| Title |
Character Recognition in Smart Glasses Using Artificial Intelligence Learning |
| Authors |
박재훈(Jaehun Park) ; 김재욱(Kim, Jaeuk) ; 김수찬(Soochan Kim) |
| DOI |
https://doi.org/10.5573/ieie.2019.56.6.55 |
| Keywords |
Gaze-tracking ; Deep-learning ; OCR ; Image-processing ; Smart-glasses |
| Abstract |
We applied the gaze point estimation technique to smart glasses such as Google Glass to enhance user’s convenience by automatically extracting gazed information when the gaze is fixed to the user's interest. The gaze point was estimated by applying the pupil center and the glint direction vector, and the characters at gazed area were recognized from the image of the gaze area by the deep learning technique. The recognized letters and sentences are output as voice for user convenience. The system consists of two USB cameras and speakers based on the Raspberry Pi 3 model. In the proposed system, the accuracy of the character recognition model was 92% in the learning set and 91% in the test set. When the distance to the monitor and the movement of the head were fixed during the calibration process, the gaze point could be processed at 10 frames per second within ± 25° in the measurement range with an error of 0.8° horizontally and 1.1° vertically. The proposed system is advanced from conventional fixed-system approaches and it informs vocally the semantics of recognized characters when the smart glass wearer moves freely and fixes gazes. Additionally, it can be applied in assisting the visual function or foreign language. |