Deep homography estimation in dynamic surgical scenes for laparoscopic camera motion extraction

Current laparoscopic camera motion automation relies on rule-based approaches or only focuses on surgical tools. Imitation Learning (IL) methods could alleviate these shortcomings, but have so far been applied to oversimplified setups. Instead of extracting actions from oversimplified setups, in this work we introduce a method that allows to extract a laparoscope holder's actions from videos of laparoscopic interventions. We synthetically add camera motion to a newly acquired dataset of camera motion free da Vinci surgery image sequences through a novel homography generation algorithm. The synthetic camera motion serves as a supervisory signal for camera motion estimation that is invariant to object and tool motion. We perform an extensive evaluation of state-of-the-art (SOTA) Deep Neural Networks (DNNs) across multiple compute regimes, finding our method transfers from our camera motion free da Vinci surgery dataset to videos of laparoscopic interventions, outperforming classical homography estimation approaches in both, precision by 41 % , and runtime on a CPU by 43 % .

Region-Based Static Video Stitching for Reduction of Parallax Distortion

In this paper, we propose a semantic segmentation-based static video stitching method to reduce parallax and misalignment distortion for sports stadium scenes with dynamic foreground objects. First, video frame pairs for stitching are divided into segments of different classes through semantic segmentation. Region-based stitching is performed on matched segment pairs, assuming that segments of the same semantic class are on the same plane. Second, to prevent degradation of the stitching quality of plain or noisy videos, the homography for each matched segment pair is estimated using the temporally consistent feature points. Finally, the stitched video frame is synthesized by stacking the stitched matched segment pairs and the foreground segments to the reference frame plane by descending order of the area. The performance of the proposed method is evaluated by comparing the subjective quality, geometric distortion, and pixel distortion of video sequences stitched using the proposed and conventional methods. The proposed method is shown to reduce parallax and misalignment distortion in segments with plain texture or large parallax, and significantly improve geometric distortion and pixel distortion compared to conventional methods.

Multi-Frame Based Homography Estimation for Video Stitching in Static Camera Environments

In this paper, a multi-frame based homography estimation method is proposed for video stitching in static camera environments. A homography that is robust against spatio-temporally induced noise can be estimated by intervals, using feature points extracted during a predetermined time interval. The feature point with the largest blob response in each quantized location bin, a representative feature point, is used for matching a pair of video sequences. After matching representative feature points from each camera, the homography for the interval is estimated by random sample consensus (RANSAC) on the matched representative feature points, with their chances of being sampled proportional to their numbers of occurrences in the interval. The performance of the proposed method is compared with that of the per-frame method by investigating alignment distortion and stitching scores for daytime and noisy video sequence pairs. It is shown that alignment distortion in overlapping regions is reduced and the stitching score is improved by the proposed method. The proposed method can be used for panoramic video stitching with static video cameras and for panoramic image stitching with less alignment distortion.