A Wearable Cognitive Vision System for Navigation Assistance in Indoor Environment

Existing mobile navigation techniques are not applicable for indoor navigation. Obviously, the best navigator is a human companion. In this paper, we explore to build a wearable virtual navigator for indoor navigation. A novel cognitive vision system is designed which consists of long-term memory and working memory for complicated vision tasks in dynamic environments. The long-term memory mimics the flexibility and scalability of human cognitive memory for domain knowledge representation, and the working memory emulates the routine process and attention selection in human cognitive model for online visual perception. Efficient algorithms for image classification and object detection are organized and performed under cognitive perception framework to achieve real-time performance. Field tests demonstrate its effectiveness and efficiency by recognizing scenes, locations, and landmark objects in real-time, and subsequently providing context-aware assistant to guide the user in the navigation of a complex office environment.