In order to improve the accuracy and authenticity for the capsule endoscopic examination,we present the 3D recovery technique for gastrointestinal tract based on the sequential endoscopic images. First,the corresponding feature points are extracted from the two sequential images by SIFT method,and their 2D coordinates on the imaging plane are acquired. Then the position and orientation shifts are calculated by the 8-point algorithm, which are related to the rotation matrix and transition vector. Further,the 3D relative and global coordinates are computed for all the feature points. Finally,the 3D feature points are gridded by Delaunay triangular subdivision algorithm and the 3D recovery is realized for the real scene. The experimental results show that the depth error is smaller than 1mm in case that the distance is 100 mm from camera to the tested point, and smaller than 3 mm the distance 250 mm. These results validate the feasibility of the proposed method.%为了改进胶囊内窥镜观测的准确性和真实性,提出了基于胶囊内窥镜序列图像的胃肠道三维重建的方法.首先利用SIFT算法提取前后两幅序列图像中尽可能多的对应特征点;计算获取各特征点在成像面上的二维坐标;进一步利用8点算法计算胶囊内镜运动变化的旋转矩阵和平移矢量.进而计算得到每个特征点的相对三维坐标和世界三维坐标;然后,采用Delaunay三角剖分算法对各三维点进行网格化,并完成场景的三维重建.实验表明相机与被测点距离在100 mm之内时,得到的深度误差小于1 mm;距离250 mm内时,相对误差在3％之内.说明所提出的算法是可行的.
展开▼
