Stereo infrared measurement of surgical target pose and continuous depth-based occlusion quantification for robot-assisted intraoperative imaging

Accurate measurement of surgical target pose and real-time quantification of camera-view occlusion are two fundamental challenges in autonomous intraoperative imaging systems. This paper presents measurement methods for both problems, validated on a robot-assisted imaging platform. For surgical target pose measurement, a stereo infrared measurement system is proposed that combines an actively emitting infrared sphere with temporal frame differencing, which suppresses ambient interference from surgical lights and metallic instruments. Target position is recovered via stereo triangulation of the RealSense D435 depth cameras, and orientation is simultaneously measured through an embedded IMU, yielding a position measurement error of 78.30 pixels (3.55% of image diagonal) and an orientation error of 3.05°, with a total processing time of 396.43 ms. For occlusion quantification, two continuous measurement metrics are derived from depth-camera point cloud projections: an inverse-distance-weighted occlusion value that quantifies obstruction severity, and a potential-field-based occlusion vector that encodes avoidance direction. These continuous metrics overcome the limitations of binary occlusion judgments in prior work, with a computation time of 3.45 ms per frame. To validate the utility of the proposed metrics, a Heuristic Next-Best-View (HNBV) algorithm employing a dual-key priority queue is developed, achieving an 87.1% success rate with a median search time of 234 ms across 172 test instances. Prototype experiments confirm successful occlusion avoidance under both single-sided and bilateral scenarios while maintaining continuous target lock.