Laparoscopic Scene Reconstruction Based on Multiscale Feature Patch Tracking Method

In minimally invasive laparoscopic surgery, doctors use laparoscopic images to guide surgical operations. During the operation, the visual field is narrow and the operation space is limited, which is not conducive to the observation and operation for doctors. Comprehensive intra-abdominal scene information exerts an enormous function on understanding anatomical structure for doctors, compensating errors in surgical navigation, and displaying in augmented reality surgery. In this paper, a laparoscopic reconstruction method based on multiscale feature patch tracking is proposed. The new method uses a multiscale Kernel Correlation Filter (KCF) to track feature patches in Simultaneous Localization and Mapping (SLAM) framework, which is called MKCF-SLAM. The proposed method is applied in stereoscopic laparoscopy, which can reconstruct laparoscopic scene under the conditions of less texture, high reflection, and target scale changes. The new method is validated on a public invivo data set. Compared with feature-based SLAM methods and our previous work, the proposed method can get better and more stable work in some special difficult environments. The results suggest that the proposed method can reconstruct the laparoscopic scene accurately and robustly.