Blind, or low (Blv) people do not hesitate to travel independently in unknown areas because of uncertainty in the form of the body. While many tools focus on In-Situ wandering, those who examine pre-travel aid usually provide only instructions and turning down instructions, lacking visible context. Pictures of road views, containing rich information to look and has the ability to express numerous environmental information, remains inaccessible to BLV. In this work, we present a picture, a great Multimodal Language (MLM) model in Ai agent that allow access to road viewing. Scenescout supports two methods: (1) Views, enabling us to familiarize themselves with visual information on the side of the road, (2) Visual assessment, enabling free movements within the Street View picture. Our usernial research (n = 10) shows that Scenescout scenes help BLV users open visual information not available in existing ways. Technical examination shows that many explanations are accurate (72%) and describe the objects that are in the oldest (95%) even in old photos, although smaller errors make it difficult to ensure without seeing. We discuss future opportunities and challenges to use a road view to improve navigation.
- † work done while in apple
- ‡ Columbia University
