Autonomous navigation is a key challenge for mobile robots, and robots that can independently and safely travel from place to place have the capacity to achieve many beneficial tasks. Successful autonomous navigation requires robots know where they are, so it is important the robot can recognize locations within the environment at a later date. Visual imagery captured using cameras provides valuable information about how the world appears. However, the appearance of the world can change a great deal due to the changing time of day, shadows, different weather conditions, or seasonal variations.

We published a survey paper that summarizes the state-of-the-art in visual place recognition. It investigates the key modules that make up a place recognition system: image processing, mapping, and belief generation. It then investigates how these modules need to be adapted to incorporate the notion of appearance change into a system’s model of the world. The survey is available here.