Human-like Guidance for Driving Navigation in an Urban Environment

Driving is a cognitively demanding task, and many current navigation systems present confusing guidance instructions that add to the distraction. Human navigators, by contrast, schedule their advice to minimise distraction, and phrase instructions in terms of visible landmarks to avoid confusion. In this paper, we present the basis for a 'natural navigation' system which interprets distances as references to landmarks. We use Extended Kalman Filtering to integrate visual odometry with other sensor data in order to obtain precise vehicle motion, then, based on the filtered motion parameters, we characterize recognised visual landmarks as locations on the navigational map. The navigation system can then use references to these landmarks in its driver instructions rather than absolute distances. Experimental results show that landmarks can be located on the navigational map with sufficient accuracy using normal vehicle telemetry and a dashboard camera.