RGB-D based human action recognition using evolutionary self-adaptive extreme learning machine with knowledge-based control parameters

Human Action Recognition (HAR) has gained considerable attention due to its various applications such as monitoring activities, robotics, visual surveillance, to name a few. An action recognition task consists of feature extraction, dimensionality reduction, and action classification. The paper proposes an action recognition approach for depth-based input by designing Single Layer Feed forward Network (SLFN) using Self-adaptive Differential Evolution with knowledge-based control parameter-Extreme Learning Machine (SKPDE-ELM). To capture motion cues, we have used Depth Motion Map (DMM) wherein to obtain compact features, Local Binary Pattern (LBP) is applied. Thereafter, for dimensionality reduction, Principal Component Analysis (PCA) is applied to reduce the feature dimensions. For the action classification task, Extreme Learning Machine (ELM) achieves good performance for depth-based input due to its learning speed and good generalization performance. Further, to optimize the performance of ELM classifier, an evolutionary method named SKPDE is used to derive the hidden parameters of ELM classifier. The performance of the proposed approach is compared with the existing approaches Kernel ELM (KELM), L2-Collaborative Representation Classifier (CRC), and Probabilistic CRC (Pro-CRC) using datasets MSRAction3D (with 557 samples), MSRAction3D (with 567 samples), MSRDaily Activity3D, MSRGesture3D, and UTD-MHAD. The proposed approach is also statistically tested using Wilcoxon signed rank-test.