Improvements in 3D Hand Pose Estimation Using Synthetic Data
| Autor: | Dmitry Ryumin, Jakub Kanis, Zdeněk Krňoul |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Matching (statistics)
Training set Artificial neural network Computer science business.industry 05 social sciences Process (computing) 020207 software engineering 02 engineering and technology Real image Synthetic data Image (mathematics) 0202 electrical engineering electronic engineering information engineering 0501 psychology and cognitive sciences Computer vision Artificial intelligence business Pose 050107 human factors |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783319995816 ICR |
| DOI: | 10.1007/978-3-319-99582-3_12 |
| Popis: | The neural networks currently outperform earlier approaches to the hand pose estimation. However, to achieve the superior results a large amount of the appropriate training data is desperately needed. But the acquisition of the real hand pose data is a time and resources consuming process. One of the possible solutions uses the synthetic training data. We introduce a method to generate synthetic depth images of the hand closely matching the real images. We extend the approach of the previous works to the modeling of the depth image data using the 3D scan of the subject’s hand and the hand pose prior given by the real data distribution. We found out that combining them with the real training data can result in a better performance. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|