同步定位與建圖系統之設計與實現

Abstract

本研究提出一套改良式同步定位與建圖（Simultaneous Localization and Mapping, SLAM）系統架構，以擴展卡爾曼濾波器（ExtendedKalmanFilter,EKF）為核心，針對感測器量測數據進行雜訊抑制與數據關聯機制之設計。在感測數據前處理階段，引入基於密度之聚類演算法（Density-basedspatial clustering of applications with noise, DBSCAN），以降低感測器因不可抗力因素所產生之雜訊對後續地圖建構的影響。於數據關聯階段，本研究結合馬哈拉諾比斯距離（Mahalanobis distance）與歐幾里德距離（Euclidean distance）進行雙重判斷，達到同時考量估測不確定性與幾何距離關係，以提升地標匹配之穩定性。此外，於新地標生成流程中加入觀測次數門檻機制，僅當同一觀測連續出現達所設定之次數時，才將其納入地圖建構。實驗結果顯示，本研究所提出之方法可有效減少地圖中重複地標生成的情形，且能在雜訊環境下維持SLAM系統運作之強健性。

This study proposes an improved Simultaneous Localization and Mapping (SLAM)system architecture, centered on the Extended Kalman Filter (EKF), with a focus on noise suppression and data association mechanisms for sensor measurements. In the sensor data preprocessing stage, a Density-Based Spatial Clustering of Applications with Noise (DBSCAN)algorithm is introduced to mitigate the impact of sensor noise caused by uncontrol lable factors on subsequent map construction. During the data association stage, a dual criteria approach combining Mahalanobis distance and Euclidean distance is employed to simultaneously consider estimation uncertainty and geometric proximity, thereby enhancing the stability of landmark matching. Furthermore, a threshold mechanism based on the number of observations is incorporated into the new landmark generation process, such that only observations occurring consecutively up to a preset count are incorporated into the map. Experimental results demonstrate that the proposed method effectively reduces the occurrence of duplicate landmarks in the map and maintains the robustness of the SLAM system in noisy environments.