Video forms the major digital data being consumed. From entertainment to medical research videos from key applications. Appropriate moderation and classification of data is required to ensure appropriate use. Complex data representation makes this task difficult for traditional algorithms. Manual classification which involve monitoring long duration video, poses many limitations. A high performance automated recognition system can reduce such limitations posed. Deep learning, has capability to learn complex features and build models which can be generalized for such activities. We propose a two streamed 3 D deep learning model for automated video behavior analysis. The model is trained on the mice data-set to classify different mice behavior such as drink,eat,groom,hang,micro-movements,rear,rest and walk. Model makes use of features derived from randomly sampled video frames and optic-flow using 3D (2D + 1D) convolution network. With appropriate data and fine tuning the model can be generalized for behavior analysis in video data
Reference to the pdf
Refer to file