Human Activity Recognition by Using Nearest Neighbor Algorithm from Digital Image

Abstract

Human activity recognition is done by using computer vision technique. The feature extraction method is performed to recognize human activity by analyzing the poses. This research proposes a ratio comparison of human shape as a feature extraction method. Feature extraction, data for human activity recognition was generated by measuring object height and width. Nearest neighbor algorithm was performed to recognize the human activity. The best nearest neighbor configuration (k value) used to design human activity recognition system. Data testing was constructed by using Cross Validation and k-Fold Cross Validation methods. Nearest Neighbor with k=3 reaches the highest accuracy to identify the poses. This configuration provides 90% recognition accuracy in Cross Validation and 60% recognition accuracy in k-Fold Cross Validation. This research argues the best Nearest Neighbor configuration of this research is k=3.