基于视觉1-Point RANSAC EKF-SLAM的大肠微型机器人内镜定位

发布时间：2018-04-14 11:19

本文选题：医学机器人 + 同时定位与建图(SLAM)　；参考：《中国科学技术大学》2017年硕士论文

【摘要】：无线胶囊内窥镜(WCE)技术虽然获得了患者的信赖,但其定位和控制问题仍然是研究人员面临的重大挑战。作为WCE检查中的关键部分,医生需要知道其确切的位置,以便在视频源检测中找到胃肠道疾病位置。要找到WCE的位置,需要一张人体内部地图。然而,由于胃肠道(GI)形状的复杂和RF信号在异质体组织中扩散方式的不同,正确地建图并对人体内的定位算法进行验证较为困难。本文验证了基于运动到结构(Structure from Motion,SfM)的扩展卡尔曼滤波框架的人体大肠内部的运动跟踪同步定位和建图技术,SfM算法对容纳来自输入图像唯一序列的所有可访问的先验概率数据源进行鲁棒估计,并且使用1-Point RANSAC将样本大小减小到1,以增强大肠内WCE的定位准确性和重建WCE运动轨迹。在此方式下,胃肠道疾病的位置可以在人体内地图上准确看到,从而促进治疗的后续过程。该方法将WCE两个源数据融合,由带式摄像机捕获序列图像,并变换WCE发射的RF信号,通过分析序列图像间特征点(FPs)位移来估计内窥镜胶囊的速度和方向,然后利用扩展卡尔曼滤波将运动信息与RF测量结合,以平滑定位结果并产生该胶囊运动轨迹。在仿真环境和恒速模型下,使用建立的大肠消化系统通道特征的虚拟试验台验证了所提出的运动跟踪算法的性能。实验结果表明,1-Point RANSAC EKF-SLAM定位算法可以给出WCE更准确的运动跟踪,并重建了 3D地图,实时显示了高典型的平均绝对误差约为0.0161的异常值,在利用卡尔曼滤波将EKF-SLAM与RF定位结合后成为0.0098363,距离误差约为1.1029。工作性能测量和与现有成果的比较描述了所提出的方法的鲁棒性和适用性。
[Abstract]:Although wireless capsule endoscope (WCE) technique is trusted by patients, its localization and control are still a major challenge for researchers.As a key part of WCE, doctors need to know its exact location in order to find the location of gastrointestinal diseases in video source detection.To find the location of WCE, you need a map of the body's interior.However, due to the complexity of the GI shape and the diffusivity of RF signals in heterogeneous tissues, it is difficult to build the map correctly and verify the localization algorithm in human body.This paper verifies the motion tracking synchronous location and mapping technology in human large intestine based on the extended Kalman filter framework based on moving to structure structure from motion (SfM). The SfM algorithm can be used to accommodate all accessible preemptions from unique sequences of input images.A probabilistic data source is used for robust estimation.1-Point RANSAC was used to reduce the sample size to 1 in order to enhance the localization accuracy of WCE in large intestine and reconstruct the WCE motion track.In this way, the location of gastrointestinal diseases can be accurately seen on the human body map, thus facilitating the treatment process.In this method, the two sources of WCE data are fused, the sequence images are captured by a band camera, the RF signals transmitted by WCE are transformed, and the velocity and direction of the endoscope capsule are estimated by analyzing the displacement of the feature points between the sequence images.Then the extended Kalman filter is used to combine the motion information with the RF measurement to smooth the positioning results and generate the capsule motion trajectory.Under the simulation environment and the constant speed model, the performance of the proposed motion tracking algorithm is verified by using a virtual test-bed based on the characteristics of the digestive system of the large intestine.The experimental results show that the 1-Point RANSAC EKF-SLAM localization algorithm can provide more accurate motion tracking for WCE, and reconstruct 3D map, showing a high typical outlier value of about 0.0161 in absolute error in real time.The EKF-SLAM and RF positioning are combined with Kalman filter to form 0.0098363, and the range error is about 1.1029.Performance measurement and comparison with existing results describe the robustness and applicability of the proposed method.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】