Traditional Features based Automated System for Human Activities Recognition

Human Activities Recognition (HAR) is an important research topic and its applications are spread in all the fields of computer vision and machine learning including video surveillance, robotics, and name a few more. In this paper, a new traditional feature fusion and selection-based method is proposed for automated HAR. The proposed methodology consists of three core steps- optical flow-based motion region extraction and later ROI detection, shape and gray level difference matrix (GLDM) features are combined in one matrix based on seniority value indexes, and finally, Reyni entropy-controlled Euclidean classifier based best features selection. The final selected features are put to Cubic SVM for final recognition. The validation of the proposed technique is conducted on three datasets- KTH, YouTube, and Weizmann and achieved an accuracy of 99.30%, 99.80%, and 99.60%, respectively. Overall, Cubic SVM outperforms among existing techniques.