Abstract
BACKGROUND: Enhancing the safety of robot-assisted minimally invasive surgery (RAMIS) is critically dependent on improving the robot's spatial understanding of the surgical scene. However, the quality of laparoscopic images is often negatively affected by factors such as uneven lighting, blurred textures, and occlusions, all of which can interfere with the accurate acquisition of depth information. METHODS: To address these challenges, we develop a depth estimation and pose tracking method that incorporates a dual-stream Transformer stereo matching network and a vision-based tracking technique. RESULTS: Experimental results indicate that the proposed method can effectively maintain the boundary information of anatomical structures and demonstrate better performance in the robustness of laparoscope pose tracking. CONCLUSIONS: This paper presents a robotic-assisted minimally invasive surgery navigation framework that achieves accurate scene depth estimation and pose tracking, thereby enhancing the robot's spatial understanding of the surgical environment.