建物モデルの属性データ生成における近接写真測量を用いた建物開口部の抽出

Abstract

航空機等の上空からの計測データを用いた建物モデリングは効率的である反面，建物側面の属性データが得られず，精緻なモデリングのためには地上からの計測が不可欠である．本論文では，3次元建物モデルにおける，近接写真測量を用いた効率的な建物開口部データ生成手法を提案する．まずscale invariant feature transform (SIFT)を用いたパスポイント選定の高速化と誤抽出の低減のために，重複撮影された画像間のずれ幅を利用して対応点探索の範囲に制限を設ける．次に，建物開口部領域内部に手動で与えた1点の位置を手掛かりに開口部の輪郭を推定する．最後に，単写真におけるモデル座標計算を通じて開口部端点の3次元座標や開口部面積を推定する．京都市東山区で計測した写真群に適用し検証した結果，パスポイントの選定時間を大幅に短縮し，開口部の3次元座標のRMSEは約10cmで，防災分野で利用できる妥当な結果が得られた．

While modeling using the data measured by airborne is effective, it cannot provide the attribute of building side and more detailed modeling requires ground measurement. This paper proposes a methodology to effectively generate building opening data using close-range photogrammetry for three-dimensional (3D) building modeling. First, to reduce computation time, it limits an area to select passpoints between a pair of images, that are obtained by using scales invariant feature transform (SIFT). Then, it estimates boundaries of building opening by referring to an intial point in a region given by manual selection. Finally, it calculates 3D coordinate values of vertices and area of the openings through an equation in photogrammetry. As a result of validation using the data measured in Higashiyama ward, Kyoto, the proposed methodology succeeded in drastically reducing the computation time for passpoint selection, and root mean squares of errors for 3D coordinate values of the estimated vertices was approximately 10 cm. It was found that the proposed methodology can generate the building opening data whose accuracy is acceptable in the field of disaster prevention.