摘要
由于单目手术视频的镜头参数未知,仅凭该单目图像难以准确估算内镜镜头下的手术器械的三维坐标.针对这一问题,提出一种手术器械相对工作空间的估计方法.首先,对用U-Net网络分割后的掩膜图像进行直线检测,对图像坐标系下的直线映射到参数空间上的点进行聚类,从而优化直线检测效果.然后,利用边缘线段的相对几何位置特征对因网络分割效果欠佳导致的边缘线段残缺进行补足,从而提取出与目标边缘高度拟合的完整线段.结合目标器械实际尺寸与图像尺寸间的比例关系,利用平行线在投影平面上的消隐点特性计算出目标器械目标点处的实际三维坐标.最后,进行测距实验和轨迹跟踪实验.实验结果表明:该方法较磁导航传感器准确度高、鲁棒性强,距离测量精度可达2.6581mm,轨迹拟合度达91.2%,满足实际情况下手术器械相对工作空间的估算要求.
It is difficult to accurately estimate the three-dimensional coordinates of surgical instruments under an en-doscopic lens based on the monocular image because the lens parameters of monocular surgical video are unknown.Consequently,a method for estimating the relative workspace of surgical instruments is proposed to address this dif-ficulty.First,the straight lines in the mask image segmented by U-Net were detected,and the points mapped from straight lines in the image coordinate to parameter space were grouped to maximize the effect of straight-line detec-tion.Second,to make up for the incomplete lines caused by poor network segmentation,the edge segment’s geomet-ric position feature was used by extracting the complete segment with a high degree of fit with the edge of the target.Furthermore,the actual 3D coordinate estimation of the target equipment was calculated using the proportional relationship between the actual and image size of the target instrument and the characteristics of the parallel line’s vanish-ing point.Finally,ranging and trajectory tracking experiments were conducted.The experimental results show that this method has higher accuracy and robustness than magnetic navigation sensors,and it achieves a distance measurement accuracy of 2.6581mm and a trajectory fitting degree reaching 91.2%,which meets the requirements of the relative workspace estimation of surgical instruments in the actual situation.
作者
苏赫
白雪
Su He;Bai Xue(School of Mechanical Engineering,Tianjin University,Tianjin 300354,China)
出处
《天津大学学报(自然科学与工程技术版)》
EI
CAS
CSCD
北大核心
2023年第8期785-795,共11页
Journal of Tianjin University:Science and Technology
基金
国家自然科学基金资助项目(52175028).