To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Comparison of Two Approaches to Task-specific Real-Time Hand Pose Estimation |
|
Author |
Guram Chaganava† and David Kakulia |
| Citation |
Vol. 21 No. 10 pp. 231-239 |
|
Abstract |
Real-time hand pose estimation in an image plays an important role in systems that require human-computer interaction (HCI). In some cases, a task requires hand pose estimation, not in any, but images with specific content. For example, such a task may require hand pose estimation in images showing one person speaking sign language near the camera. The goal of the study presented in this paper is to experimentally test the assumption that, for the aforementioned specific tasks it will be better than the standard approach to perform hand pose estimation directly in the original image, without hand detection. This approach can result in higher speed and nearly the same accuracy of hand pose estimation as in the case of the standard approach. To determine the advantage of the direct approach for specific tasks, it is necessary to compare the methods in terms of accuracy and speed. For this, a comparative analysis of the standard and direct approaches is carried out. The efficiency coefficients of the methods are quantitatively evaluated to find the optimum between accuracy and speed. It is also examined how the accuracy of hand pose estimation depends on the content of the dataset used to build such a system. As a result, the direct approach proves to be more efficient when using a dataset consisting of images with specific content. |
|
Keywords |
Hand Pose Estimation, Keypoint, Keypoint Detection, Hand Detection. |
|
URL |
