Abstract:
For people with visual disabilities, the use of video-call application on mobile devices can be an essential means to remotely communicate with their assistants whenever tasks require help from sighted person. However, such applications on the market were not readily usable by blind users. One major drawback is that the assistant can only view current video stream. When the blind fails to take necessary scenes, the assistant must direct the blind to re-shoot the video until required task can be accomplished. Apparently, it is highly challenging for the blinds to capture the videos at the right spot. Thereby, this thesis focuses on constructing large scene by stitching images real-time from video streams taken by the users. A new system, Tor Sang Suk, was developed to include our proposed algorithm. It comprises three components. The first component, for the blind users, is a WCAG complied video-call application on Android that simply captures and sends videos. The second component is the communication between Android device the PC, enabled through WebRTC. The third component for the assistants contains our proposed remote scene construction algorithm based on SIFT (Scale-Invariant Feature Transform) that allows up to five threads of disconnected video streams. It also features real-time zooming function and graying of older portion of images. We evaluated Tor Sang Suk by one blind user, two assistants, and two scenarios. For each of the scenarios, the blind user performed the same task twice per one assistant, once using Tor Sang Suk with remote scene construction feature, and another without. For the two times the tasks were performed, it was done with different assistant to avoid prior knowledge from the assistant. We measured the time to complete each task and found that Tor Sang Suk with remote scene construction showed 31% time reduction, or an average of 1.75 minutes.