Low Cost Perception of Dense Moving Crowd Clusters for Appropriate Navigation

Abstract

This paper describes an algorithm for rapidly and cheaply clustering humans moving in clusters during dense crowd conditions. The algorithm differs from other methods due to a focus on low cost, use of a single robot-mounted RGB-D sensor, and design choices driven by human perception and expectations of appropriate behavior. Use of a human-centered algorithm design process should lead to actions and decisions that are more aligned with human expectations since robots will mimic human perception and guesses during dense crowd motion.