A Nearly Optimal Landmark Deployment for Indoor Localisation with Limited Sensing

Indoor applications based on vehicular robotics require accurate, reliable and efficient localisation. In the absence of a GPS signal, an increasingly popular solution is based on fusing information from a dead reckoning system that utilises on-board sensors with absolute position data extracted from the environment. In the application considered in this paper, the information on absolute position is given by visual landmarks deployed on the floor of the environment considered. This solution is inexpensive and provably reliable as long as the landmarks are sufficiently dense. On the other hand, a massive presence of landmark has high deployment and maintenance costs. In this paper, we build on the knowledge of a large number of trajectories (collected from environment observation) and seek the optimal placement that guarantees a localisation accuracy better than a specified value with a minimal number of landmarks. After formulating the problem, we analyse its complexity and describe an efficient greedy placement algorithm. Finally, the proposed approach is validated in realistic use cases.