Abstract
Video classification is the task of producing an index or a label that is relevant to the given video. For this classification input can be in frames, heatmaps of respective videos, etc., depending on the application. Video classification is more than just simple image classification, it can be done using the concept of Convolutional Neural Network (CNN). Based on the application, and dataset limitations there are many extensions to this method such as CNN- Long Short-Term Memory (CNN-LSTM), Convolutional 3D network, Pose detection and LSTM, etc. In this project, we used the UCF11 (YouTube Action) dataset and classified them based on the action. A CNN model has been developed and trained with the dataset and the tested results are improved with a better approach called Ensemble Learning. Using this approach, the same CNN model having the same architecture gave better prediction results.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have