Gridding place recognition for fast loop closure detection on mobile platforms

Loop closure detection is an indispensable component to reduce the error accumulation in localisation applications, however, it is hard to be implemented as a real-time application with high accuracy on the mobile platform. A place recognition method called gridding place recognition (GPR) is proposed for ultra-fast loop closure detection with high accuracy maintained. The GPR splits the images into multiple grids for high parallelism and uses bags of word features to summarise each grid. The matching results for each grid are then retrieved and merged for further determination of loop closure under extra temporal and spatial constraints. The authors implement the method on a mobile platform, and the best accuracy and fastest execution time of 47 frames per second are achieved on several public benchmarks compared with other typical or fast loop closure algorithms.