| Presentation | 2019-09-06 A Machine Learning-based Character Identification System for the Visually Impaired to Enjoy Broadcast Animations Yu Yoshino, Kazuki Nakada, Makoto Kobayashi, Iwao Sekita, Hisayuki Tatsumi, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | This study aims to assist visually impaired individuals by focusing on the following problems that arise at the time of viewing animation videos and images. (1) difficulty of understanding behaviors and situations, (2) difficulty of discriminating animation characters, and (3) confusion caused by animation characters with similarities. To identify the target animation character on the above problems, we are going to make a support equipment for character identification using machine learning. The machine learning framework has been constructed so that students as users can achieve machine learning with desired animation datasets themselves. In our framework, we combined the area detection using a cascade classifier and the face discrimination using deep learning to balance of the easiness of learning and the robustness of discrimination accuracy. For efficient learning from desired small dataset of animation characters, we applied the transfer learning to the extended convolutional neural network (CNN) model pre-trained with ImageNet, and we confirmed that the bottleneck features of the learned CNN model are effective for identifying animation characters. Moreover, we implemented the customized CNN model trained by a student himself on a hardware accelerator and verified the real time operation in practical environments. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Visual impairment assistance / Animation character / Automatic identification / Deep Neural Network / Transfer learning |
| Paper # | ET2019-31 |
| Date of Issue | 2019-08-30 (ET) |
| Conference Information | |
| Committee | ET |
|---|---|
| Conference Date | 2019/9/6(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Tsukuba University of Technology |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Special Needs Education and Welfare Support, etc. |
| Chair | Hideyuki Suzuki(Ibaraki Univ.) |
| Vice Chair | Ryo Takaoka(Yamaguchi Univ.) |
| Secretary | Ryo Takaoka(Waseda Univ.) |
| Assistant | Megumi Kurayama(National Inst. of Tech., Hakodate College) / Ryo Oonuma(Fukushima Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Educational Technology |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A Machine Learning-based Character Identification System for the Visually Impaired to Enjoy Broadcast Animations |
| Sub Title (in English) | |
| Keyword(1) | Visual impairment assistance |
| Keyword(2) | Animation character |
| Keyword(3) | Automatic identification |
| Keyword(4) | Deep Neural Network |
| Keyword(5) | Transfer learning |
| 1st Author's Name | Yu Yoshino |
| 1st Author's Affiliation | Tsukuba University of Technology(NTUT) |
| 2nd Author's Name | Kazuki Nakada |
| 2nd Author's Affiliation | Tsukuba University of Technology(NTUT) |
| 3rd Author's Name | Makoto Kobayashi |
| 3rd Author's Affiliation | Tsukuba University of Technology(NTUT) |
| 4th Author's Name | Iwao Sekita |
| 4th Author's Affiliation | Tsukuba University of Technology(NTUT) |
| 5th Author's Name | Hisayuki Tatsumi |
| 5th Author's Affiliation | Tsukuba University of Technology(NTUT) |
| Date | 2019-09-06 |
| Paper # | ET2019-31 |
| Volume (vol) | vol.119 |
| Number (no) | ET-200 |
| Page | pp.pp.35-40(ET), |
| #Pages | 6 |
| Date of Issue | 2019-08-30 (ET) |