基于单目视觉的手术器械位姿估计模型研究

Research on the pose estimation model of surgical instruments based on monocular vision

手术器械管理是整个手术过程中重要的一环,科学高效的管理手术器械对提高医疗质量具有重要意义。为有效解决人工传递和清点手术器械效率低、易出错的情况,提出了一种基于单目视觉的手术器械位姿估计模型。首先将手术器械数据集图像进行Gamma适应校正,其次把校正后的数据集放入YOLOV5进行模型权重训练与识别验证,接着将识别所得的单枚手术器械图像进行局部增强与主体提取,最后把处理结果接入到位姿估计模块中,得到手术器械的中心点与4D位姿。实验结果表明,手术器械识别精度可达到89.4%,平均平移位姿误差为3.482 mm,平均旋转角度误差为2.048度,并且在不同分辨率和灰度值下均有良好的表现,验证了该算法具有良好的精度和鲁棒性。

Surgical instrument management is an important part of the entire surgical process.Scientific and efficient management of surgical instruments is of great significance to improving the quality of medical care.In order to effectively solve the inefficiency and error-prone situation of manual delivery and inventory of surgical instruments,a pose estimation model of surgical instruments based on monocular vision was proposed.First,the images of the surgical instrument data set are subjected to Gamma adaptation correction,and then the corrected data set is put into YOLOV5 for model weight training and identification verification.After that,the identified single surgical instrument image is subjected to local enhancement and subject extraction.Finally,the processing result is connected to the pose estimation module to obtain the center point and 4D pose of the surgical instrument.The experimental results show that the recognition accuracy of surgical instruments can reach 89.4%,the average translation orientation error is 3.482 mm,the average rotation angle error is 2.048 degrees,and the algorithm has good performance under different resolutions and gray values,which verifies the algorithm has good accuracy and robustness.