Map building from laser range sensor information using mixed data clustering and singular value decomposition in noisy environment
Ankit A. Ravankar, Yohei HOSHINO, Takanori Emaru, Yukinori Kobayashi · 2011
This paper presents the results of using mixed clustering (k-means and DBSCAN clustering) with singular value decomposition to build map in a noisy environment from laser range sensor information. Sensors are prone to errors, moreover, environmental and other factors may affect the sensor sensitivity. This may generate a lot of noise which must be removed before building the map. The study shows how density based clustering techniques can, without losing critical information, greatly reduce noise from sensor data. Further applying k-means clustering, the results with various cluster sizes are discussed. Singular value decomposition on the centroids obtained with the k-means clustering is applied to obtain straight regression lines. The %error in the generated maps were analyzed with different sizes of clusters. The experimental results confirmed that the proposed approach can generate accurate maps even in noisy environments.