Human Activity Recognition using Machine Learning Approach


The growing development in the sensory implementation has facilitated that the human activity can be used either as a tool for remote control of the device or as a tool for sophisticated human behaviour analysis. With the aid of the skeleton of the human action input image, the proposed system implements a basic but novel process that can only recognize the significant joints. The proposed system contributes a cost-effective human activity recognition system along with efficient performance in recognizing the significant joints. A template for an activity recognition system is also provided in which the reliability of the process of recognition and system quality is preserved with a good balance. The research presents a condensed method of extraction of features from spatial and temporal features of event feeds that are further subject to the mechanism of machine learning to improve the performance of recognition. The significance of the proposed study is reflected in the results, which when trained using KNN, show higher accuracy performance. The proposed system demonstrated 10-15% of memory usage over 532 MB of digitized real-time event information with 0.5341 seconds of processing time consumption. Therefore on a practical basis, the supportability of the proposed system is higher. The outcomes are the same for both real-time object flexibility captures and static frames as well.