連續文字筆跡辨識

Air-writing Recognition

吳俊弘(Chun-Hung Wu)

指導教授 ： 王昱舜

國立交通大學/資訊學院/多媒體工程研究所/碩士(2014年)

https://doi.org/10.6842/NCTU.2014.00396

摘要

因現今微型化的穿戴式裝置欠缺適合且便利的輸入方式，介面太小導致觸控輸入的速度與準確率不高，語音輸入則是缺乏隱私性，因此我們提出air-writing 來當作新的輸入方法。使用者在空中書寫時產生的文字筆跡是連續且不中斷的，文字間將以連接線相接，導致筆跡產生字型上的混亂，增加了辨識文字的困難性。本研究提出一個針對air-writing的文字辨識方法，並使用leap motion來偵測使用者手指的軌跡。我們的方法會將使用者a 至z 的筆跡作為ground truth，再利用dynamic time warping (DTW) 比對input 與ground truth 的相似度，從連續筆跡中找出文字的部分並移除連結線，藉此辨識出文字。由於文字間外觀的相似度高，我們還利用文字的位置、高度、書寫方式來加強辨識的準確性。我們可以將air-writing 應用在微型化的穿戴式裝置上，即使沒有足夠大小的輸入空間，使用者仍可利用air-writing 在任意的時機與地點書寫文字。與常用的語音輸入相比，air-writing 更適合在公共場合使用，這是一種直覺且保有隱私的書寫方式。

關鍵字

手寫辨識 ； 空中書寫

並列摘要

We present a method to recognize text from air-writing. This new input interface will be very helpful to wearable computers such as iWatch or Google glass because the interface is not limited by small screen size. Specifically, our system applies a 3D leap motion controller to capture the user’s finger motion and transfers the motion path line to a sequence of letters. Considering that the air-writing input does not support pen-down and pen-up actions, where all letters are connected and the connecting lines are unclear, the text recognition becomes a challenging problem. To recognize letters at an interactive performance, we first measure the similarity between the pre-collected ground truth letters and the captured finger motion path line. We then solve a letter composition problem, which minimizes the sum deviation of letter shapes, alignments, and collisions, to achieve letter recognition. The experimental results show the feasibility of our algorithm, where 90.5% of the inputs were ranked in the first three positions without the need of vocabulary checking.

並列關鍵字

handwriting recognition ； air-writing

參考文獻

[1] S. Agrawal, I. Constandache, S. Gaonkar, R. Roy Choudhury, K. Caves, and F. DeRuyter.

Using mobile phones to write in air. In Proceedings of the 9th International Conference

NY, USA, 2011. ACM. ISBN 978-1-4503-0643-0. doi: 10.1145/1999995.1999998.

[2] C. Amma, D. Gehrig, and T. Schultz. Airwriting recognition using wearable motion sensors.

[3] C. Bahlmann and H. Burkhardt. The writer independent online handwriting recognition