Human Activity Recognition System Using Angle Inclination Method and Keypoints Descriptor Network
May Phyo Ko, Chaw Su, Htut Shie · 2024
Video based Human Activity Recognition (HAR) system is a challenging research topic in the field of computer vision and pattern recognition. The success of many surveillance HAR applications depends on how effectively the learning algorithms are designed. The video-based HAR system have difficult research problem which are complicating nature of human actions such as pose, motion, appearance variation or occlusions. The keypoints descriptor based on deep learning technology offers a competitive solution to these difficulties. In this paper, the HAR system based on deep learning and statistical learning are proposed for detecting the important keypoints that is applicable to scale and describe the orientation or movement of a person. Experiments are conducted on our own dataset. The mean average accuracy of 93.0% is achieved by using the proposed HAR system that can run on lightweight, low computation-intensive device.