視覺慣性里程計之一致性分析

Consensus Analysis for Visual-Inertial Odometry

陳萱(Hsuan Chen)

指導教授 ： 孫崇訓

淡江大學/工學院/機械與機電工程學系碩士班/碩士(2016年)

https://doi.org/10.6846/TKU.2016.01098

摘要

本論文探討的主要重點為整合視覺感測元件與慣性量測元件(Inertial Measurement Unit)之里程計(Visual-Inertial Odometry, VIO)及其一致性分析。以SURF(Speeded-Up Robust Feature)搜尋雙眼影像上之特徵點，將較為強健的特徵設為地標點，並利用立體幾何關係給定三維座標位置。透過移動前後地標點的位置，經由三點透視問題(perspective-3-point, P3P)及隨機取樣一致(random sample consensus, RANSAC)演算法求算出攝影機位置。搭配慣性量測元件(IMU)推算出旋轉角度，達成視覺慣性里程計的目的。此外，加入群聚分析剔除地標點位置的雜訊，並利用菁英制選取地標點，減少P3P運算時間。而後疊代修正攝影機與地標點位置完成一致性的分析。最後以實驗結果驗證所提出的視覺慣性里程計之一致性分析的定位準確性與即時性。

關鍵字

視覺慣性里程計 ； 群聚分析 ； 菁英制

並列摘要

This thesis focuses on integrating a vision system and an inertial measurement unit (IMU) into a visual-inertial odometry (VIO). Also, the consensus analysis is discussed in this thesis. First, the visual features in binocular images are detected by the speeded-up robust feature (SURF) algorithm. And, the robust features are defined as the landmarks and be located by the stereo geometry method. The camera can be located by the perspective-3-point (P3P) and random sample consensus (RANSAC) algorithms with the located landmarks. The IMU is used to measure the rotation of the camera. Then, the visual-inertial odometry is implemented. Besides, the clustering analysis for the position of landmarks removes the outliers. And, elitist strategy effectively reduces the computing demand for the P3P. The positions of the camera and landmarks are iteratively corrected to complete the consensus analysis. Finally, the experimental results demonstrate the accuracy and instantaneity of the proposed consensus analysis for the visual-inertial odometry.

並列關鍵字

visual-inertial odometry ； clustering analysis ； elitist strategy

參考文獻

[4] 林冠瑜，"使用低階攝影機實現機器人視覺式SLAM"，淡江大學機械與機電工程學系碩士論文，2012。

[11] 蕭恩哲，"基於視覺式里程計之二輪式機器人控制"，淡江大學機械與機電工程學系碩士論文，2013。

[2] H. Bay, A. Ess, T. Tuytelaars, and L. Van Gool, “Speeded-up robust features (SURF),” Computer Vision and Image Understanding, vol. 110, pp. 346-359, 2008.

[3] M. Agrawal, K. Konolige, and M. Rufus Blas, “CenSurE : Center Surround Extremas for Realtime Feature Detection and Matching,” Computer Visio ECCV2008, vol. 5305, pp. 102-115, 2008.

[5] M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography, ” Commun of the ACM, vol. 24, pp. 381-395, 1981.