基于双目视觉的医疗器械3D数据模型获取方法研究

The Research of the 3D Data Model Acquiring Method for Medical Instruments Based on Binocular Stereo Vision

目的:为了高效率、低创伤地完成外科手术,计算机辅助外科手术得到了广泛应用。医疗器械3D数据模型可有效地结合病人的影像数据,指导外科医生完成顺利外科手术,从而降低手术的风险,减少对病人的伤害。方法:基于双目视觉基本原理,首先对双目摄像机进行立体标定,利用标定后获得的摄像机参数对双目立体图像进行畸变矫正和立体校正,然后为了提高后续匹配精度,对医疗器械目标进行分割后再进行立体匹配,最后将获取的匹配视差信息重投影到3D空间,获得医疗机械的3D数据。结果:采用交互式分割方法对医疗器械目标进行先分割再立体匹配,能避免遮挡造成的误匹配,大大提高匹配精度。通过实验,将基于双目视觉获取的医疗器械3D数据与实际测量获得的数据进行比较后发现,本文方法获取的医疗器械3D数据精确度较高。结论:基于双目视觉理论可以高效率、高精度地进行医疗器械3D模型还原,其模型可用于模拟手术、手术导航等实际应用。

Objective： The technique of computer assisted surgery is widely applied in decades to accomplish surgeries with high efficioncy, small trauma and low risk. The 3D data of medical instruments, medical imaging data of patients as well as their fu- sions play a significant role in directing surgeons for successful operations, decreasing surgery risk and reducing damages for pa- tients, Methods： After calibrations for the stereo cameras, according to the distortion model, the method first makes use of the calibration results - cameras parameters to correct distorted images and to rectify stereo images. Then the method matches the loft and right images from two visions of stereo cameras to gain the disparities of the medical instruments segmented in the im- agos by interactive segmentation methods to improve the matching accuracy. At last, based on the principle of binocular stereo vision, th~ disparity data is re-projected into 3D space, i.e, 3D data for medical instruments is acquired. Results： The processing of segmentation for instruments greatly increases the matching precisions, and consequently so does the 3D data. By the compar- ison between the 3D data fi＇om the method and from the actual measurement results of the instruments, it shows that the obtained 3D data is with high accuracy. Conclusions： The theory of binocular stereo vision is effective to obtain 3D data for medical in- struments and the data is of high practical value in fields of virtual surgery simulation, images guided surgeries, etc.