Case study of human pointing gestures and their data analysis

摘要

The main contribution of this work is a method to generate datasets of pointing gestures. A person points to tracked objects within a motion capture environment and a neuronal network further processes the tracking data. We found that different input data combination has no significant effect on the performance of the networks. Therefore, it is possible to train a network and obtain correct results, regardless of the complete availability of data for head, shoulder, elbow or hand. With the presented method, we could achieve an overall accuracy of 35 mm within a 2D plane over a distance of 2 m.