Abstract: Aiming at problem that the hand joint is difficult to identify and different joint is difficult to distinguish based on depth image, an algorithm based on rotation invariant depth comparison feature is proposed. First, the rotation invariant depth comparison feature is extracted to train random forest classifier and identify the part category of pixel. Then density of centroid of perpoint mapping is calculated to achieve joint recognition. The experimental result shows that the algorithm is of high accuracy and robustness.

Key words: depth image, part, rotation invariant depth comparison features, random forest, algorithm