MobileASL: Overcoming the technical challenges of mobile video conversation in sign language
As part of the ongoing MobileASL project, we have built a system to compress, transmit, and decode sign language video in real-time on an off-the-shelf mobile phone. Compression and transmission of sign language video presents unique difficulties. We must overcome weak processing power, limited bandwidth capacity, and low battery life. We also must ensure that the system is usable; that is, that the video is intelligible and the algorithms that we employ to save system resources do not irritate users.
We describe the evolution of the MobileASL system and the algorithms we utilize to achieve real-time video communication on mobile phones. We first review our initial user studies to test feasibility and interest in video sign language on mobile phones. We then detail our three main challenges and solutions. To address weak processing power, we optimize the H.264 encoder to work on mobile phones, adapting a fast algorithm for distortion-complexity optimization to choose the best parameters. To overcome limited bandwidth capacity, we utilize a dynamic skin-based region of interest, which encodes the face and hands at a higher bit rate at the expense of the rest of the image. To save battery life, we automatically detect periods of signing and lower the frame rate when the user is not signing. We implement our system on off-the-shelf mobile phones and validate it through a user study. Fluent ASL signers participate in unconstrained conversations over the phones in a laboratory setting. They find the conversations with the dynamic skin-based region of interest more intelligible. The variable frame rate affects conversations negatively, but does not affect the users’ perceived desire for the technology.