计算机辅助设计与图形学学报
計算機輔助設計與圖形學學報
계산궤보조설계여도형학학보
JOURNAL OF COMPUTER-AIDED DESIGN & COMPUTER GRAPHICS
2015年
5期
824-831
,共8页
摄像机标定%摄像机模型%镜头畸变%模型选择
攝像機標定%攝像機模型%鏡頭畸變%模型選擇
섭상궤표정%섭상궤모형%경두기변%모형선택
camera calibration%camera model%lens distortion%model selection
合理选择摄像机模型是摄像机标定的首要任务, 而传统的模型选择方法主要采用单一的重建精度指标来评判模型的合理性, 存在明显的局限性. 针对摄像机模型的比较和选择问题, 分析和定义了模型的重建精度与泛化性能、模型估计的稳定性和标定参数的不确定性 3 个性能指标. 重建精度与泛化性能指标不仅度量了模型对标定控制点范围内数据的重建误差大小, 而且还反映了模型对于控制点范围之外数据的适应程度; 稳定性指标描述了标定结果对于标定数据空间分布的依赖程度; 不确定性指标则反映了标定结果的偶然性. 这3个指标具有一定的互补性, 且稳定性和不确定性指标无需进行重建, 可以直接通过标定结果进行评判. 最后以3个不同特点的摄像机及8种不同类型的畸变模型为例, 通过实验验证了3个指标的合理性和可行性, 为合理选择摄像机模型提供了一种有效途径.
閤理選擇攝像機模型是攝像機標定的首要任務, 而傳統的模型選擇方法主要採用單一的重建精度指標來評判模型的閤理性, 存在明顯的跼限性. 針對攝像機模型的比較和選擇問題, 分析和定義瞭模型的重建精度與汎化性能、模型估計的穩定性和標定參數的不確定性 3 箇性能指標. 重建精度與汎化性能指標不僅度量瞭模型對標定控製點範圍內數據的重建誤差大小, 而且還反映瞭模型對于控製點範圍之外數據的適應程度; 穩定性指標描述瞭標定結果對于標定數據空間分佈的依賴程度; 不確定性指標則反映瞭標定結果的偶然性. 這3箇指標具有一定的互補性, 且穩定性和不確定性指標無需進行重建, 可以直接通過標定結果進行評判. 最後以3箇不同特點的攝像機及8種不同類型的畸變模型為例, 通過實驗驗證瞭3箇指標的閤理性和可行性, 為閤理選擇攝像機模型提供瞭一種有效途徑.
합리선택섭상궤모형시섭상궤표정적수요임무, 이전통적모형선택방법주요채용단일적중건정도지표래평판모형적합이성, 존재명현적국한성. 침대섭상궤모형적비교화선택문제, 분석화정의료모형적중건정도여범화성능、모형고계적은정성화표정삼수적불학정성 3 개성능지표. 중건정도여범화성능지표불부도량료모형대표정공제점범위내수거적중건오차대소, 이차환반영료모형대우공제점범위지외수거적괄응정도; 은정성지표묘술료표정결과대우표정수거공간분포적의뢰정도; 불학정성지표칙반영료표정결과적우연성. 저3개지표구유일정적호보성, 차은정성화불학정성지표무수진행중건, 가이직접통과표정결과진행평판. 최후이3개불동특점적섭상궤급8충불동류형적기변모형위례, 통과실험험증료3개지표적합이성화가행성, 위합리선택섭상궤모형제공료일충유효도경.
The first step in camera calibration is to make a right choice of the camera model. Traditional ways of the model selection share an obvious limitation by mainly relying on a single evaluation measure, i.e. the reconstruction precision, to estimate a model's soundness. In this paper, aiming at the comparison and selection of different cameral models, we define and analyze three evaluation measures. The first one is the 3D reconstruction precision and generalization capability, measuring not only the reconstruction error for calibrated points but also the adaptability on points deviated from the calibrated ones; the second one is the stability, describing the extent to which the calibration result depends on the spatial distribution of the points; the third one is the uncertainty which reflects the randomness of one trial of calibration. These three meas-ures are mutually complementary, and the stability and uncertainty can be directly estimated through the calibration result without reconstruction. Finally, intensive experiments on three cameras with eight distor-tion models are conducted to verify the validity and feasibility of three proposed evaluation measures, which offers a helpful guidance for the selection of camera models.