files

2. Define model load .npy files and train the model :

a) Load numpy files(train_data, train_labels,test_data,test_labels)

b) Pass them into Conv3D model written in Keras

c) Compile and observe results

It gave a testing accuracy of around 50% but greatly reduced time to train the model by already saving data in a numpy array. Since the training data remains the same, there is no need to convert videos into numpy arrays every time we train the model.