基於增強型ICP演算法之雲端多機器人建圖

Abstract

迭代最近點演算法是一種用來將兩群點集合對齊的方法，常使用於 2D 和 3D 幾何圖形的對齊。本論文使用機器人搭載雷射測距儀收集雷射資料，透過其掃描資訊作為點集合資訊，再使用迭代最近點演算法疊合，完成一個未知環境地圖的建立。但原始 ICP 演算法容易因為雜訊和離散點的關係，使得對齊效果不準確，尤其是在連續掃描的狀況下，對齊誤差越大，導致疊合精確度低和運算時間龐大。故本論文提出基於增強型ICP演算法實現於雲端運算架構，將原本序列進行所有雷射資料的過程，提出一分散式計算架構，使得所有雷射資料可以透過平行化的過程進行增強型ICP演算法，此演算法可大幅降低計算負擔並提升對齊的精確度，獲得更準確的環境地圖。接著將單機器人延伸至多機器人系統，將增強型ICP演算法結合加速強健特徵法，主要利用影像資訊判斷多機器人是否於相同的環境，在未滿足影像特徵門檻值前，單機器人將於各自的環境建立區域地圖，一旦滿足特徵匹配後，將多機器人的區域地圖資訊再以增強型ICP演算法疊合，進而增加建圖的效率。

The Iterative Closest Point algorithm (ICP) is to align the two point set, which is widely used in 2D and 3D geometric figure alignments. By using a Laser Range Finder (LRF) to collect data, it is capable of building a map in uncertain environments through a mobile robot. However, the original ICP algorithm can be easily affected by noise and discrete points, thus, increasing the error of alignment. In a row scanning by the laser date, the more data points are accumulated, the larger the errors of alignment become, which leads to an unpreferable map, and the process would be time consuming. This thesis proposes a cloud computing architecture based on Enhanced ICP, which parallelizes the original procedure. By doing so, it can significantly reduce the computational burden, improve the accuracy of alignment, and provide a more accurate environmental map. Furthermore, this thesis improves the use of a single robot to multi-robot system which combines the ICP alignment and Speeded Up Robust Features (SURF), resulting in increasing the efficiency of map building.