Annotated Hand-Depth Image Dataset and its
Performance Evaluation

This website contains a set of annotated depth images of various hand poses taken from multiple volunteers. This can be used to evaluate the performance of a hand pose estimation algorithm. The depth images are acquired through a time-of-flight (TOF) camera (SoftKinetic DS 325). The annotations are obtained using a dataglove (Measurand ShapeHand). Without loss of generality, we focus only on the right hand. The images are collected from 30 volunteers varying in age (18 - 60 years), gender (15 male and 15 female), race and hand size. 29 images are obtained for each volunteer during the capture sessions, which can be categorized into two groups. 8 of these images are from the Chinese Number Counting system (from 1 to 10, excluding 3 and 7), while the remaining are from the American Sign Language alphabet (from A to Z, excluding J, R, T, W and Z). Together, these amount to 870 annotated examples, with each example consisting of a hand depth image and its label (i.e. 3D positions of the hand joints following the hand kinematic model). Aiming at a fair performance evaluation of hand pose estimators, the data glove annotations are removed from examples on which the online evaluation will be conducted. For the rest of the examples, annotations are provided to facilitate model training.


When using this data in your work, you may cite the following paper:-

 author = {C. Xu and L. Cheng},
 title = {Efficient Hand Pose Estimation from a Single Depth Image},
 booktitle = {ICCV},
 year = {2013},

Related Publications

Xu, C. & Cheng, L. Efficient hand pose estimation from a single depth image. International Conference on Computer Vision (ICCV), 3–6 December 2013. | PDF | Website |

Chi Xu, Ashwin Nanjappa, Xiaowei Zhang, Li Cheng. Estimate Hand Poses Efficiently from Single Depth Images. In In International Journal of Computer Vision (IJCV), 2015. | PDF | Website |