開發球面全景圖立體成像法以測量樹木之距離及直徑

Abstract

這項研究是用球形全景圖的立體成像去測量林分下的樹木距離和直徑，採用兩張用已知相對位移及旋轉的全景圖，在人工林中隨機建立了23個樣區，樣樹選擇具有小的底面積因子（BAF）作為胸徑和水平距離測量的直徑，使用較大的BAF對上枝直徑做再取樣分析，照片中有被遮擋的樣木不會採用，研究結果發現精度隨著與樣區中心的距離增加而下降，距離樣區中心10公尺範圍內的樹木距離可以非常精確地進行估計，對於樹木的胸高直徑來說，相機與樹木距離5公尺以內或者樹木胸高直徑不足20公分均表現出更高的精確度，由於地形和設備架設上的操作，保持全景圖之間所需的垂直位移是個困難的議題且可能是造成誤差的原因之一，為了使測量得到更好的結果，應用對極幾何的理論來找全景圖之間的關係，進一步的減小設備架設或攝影上的不確定性。在良好控制變因的區域中進行對極幾何的初步測試和理解，隨著現代科技以及攝影設備的不斷進步，拍攝球形全景圖變得更加容易，這項研究也許可以使公民科學以及森林資源的監測系統更加進步茁壯。

This study is the first to adapt stereoscopic imaging to measure tree distance and diameter under forest canopy with spherical panoramas. The proposed stereoscopy method requires two spherical panoramas taken at a known distance directly on top of each other. The geometry was comprehensively laid out. Twenty-three plots were randomly established in planted forests. Sample trees were selected with a small basal area factor (BAF) for diameter at breast height and horizontal distance measurement. A larger BAF was used to subsample trees for upper stem diameters. Only fully visible trees were used in the analysis. Results showed that precision decreased with increasing distance from a plot center. Tree distance could be accurately and precisely estimated within 10 m from a plot center. For tree diameters, better accuracy and precision was achieved either a tree was within 5 m away or its diameters were less than 20 cm. Maintaining the required vertical displacement between panoramas was challenging in the field due to difficult terrain and heavy equipment, which is a likely source of error. To relax the assumption of direct vertical displacement, theory of epipolar geometry was explored for the potential of identifying and correcting the degree of misalignment due to errors in the field. A preliminary test and understanding of epipolar geometry was carried out in an open area with targets of known sizes. With continued advancement in photography equipment, shooting spherical panorama becomes easier. Our proposed method could potentially lead to a citizen science based forest resources and monitoring program.