基于模板匹配的医用内窥镜影像目标识别算法

Target Recognition Algorithms for Medical Endoscope Image Based on Template Matching

为了减轻医务人员劳动强度,同时为患者提供有效的辅助诊断信息,将自动识别追踪技术应用于医用内窥镜中,以辅助外科医生诊断与治疗,并为后续持镜机器人及手术机器人研发打下基础。对比模板匹配和边缘检测匹配两种算法之后,发现模板匹配方法容易受到光照影响,将两者综合后的算法对光照和像素迁移有很强的抗干扰能力,适合于医用内窥镜光照条件不足的应用环境。此外,引入的CamShift算法以颜色特征作为第二匹配依据,可应对边缘不明显的情况。实验中分别对手术器械和胆囊进行模拟识别跟踪,实验结果表明,在该运动目标检测跟踪算法下,视频画面的帧速率稳定在30fps,不会出现卡顿情况,识别准确率达到了95%,并且在追踪过程中不会丢失目标。该算法原理简单、机理清晰,在实时性、鲁棒性等方面均可满足临床需求。

In order to reduce the work intensity of medical staff and provide effective auxiliary diagnostic information for patients,automatic recognition and tracking technology is applied to medical endoscopy to assist surgeons in diagnosis and treatment,which lays a foundation for the follow-up research and development of mirror-holding robots and surgical robots.After comparing the two algorithms of template matching and edge detection matching,we find that the template matching method is susceptible to illumination.The combined algorithm has strong anti-interference ability to illumination and pixel migration,which is suitable for the application environment of medical endoscopes with insufficient illumination conditions.In addition,the CamShift algorithm is introduced to deal with the situation that the edge is not obvious,taking the color feature as the second matching basis.Simulated recognition,tracking of surgical instruments and gallbladder were carried out in the experiment.Experiment result have shown that the moving target detection and tracking algorithm has a stable frame rate of 30 frames per second,and does not produce a pause.The recognition accuracy was 95%,and the target was not lost in the process of tracking,thus we can draw the conclusion that the principle of the algorithm is simple and the mechanism is clear.It can meet the clinical needs in real-time and robustness.