Automatic Appearance-Based Loop Detection from 3D Laser Data Using the Normal Distributions Transform

We propose a new approach to appearance-based loop detection for mobile robots, using 3D laser scans. Loop detection is an important problem in the SLAM domain, and, because it can be seen as the problem of recognizing previously visited places, it is an example of the data association problem. Without a flat floor assumption, 2D laser-based approaches are bound to fail in many cases. Two of the problems with 3D approaches that we address in this paper are how to handle the greatly increased amount of data and how to efficiently obtain invariance to 3D rotations. We present a compact representation of 3D point clouds that is still discriminative enough to detect loop closures without false positives (i.e., detecting loop closure where there is none). A low false positive rate is very important because wrong data association could have disastrous consequences in a SLAM algorithm. Our approach uses only the appearance of 3D point clouds to detect loops and requires no pose information. We exploit the NDT surface representation to create feature histograms based on surface orientation and smoothness. The surface shape histograms compress the input data by two to three orders of magnitude. Because of the high compression rate, the histograms can be matched efficiently to compare the appearance of two scans. Rotation invariance is achieved by aligning scans with respect to dominant surface orientations. We also propose to use expectation maximization to fit a Gamma mixture model to the output similarity measures in order to automatically determine the threshold that separates scans at loop closures from non-overlapping ones. We discuss the problem of determining ground truth in the context of loop detection and the difficulties in comparing the results of the few available methods based on range information. Furthermore, we present quantitative performance evaluations using three real-world data sets, one of which is highly self-similar, showing that the proposed method achieves high recall rates (percentage of correctly identified loop closures) at low false positive rates in environments with different characteristics.