A Novel Image Model of Point Clouds and its Application in Place Recognition

In this paper, we present a novel panoramic image model for scattered point clouds, and apply it to the problem of place recognition. We project a point cloud onto a sphere, and then the sphere is divided into a set of individual grids by longitudes and latitudes. Each grid is regard as a pixel and its value is computed using the geometrical relationship among the points in the grid and its neighbors. For convenience, the sphere is transferred into a flat. Since point clouds are converted to 2D images, we use ORB features and bag of words technique to solve place recognition problem. Our experimental results show that our image model is a more universal one and achieve a good performance in place recognition in both accuracy and efficiency.