Randomized decision bush: Combining global shape parameters and local scalable descriptors for human body parts recognition

摘要: This paper presents a novel method which combines global shape parameters and scalable local descriptors for accurate body parts recognition from single depth image in real-time. Human poses are of extremely large variation aspects visual shapes, because human can take daily activities to gymnastic actions. In order cover wide-range the poses, proposed algorithm employs unified structure pose clustering classification. We name Randomized Decision Bush (RDB). Specifically, discriminate coarse level shapes utilized while employed RDB splits various into multiple clusters contain similar poses. As result, it provides robust enables fine classification within cluster. The experimental results show improvements on recognizing due with descriptors. Additionally, we significantly reduce complexity training number shapes.