以無人載具空拍影像進行校正及分類之研究－以筏子溪為例

Abstract

ABSTRACT RPV(Remotely Piloted Vehicle)has been widely used in different mission, puck as in recent years.(surveying on the forestry monitor, researching agriculture, fishery and farming, monitoring on environmental protection and disaster prevention like floods or debris flows)However, RPV is restricted to congenital conditions, such as inaccuracy of RPV, and acquiring data is used for differentiation only. In order to increase the accuracy of RPV image , it is necessary for use to use GCPs to process correct necessary. But it is bad of using too much GCP for correcting. This study compares mosaicking photos with periods of increasing practicability. By way of grasping probably problems in mosaicking photos, and increasing precision by GPS. This study uses Erdas imagine 8.5 polynomial to understand effects of corrections by GCP and geometric distribution. In the end, we proceed the images classification,s study in geometric corrections. We extract the region of being corrected and then compare it with data by measuring. It also shows that 1st-order Polynomial GCP numbers in four or five is more perfect, and the 2nd-order Polynomial GCP numbers in seven or eight is more perfect. Both of the precisions of plane surface coordinate are also increased. Bilinear interpolation is the better photo,s correction in land uses classification.

中文摘要 無人載具直昇機近年來廣泛被應用於森林監測、農漁牧之調查、環保監控、坡地防災監測上。然因基於從事空拍攝影作業受限於平台的姿態誤差尚無法避免…等先天條件所限，以致該影像資料僅能用於判釋，必須以地面控制點進行地面控制點Ground control point，幾何校正以提高影像之價值，方能將該影像資料做充分之使用，然採用GCP之數量過多反會造成顧此失彼之缺失。 為提升無人載具直昇機航拍影像之實用性，本研究以前後兩期之空拍鑲嵌影像進行比較，藉以了解空拍影像鑲嵌之可能問題，並藉由全球衛星系統（GPS）輔助GCP 座標資料的獲得，以提高座標的精度，並使用Erdas公司生產之 Imagine8.5 影像處理軟體中之多項式轉換法（Polynomial） 進行GCP 數量及分布形狀對幾何校正影響之研究。最後對雙線性內插法（Bi-linear BL）、立方迴旋法（Cubic- convolution, CC）及最鄰近像元法（Nearest-neighbor, NN）三種幾何校正內差法進行影像分類之研究，將校正完成之影像資料萃取出欲分類之區域，與現地資料進行比對。經研究得知一次轉換式所採用之GCP數量以4至5點為佳，二次轉換式所採用之GCP數量則以7、8點為佳，不但總RMS（均方根）值會明顯降低，平面座標之精度也相對提高；在土地利用分類上具有較高利用價值的影像校正方法為雙線性內插法（BL）。