以權重式重疊區域影像平面投射及動態規劃建立適用於視覺化定位系統之全景影像疊合演算法

Abstract

此篇論文最主要的討論內容是如何使用單個攝影裝置拍攝周圍360°環狀影像之後，把它們疊合成適合用於視覺化定位系統的全景環境地圖。疊合全景環境地圖的過程主要分為三個部份，即重疊區域估測、影像平面投射、和影像疊合。首先，我們先預估兩張影像之間的重疊區域，接著使用SURF特徵點偵測和比對來校正重疊區域的位置與範圍。透過SURF特徵點校正重疊區域位置，我們可以減低系統對攝影機架設和移動量估計的依靠性，使得系統更加能夠抵抗移動量計算的誤差。考慮到現實場景中很多時候會出現重複性或對稱性很高的場景，因此採用區域性SURF特徵點偵測以減低錯誤的特徵點比對。除此之外，我們採用了權重式重疊區域影像平面投射以減低影像平面投影之後所造成的誤差，同時也提高疊合全景地圖的成功率。疊合影像的時候，爲了避免幽靈效果，重疊區域的影像不會直接進行線性混合，而是使用動態規劃法來規劃出一條影像切割的最佳路徑，把影像切割之後再做疊合。本論文將針對此系統於室內和室外之全景地圖疊合效果進行定性和定量分析與討論。

This thesis presents a panorama stitching system using only one single capturing device to capture 360° scenery of the surrounding environment to create an environment map for the uses of vision localization. Basically, there are mainly three parts in the panorama stitching system, consisting of overlapping area estimation, image plane projection, and image stitching. Instead of assuming that the overlapping area is kept constant for every two adjacent images, the overlapping area is estimated and further refined by using matched SURF feature pairs. By doing this, the dependency of the system on capturing device setup and its motion model estimation can be reduced, thus resulting it to become a more robust system. To consider an environment with highly symmetric or repeated features, the regional SURF feature detection is applied to lower down the fault error during SURF feature matching. To reduce distortion within the resulting panorama image, the overlapping area weighted image plane projection is used to successfully create a panorama image. Furthermore, to prevent ghost effect during stitching panorama image, the dynamic programming algorithm is used to find an optimal path to stitch two adjacent images. The proposed algorithm has been tested at indoor and outdoor environments. Related qualitative and quantitative results will be analyzed in detail.