應用 SLAM 自走機器人自動建置場域 Google 街景地圖

Application of SLAM-based Autonomous Robot for Building Custom Field Street View in Google Maps

陳世軒(Shih-Hsuan Chen)

指導教授 ： 林達德

國立臺灣大學/生物資源暨農學院/生物產業機電工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201704016

摘要

本研究致力於發展一套自動化的街景拍攝系統。為達成此目的所開發之系統中搭載了許多不同的感測器，各個感測器各司其職，在機器人視覺領域中，即時定位與場景建構SLAM 是一項能夠針對未知環境探索時所採用的技術，透過重複的觀測以及控制的資訊，能夠估計出環境的資訊以及機器人的姿態。立體視覺是由兩個攝影機所組成的感測器，能夠透過影像處理以及立體視覺理論能夠計算出影像的空間資訊，缺點是容易受到環境狀態像是光源、成像品質、距離影響深度距離的計算，因此加入雷達感測器進行輔助，其透過都卜勒效應的原理來量測感測器與物體的距離及其他資訊，透過立體視覺與雷達感測器利用感測器融合的方式提高量測的可靠度，藉由感測器融合能夠提供系統進行障礙物偵測、障礙物追蹤，甚至提供系統利用 A* 演算法進行路徑規劃達到閃躲的能力。機器人在行走的狀況下容易因為路面品質不同、輪胎的打滑造成移動時會產生不同程度的誤差，本研究利用 SLAM 得到機器人的姿態提供給予 PID 控制器作為回授訊號，讓機器人在移動的過程中隨時修正路線。在室內及戶外的情況下，實驗得到直行之平均偏移量分別為 1.7 cm 及 2.0 cm。GPS (Global Positioning System) 是透過衛星定位出當下的位置，其因不同環境下有數公尺到數十公尺不等的誤差，本研究透過 SLAM 的方式搭配特徵點的選取藉此推算出拍攝當下實際之位置，與 GPS 所記錄資訊相較之下能夠大大提高精確度。本研究能夠於自動導航並拍攝街景資訊，將影像中的 GPS 資訊經過 SLAM 進行座標校正後，街景能夠上傳至公開的雲端平台上讓世界上各個角落的使用者能夠觀看到當地的場景資訊，或是儲存至私人的後端資料庫中方便使用者進行觀察以及記錄當時的環境資訊。

關鍵字

即時定位與地圖建構 ； 最近疊代點 ； 場景重建 ； 場景接合 ； 立體視覺 ； 雷達感測器 ； 感測器融合 ； 障礙物偵測 ； 避障系統

並列摘要

This study is dedicated to the development of an automated street shooting system. In order to achieve this purpose, the system is equipped with a number of different sensors with different usage. In the field of robot vision, SLAM (Simultaneous localization and mapping) is a technology be used to explore the unknown environment, can estimate the information of the environment and the robot's location through repetitive observation and controlled information. Stereo vision system, which is composed of two cameras, can calculate the spatial information of the image through image processing and stereo vision theory. The disadvantage is that the accuracy of the depth of the distance calculation is easily affected by the environment such as light source, image quality, and distance. Therefore, we add the radar, which measure the distance between the sensor and the object and other related information through the Doppler effect principle. To improve the overall reliability of the measurement through the stereo vision and radar sensor using the sensor fusion approach. The results obtained by the sensor fusion can provide obstacle detection, obstacle tracking, and even provide the system to do A * path planning algorithm to dodge them. The robot will have different degrees of error because of the different quality of the road or the tire slip caused during the movement. In order to fix this, this study use SLAM to provide the location of robot to gain PID control feedback signal, which can give the robot the ability to correct the route in the process of moving indoor and outdoor. The results of the experiment get the average offset of 0.084 m and 0.125m.GPS can locate the current position through satellite, with an error from a few meters to tens of meters due to the environment. This study combines SLAM with the selection of feature points to calculate the actual location of the shooting, and the accuracy compared with the information recorded by GPS is greatly improved. This study can automatically navigate and capture the scene information on the general road or in a specific field. After the GPS information of the image is corrected by SLAM, the 360 ° panoramic image can be uploaded to the open cloud platform to let the worldwide users view local scene information or store them in a private back-end database, which make it easier for users to view and record environmental information at the time.

並列關鍵字

Simultaneous localization and mapping (SLAM) ； iterative closest point (ICP) ； scene reconstruction ； stereo vision ； radar ； sensor fusion ； obstacle detection ； obstacle avoidance

參考文獻

李榕修。2015。基於SLAM自動導航及色彩點雲演算法之大尺度場景重建方法。碩士論文。臺北：國立臺灣大學生物產業機電工程學研究所。

翁立剛。2015。立體視覺與雷達感測器融合系統於車輛避障之應用。碩士論文。臺北：國立臺灣大學生物產業機電工程學研究所。

莊凱強。2013。基於立體視覺之即時障礙物追蹤與避障方法。碩士論文。臺北：國立臺灣大學生物產業機電工程學研究所。

賴宗誠。2012。應用多組雙眼攝影機系統進行車前三維環境模型重建。碩士論文。臺北：國立臺灣大學生物產業機電工程學研究所。

Lin, K., C. Chang, A. Dopfer and C. Wang. 2012. Mapping and Localization in 3D Environments Using a 2D Laser Scanner and a Stereo Camera. Journal of Information Science and Engineering. 28(1), 131-144.