DENDT: 3D-NDT scan matching with Differential Evolution
In this paper, we introduce a fast and robust scan matching method that combines the Multi-Layered Normal Distributions Transform (ML-NDT) and a feature extraction algorithm into a single framework. This is achieved by first applying the conventional NDT generation process to the reference scan, and the plane segments are extracted with the help of Random Sample Consensus (RANSAC) algorithm for the input scan. Thus, the proposed method provides three significant advantages with respect to conventional methods. The first one is that the proposed method is more robust to outliers since it is based on the matching of certain geometric structures. The second one is that the registration step is much faster because the number of points to be matched is very less with respect to all scanned points. Therefore, this process can be considered as a special sampling strategy. Finally, it is showed that the extracted features can also be used in feature based probabilistic SLAM methods such as Kalman Filters, Information Filters, and Particle Filters after applying merging procedure. Since the plane segments are already registered, the data association problem can be easily solved even without any odometry measurement. This can be considered as the most powerful part of the algorithm because data association problem in three dimensions is quite difficult problem. As a result, on the one hand, it is obtained a robust and fast scan matching; on the other hand, it is possible to extend the method for feature extraction algorithm in SLAM problems with a little extra computation. The method is applied to real experimental data and the results are quite affirmative.