空拍影像之全景圖建置與深度估計

Abstract

近年來，由於空拍機製造技術之進步，使空拍機之價格不如以往昂貴，讓一般民眾可以輕鬆負擔。空拍機的主要用於空中攝影與拍照，其拍攝之視角遠大於傳統手持攝影機所能達到的範圍。 本論文提出一套基於空拍影像之全景圖建置與深度估計方法，透過設計一演算法，分析影片畫面的動向以估算畫面移動速度，從而產生根據速率自動調適之空拍全景圖。此外，我們利用畫面移動速度與空拍機飛行速度之關係，進一步計算建築物深度，有了空拍全景圖的深度資訊，除了可改善全景圖的視覺品質外，還可用之模擬立體模型，幫助使用者了解立體全景狀況，增加導覽系統之豐富性。我們使用空拍機錄製的影片進行實驗，其結果說明我們所提出的空拍全景圖重建法能獲得令人滿意的視覺感受。

In recent years, due to the progress of manufacturing technology of drone, the price of drone is not as expensive as ever, so that the people can easily afford. Drones are mainly used for aerial photography and taking pictures. The shooting coverage is much larger than that the traditional handheld camera can achieve. In this thesis, we propose an aerial panorama generation system using a drone with general photography devices. The aerial panorama can provide visual summary of the aerial scenes along the top of buildings. Our method provides a novel algorithm which by analyzing screen movement between consecutive frames, is able to estimate depth information and generate panorama images properly without requiring a depth camera and a constant flying speed. With estimated depth information, the produced aerial panorama images have much better visual quality than those without using depth information. In addition, the depth of aerial panorama provided in our system helps users get more 3D information about the buildings. Experiments conducted on real videos captured by drone demonstrate that satisfactory results can be obtained by the proposed aerial panorama generation system.