ViCo-MoCo-DL: Video Coding and Motion Compensation Solutions for Human Activity Recognition Using Deep Learning

This paper proposes three novel feature extraction approaches for human activity recognition in videos. The proposed solutions are based on video coding concepts including motion compensations and variables. We use these features with deep learning model generation classification, hence the ViCo-MoCo-DL abbreviation which stands Video Coding Motion Compensation Deep Learning. These fused terms of averaging their classification scores to predict In all solutions, an input is temporarily segmented into 12 non-overlapping segments equal size. first second solution each segment converted one component RGB image, thus resulting 4 images. conversion happens capture using estimate, compensation accumulating image prediction errors. Consequently, solution, generated images tiled big used train a Convolutional Neural Network (CNN) network. entered pre-trained CNN extraction. resultant FVs arranged matrix training Long Short-Term Memory network (LSTM). third customized High Efficiency Coder (HEVC) generate variables per frame. Feature Vectors (FVs) 3 numerically summarized FV, thus, represented by another LSTM Experimental results well-known datasets show superior over existing work.