Research On Endoscopic Visual SLAM For Minimally Invasive Surgery

Endoscopic vision simulation localization and mapping(SLAM)is to realize 3D reconstruction of soft tissue environment,simultaneously,estimate the endoscopic location respect to the 3D map based on the endoscopic real-time image in the dynamic minimally invasive surgery environment.Which means that the 3D spatial information of the soft tissue surface feature in the corresponding scence is abtained,and then the localzition of endoscope in the same spatial coordinates can be determined according to the relative positional relationship between the endoscope and the scence feature.The special compex environment of the minimally invasive surgery,such as intraoperative light unevenness,soft tissue bleeding,smoke of diagnosis and treatment,soft tissue deformation,soft tissue surface strong edge image feature loss and wet soft tissue surface height mirror reflection,etc.,has particularly high demands for endoscopic vision SLAM method on its real-time and robustness.In the face of above problems,this paper has carried out the study of endoscopic vision SLAM in minimally invasive surgery.The details are as follows:1.In this paper,the endoscopic vision SLAM framework based on probabilistic estimation is designed,which includes endoscopic and soft tissue motion modeling,endoscopic vision measurement model and extended Kalman filter algorithm.2.The soft tissue feature measurement is to obtain the actual measurement result of the scene feature on the two-dimensional image,including the soft tissue feature extraction and the feature matching.For the feature extraction of soft tissue,the ORB(Oriented Brief)feature extraction method is adopted to ensure the stability of the extracted soft tissue feature in the complex environment of minimally invasive surgery.For the better real-time,the current image is divided into the raster region,and the stable number of soft tissue features is extracted according to the disturbution parameter of feature in each raster region.For the feature matching of soft tissue,an active search method of soft tissue feature matching region is proposed,which avoids the global search,and greatly reduces the computational complexity of feature matching.The experimental results show that the improved feature measurement method is robust and widely applicable by comparing with a difficult soft tissue feature measurement method into the SLAM framework.3.An improved extended Kalman filter algorithm based on 1-point Random Sample Consensus(1-pointRANSAC)is proposed,and the feasibility of the improved extended Kalman filter is analyzed experimentally.Finally,this paper verifies the performance of the endoscopic visual SLAM method in deformed soft tissue by using the endoscopic sequence image data collected by DaVinci surgical robots in real minimally invasive surgery.The experimental results show that the method has good robustness on the dynamic interference of soft tissue deformation on positioning and three-dimensional reconstruction.The uncertainty elliptic region of the soft tissue feature measurement is finally converged to the estimated value,and with less soft tissue feature information,the method can achieve accurate estimation of the endoscope,and can deal with medical image sequences with 7Hz real-time processing speed.