To evaluate and compare methods in computer vision, scientists must use a benchmark dataset and unified sets of measurements. The UCI-EGO dataset is a standard benchmark dataset for evaluating Hand Pose Estimation (HPE) on depth images. To build robotic arms that perform complex operations such as human hands, the poses of the human hand need to be accurately estimated and restored in 3D space. In this paper, we standardized the UCI-EGO dataset to evaluate 3D HPE from point cloud data of the complex scenes. We also propose a method for fine-tuning a set parameter to train the estimation model and evaluating 3D HPE from point cloud data based on 3D Convolutional Neural Networks (CNNs). The CNNs that we use to evaluated currently the most accurate in 3D HPE. The results of the 3D HPE from the point cloud data were evaluated in two branches: using the hand data segment and not using the hand data segment. The results show that the average of 3D joint errors of the 3D HPE is large on the UCI-EGO dataset (87.52mm) and that the error without using the hand data segment is many times higher than the estimated results when using the hand data segment (0.35ms). Besides, we also present the challenges of estimating 3D hand pose and the origin of the challenge when estimating real image dataset.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.