单幅结构场景图像的三维建模与仿真技术
|
摘要
针对当前单幅结构场景图像三维建模结果与实际情况偏差大、建模能耗高的问题,提出基于形状检索的单幅结构场景图像的三维建模方法。利用三组相互垂直的影灭点,实现像素相机标定,通过影灭线与圆环点像校正图像各平面度量。利用K均值聚类法对场景图像中所有局部特征聚类,组建一个视觉词汇库。依据视觉词汇库的构建,对图像特征向量和三维模型的不同方向特征之间距离进行计算,获取场景图像和三维模型间相似距离。在模型数据库检索出和物体最为相似的三维模型,将其作为备选的三维模型,并对其实施坐标与尺度变换操作。基于坐标与尺度变换中的参数,引入物体于图像场景中长和宽值、图像场景长和宽值等参数,完成单幅结构场景图像三维建模。实验结果表明,所提方法建模结果与实际情况拟合度平均为95.6%,建模能耗平均为60nJ/bit。上述方法建模精度高、能耗低,鲁棒性较强。
This article focuses on a three-dimensional modeling method for single structure scene image based on shape retrieval. At first, three sets of orthogonal vanishing points were used to realize the pixel camera calibration, and then each plane metric of image was corrected by vanishing line and the image of circular points. After that, K-means clustering method was used to cluster all local features in scene image, so as to form a visual vocabulary. Based on the visual vocabulary, the distance between feature vector of image and different directional features of three-dimensional model was calculated to get the similar distance between scene image and three-dimensional model. Moreover, three-dimensional model which was most similar to the object was searched from the model database, and then it was regarded as an alternative three-dimensional model. Coordinate and scale transformation operations were performed on it. Based on parameters in coordinate and scale transformation, the parameters such as the length and width of object in image scene, the length and width of image scene were introduced. Thus, the three-dimensional modeling for single structure scene image was completed. Simulation results show that the average fitting degree between the modeling result of proposed method and the actual situation is 95.6%. The average modeling energy consumption is 60 nJ/bit. This method has high modeling precision, low energy consumption and strong robustness.
This article focuses on a three-dimensional modeling method for single structure scene image based on shape retrieval. At first, three sets of orthogonal vanishing points were used to realize the pixel camera calibration, and then each plane metric of image was corrected by vanishing line and the image of circular points. After that, K-means clustering method was used to cluster all local features in scene image, so as to form a visual vocabulary. Based on the visual vocabulary, the distance between feature vector of image and different directional features of three-dimensional model was calculated to get the similar distance between scene image and three-dimensional model. Moreover, three-dimensional model which was most similar to the object was searched from the model database, and then it was regarded as an alternative three-dimensional model. Coordinate and scale transformation operations were performed on it. Based on parameters in coordinate and scale transformation, the parameters such as the length and width of object in image scene, the length and width of image scene were introduced. Thus, the three-dimensional modeling for single structure scene image was completed. Simulation results show that the average fitting degree between the modeling result of proposed method and the actual situation is 95.6%. The average modeling energy consumption is 60 nJ/bit. This method has high modeling precision, low energy consumption and strong robustness.
引文
[1] 吴晓军,徐广毅. 基于三维模型几何信息的纹理图像压缩[J]. 计算机辅助设计与图形学学报, 2016,28(3):471-479.
[2] 霍林生,张耀文,李宏男. 图像三维重建法在震损建筑实体建模中的应用研究[J]. 世界地震工程, 2017,33(2):113-118.
[3] 金磊,曾亚武,叶阳,等. 不规则颗粒及其集合体三维离散元建模方法的改进[J]. 岩土工程学报, 2017,39(7):1273-1281.
[4] 董建伟,李海滨,孔德明,等. 基于多视图立体视觉的煤场三维建模方法研究[J]. 燕山大学学报, 2016,40(2):136-141.
[5] 王思洁,方莉娜,陈崇成,等. 基于结构化场景的单幅图像建筑物三维重建[J]. 地球信息科学学报, 2016,18(8):1022-1029.
[6] 温维亮,赵春江,郭新宇,等. 基于t分布函数的玉米群体三维模型构建方法[J]. 农业工程学报, 2018,34(4):192-200.
[7] 贺庆,李兵邓燕青,等. 构建腰椎三维图像模型兔的特点分析[J]. 中国组织工程研究, 2017,21(12):1889-1893.
[8] 黄晓洲. 景物三维立体图像真实性建模仿真研究[J]. 计算机仿真, 2017,34(1): 249-252.
[9] 李占利,高天宇,李洪安,等. 基于牙冠与牙根特征的牙齿建模方法[J]. 图学学报, 2017,38(4):483-488.
[10] 贾笛,李彦林,何川,等. 点对点图像配准技术虚拟膝关节单髁置换术后三维模型的构建[J]. 中国运动医学杂志, 2017,36(9): 760-764.
[2] 霍林生,张耀文,李宏男. 图像三维重建法在震损建筑实体建模中的应用研究[J]. 世界地震工程, 2017,33(2):113-118.
[3] 金磊,曾亚武,叶阳,等. 不规则颗粒及其集合体三维离散元建模方法的改进[J]. 岩土工程学报, 2017,39(7):1273-1281.
[4] 董建伟,李海滨,孔德明,等. 基于多视图立体视觉的煤场三维建模方法研究[J]. 燕山大学学报, 2016,40(2):136-141.
[5] 王思洁,方莉娜,陈崇成,等. 基于结构化场景的单幅图像建筑物三维重建[J]. 地球信息科学学报, 2016,18(8):1022-1029.
[6] 温维亮,赵春江,郭新宇,等. 基于t分布函数的玉米群体三维模型构建方法[J]. 农业工程学报, 2018,34(4):192-200.
[7] 贺庆,李兵邓燕青,等. 构建腰椎三维图像模型兔的特点分析[J]. 中国组织工程研究, 2017,21(12):1889-1893.
[8] 黄晓洲. 景物三维立体图像真实性建模仿真研究[J]. 计算机仿真, 2017,34(1): 249-252.
[9] 李占利,高天宇,李洪安,等. 基于牙冠与牙根特征的牙齿建模方法[J]. 图学学报, 2017,38(4):483-488.
[10] 贾笛,李彦林,何川,等. 点对点图像配准技术虚拟膝关节单髁置换术后三维模型的构建[J]. 中国运动医学杂志, 2017,36(9): 760-764.