Gastrointestinal Endoscope Localization And Feature Map Reconstruction Based On Monocular SLAM

Gastrointestinal disease has a great influence on the patient's health, so early detection and treatment of these diseases is very significant. Currently, there exists two problems in clinical treatment using the endoscope. Firstly, the endoscope usually moves in human body, so it is hard to locate the endoscope. Secondly, we can hard to reconstruct the environment of the gastrointestinal. So, research on using monocular to implement the task of localization for the endoscope and reconstruction of the gastrointestinal environment is very helpful.Firstly, several feature extract methods is analyzed under the environment of human Gastrointestinal. we compare these algorithms in several ways such as the speed of feature extraction, the correct match rate under rotation and scale transformation to find a suitable method. ORB algorithms resulted a small rotation angle error, a high correct match rate, and a fast feature extraction speed.Based on gastrointestinal endoscopy measured data,we modify the posture adjustment and spatial position calculate method in the classical SLAM algorithm, introduced a local optimization algorithm based on the visual correlation diagram and a triangulation algorithm which geometric distance is least. In order to control the growth of key frames and map points, a key frame judgment strategy and a spatial map point delete strategy are designed. The feature map of human colon is reconstructed based on the video data, and the trajectory of the endoscope is estimated. Then the feasibility of the algorithm in the real intestinal environment is verified. To acquire the ground truth of location and mapping and evaluated the effect of frame rate and Keyframe number on the accuracy of the location and mapping, a simulated gastrointestinal system is built. Based on simulated gastrointestinal system, the gastrointestinal SLAM algorithm and the traditional visual odometry calculation method were compared. When endoscopic moved in line, the number of Keyframe storage of the proposed algorithm only one eighth that of the latter, and the location accuracy is improved by three times.