Because indoor navigation is difficult for people with visual impairment, there is a need for the development of assistive technology. Indoor location sensing, the ability to identify a pedestrian's location and orientation, is a key component of such technology. We tested the accuracy of a potential crowdsourcing-based indoor location sensing method. Normally sighted subjects were asked to identify the location and facing direction of photos taken by a pedestrian in a building. The subjects had available a floor plan and a small number of representative photos from key locations within the floor plan. Subjects were able to provide accurate location estimates (median location accuracy 3.87ft). This finding indicates that normally sighted subjects, with minimal training, using a simple graphical representation of a floor plan, can provide accurate location estimates based on a single, suitable photo taken by a pedestrian. We conclude that indoor localization is possible using remote, crowdsourced, human assistance. This method has the potential to be used for the location-sensing component of an indoor navigation aid for people with visual impairment.

We report on three experiments that investigate the efficacy of a new type of interface called a virtual verbal display (VVD) for nonvisual learning and navigation of virtual environments (VEs). Although verbal information has been studied for route-guidance, little is known about the use of context-sensitive, speech-based displays (e.g., the VVD) for supporting free exploration and wayfinding behavior. During training, participants used the VVD (Experiments I and II) or a visual display (Experiment III) to search the VEs and find four hidden target locations. At test, all participants performed a route-finding task in the corresponding real environment, navigating with vision (Experiments I and III) or from verbal descriptions (Experiment II). Training performance between virtual display modes was comparable, but wayfinding in the real environment was worse after VVD learning than visual learning, regardless of the testing modality. Our results support the efficacy of the VVD for searching computer-based environments but indicate a difference in the cognitive maps built up between verbal and visual learning, perhaps due to lack of physical movement in the VVD.

For pt. XVIII see Vis. Res., vol. 38, no. 19, pp. 2949-62 (1998).Low vision is any chronic form of visual impairment, not correctable by glasses or contacts that adversely affects performance of important everyday visual tasks. Most people with low vision need magnified text to read. On a fixed-size computer screen, the magnification of text trades off against the proportion of the entire screen visible. To read hypertext, simultaneous access to the full-screen page is important for skimming text and for locating hyperlinks. Therefore, people with low vision using magnified text might encounter difficulties reading hypertext, especially when hyperlinks are placed at unpredictable locations (true for most webpages). We investigated hypertext information retrieval as the time taken and number of nodes traversed to answer a series of questions. In Experiment 1, low-vision performance for reading prose and hypertext was compared to normal performance: low-vision performance deficits in hypertext retrieval were predictable from deficits in conventional prose reading. Experiment 2 evaluated the effect of web-page layout on low-vision performance: retrieval performance was severely affected when hyperlinks had unpredictable locations. This extra deficit was eliminated when users were provided with simultaneous access to full-screen layout. Based on these findings, we discuss the accessibility of the Internet by people with low vision.