基於全景影片之室內定位及影像導航系統

當今室內導航的解決策略主要透過Beacon、Wi-Fi等作三角定位。本論文提出一基於視覺資訊的定位及導航系統，不必額外架設硬體設備，只需藉由室內空間中的環境影像建立參照影像資料庫作為後續檢索空間位置之用。建立任何定位及導航系統的必要流程之一為環境影像與實體空間位置的對應，本研究透過單眼視覺測程演算法簡化全景影像與空間平面圖疊合所需耗費的人力。本研究提出之定位系統建立在由全景影像所組成之參照資料庫之上，並且設定使用者透過行動裝置相機拍攝查詢影像來進行定位。本研究藉由MODS進行查詢影像與參照影像資料庫之比對，實驗場景台北車站捷運站B2場景中top 1的定位正確率在同時考慮人工標記答案與鄰近答案下可達61.1%。在導航系統中，本研究提出以環境全景影像在特定視角所擷取出之局部影像作為導航指示影像，利用導航指示影像引導使用者前往目的地。和使用藍點(blue dot)為導航人機介面的系統相比，真實環境畫面的反應可使使用者更確定自己是否走在正確的路上。

關鍵字

室內定位；室內導航；全景影像；單眼視覺測程；影像比對

並列摘要

Nowadays, indoor positioning and navigation problems are mainly solved by triangulation through beacon, Wi-Fi, etc. This paper proposes a vision-based indoor positioning and navigation system which only requires a reference image database constructed by images of indoor environments for localization, with no necessity to set up hardware devices. One essential step to build an indoor positioning and navigation system is to correspond panoramic images to real space. Through a monocular visual odometry algorithm, this study reduces labor needed for the correspondence. The positioning system in this study is based on a reference database constituted by panoramic images. Users are required to take a picture of the location of inquiry for localization. This study searches for the corresponding image in the reference database through MODS. Experiments of positioning at MRT Taipei Main Station at B2 show that the accuracy rate of top 1 can reach 61.1% if considering artificially marked answers and approximate answers at the same time. With respect to the navigation system, this study proposes using images extracted from panoramas at a specific angle to guide users to their destinations. Compared with the navigation system using the blue dot as the user-interface, reacting to real images of the surroundings allows users to be more certain whether they are on the right path.

並列關鍵字

Indoor positioning ； Indoor localization ； Indoor navigation ； Panoramic images ； Monocular visual odometry ； Image matching

參考文獻

1. Hu, F., Vision-based Assistive Indoor Localization. 2018.

Google Scholar

2. Liu, P., et al., A semi-supervised method for surveillance-based visual location recognition. IEEE transactions on cybernetics, 2017. 47(11): p. 3719-3732.