Active Stereo Vision Based 3D Reconstruction for Image Guided Surgery

We propose the 3D sensing system using active stereo vision to get the 3D model of human body parts and surgical tools. This article presents the 3D scanner based on active stereo vision utilized for image guided surgery. The proposed system comprises of a pair of cameras (stereo camera) and the projector to record images of the target object placed at a specific distance. The proposed approach also includes the pre-calibration of the stereo camera for removing the distortion from the image data using non-linear method. We have performed the camera calibration of stereo camera using undistorted image data to estimate camera parameters using chessboard target. This research uses NCC (normalized cross correlation) based stereo matching and this matching is reduced to single dimension by utilizing epipolar geometry. After estimating the matched points, DLT (direct linear transformation) is applied to render the 3D point clouds. The rendered 3D point clouds were further filtered to remove the background objects. This refinement stage includes removal of outliers, suppression of noise and smoothing. We tested the 3D scanning system using artificial skull and a paper box object for the demonstration of image guided surgery. The three dimensional sensing system using stereo vision and structured light was tested using the objects, skull and paper box objects. The three dimensional model (3D model) shows that the features are preserved. Post processing based filtering results in the refinement of the point clouds. To evaluate the proposed system, we used the paper box object and the 3D scans were captured. We found that the errors in the height and breadth were 9.4 μm and 23 μm respectively. Our system currently does not target real time applications, but the parallel processing and utilization of GPUs may reduce the processing time to meet the needs of real time surgical intervention.