Human Pose Estimation for Early Action Prediction using Progressive Skeleton Data

Human surveillance to analysis human activity is one of the most important for security system. In human activity is now to early action prediction to predict the action before fully execute action. Our approach is about to combine progressive scaling data with 2D skeleton data. For the dataset that matching with pose estimation and CCTV like is NTU with only RGB data to extract joint and limp in 2D frames input. Using MoViNetsA-5 to reduce memory used for the limit resource in 3D Convolutional neural network.