Dynamic hand gesture recognition on 3D virtual cultural heritage ancient collection objects using k-nearest neighbor

This paper discusses on how to prepare a specific dynamic hand gesture, modeling and testing it to interact with 3D virtual objects of cultural heritage ancient collection. Those virtual objects prepared to avoid damage on the original one. Several kinds of research work for recreation or reactivating ancient heritage for educational purposes can take place using it. The dynamic hand gesture detected using hand movement sensor. We recorded ten specific dynamic hand gesture that stands for the interaction between museum visitors and the ancient collection chosen for the test. All ten gestures consist of fingers tips coordinates, palm, and wrist movement. A Total of 14474 rows in 30 features forming fingers and palm movements information. Those gestures namely: pick-up, sweep from right to left, sweep from left to right, grab from above, grab from the right side, pinch from above, pointing, scooping, push and picking. We train ten dynamic hand gestures using K-NN classifier and using different distance metric namely Cosine, Euclidean and Cubic. The best result of trained model reaches 99.3% accuracy. Later, we use the new hand gesture to test the trained model. It consists of 15000 rows of fingers coordinates in 30 features. The results show that from all ten gestures, there are four gestures reach recognition accuracy more than 92%. One gesture reaches 100%, two gestures on 82% and 89% and three gestures below 64%. The gesture which reaches high accuracy in training and testing consider selected for default model.