An improved method for calculating image similarity between virtual and real endoscopic images in tracking bronchoscope motion

摘要

This paper presents an improved method for calculating image-similarity in the camera motion estimation process required in augmented endoscopy. An image-registration technique is utilized here for camera motion tracking. In the previous method, image similarity is computed by summing gray-level differences up for all pixels of real and virtual endoscopic images. This method could not estimate the position and orientation of a real endoscope camera properly when image similarity does not change so much due to averaging of gray-level difference for the entire of the image. The proposed method divides the original image into a set of small sub blocks and selects only the sub blocks in which characteristic shapes are observed. Image similarity is calculated only in the selected sub blocks. We applied the proposed method to three pairs of X-ray CT images and real bronchoscopic video images. In the experimental results, the proposed method showed much improvement in continuous tracking performance. Over 900 consecutive frames were tracked correctly.