PCA-based line detection from range data for mapping and localization-aiding of UAVs

This paper presents an original technique for robust detection of line features from range data, which is also the core element of an algorithm conceived for mapping 2D environments. A new approach is also discussed to improve the accuracy of position and attitude estimates of the localization by feeding back angular information extracted from the detected edges in the updating map. The innovative aspects of the line detection algorithm regard the proposed hierarchical clusterization method for segmentation. Instead, line fitting is carried out by exploiting the Principal Component Analysis, unlike traditional techniques relying on Least Squares linear regression. Numerical simulations are purposely conceived to compare these approaches for line fitting. Results demonstrate the applicability of the proposed technique as it provides comparable performance in terms of computational load and accuracy compared to the least squares method. Also, performance of the overall line detection architecture, as well as of the solutions proposed for line-based mapping and localization-aiding is evaluated exploiting real range data acquired in indoor environments using an UTM-30LX-EW 2D LIDAR. This paper lies in the framework of autonomous navigation of unmanned vehicles moving in complex 2D areas, e.g. unexplored, full of obstacles, GPS-challenging or denied.