3D Human Activity Classification with 3D Zernike Moment Based Convolutional, LSTM-Deep Neural Networks

In this paper, we propose a method for classification 3D human activities using the complementarity of CNNs, LSTMs, and DNNs by combining them into one unified architecture called CLDNN. Our approach is based on prediction Zernike Moments some relevant joints body through Kinect Activity Recognition Dataset. KARD includes 18 each activity consists real-world point clouds that have been carried out 3 times 10 different subjects. We introduce potential Moment feature extraction via cloud classification, ability to be trained generalized independently from datasets Deep Learning methods. The experimental results obtained with proposed system has correctly classified 96.1% activities. CLDNN shown provide 5% relative improvement over LSTM, strongest three individual models.