An integrated artificial vision framework for assisting visually impaired users

摘要

We present a conceptual framework inspired by biological vision which integrates low-level vision functionalities oriented to actions – typical of the so-called “where” dorsal pathway – with identification and recognition capabilities – common to the “what” ventral pathway. Although they proceed independently, these complementary vision models may provide a deeper scene understanding and a more efficient computational framework. In this work, we refer specifically to a set of video analysis modules, which include semantic annotations of the scene and 3D environment interpretation. We discuss and qualitatively evaluate possible connections and integrations between different functionalities, grounding our analysis on a set of specific use cases, depicting visually impaired users finding their way in unfamiliar environments.