Motion stabilisation for video sequences

Motion Stabilisation is the art of removing unwanted observable motion from a dynamic image sequence. Computer Vision techniques serve as a purely software based means of achieving motion stabilisation by applying a sequence of techniques including calculating an original global motion estimate using feature detection, feature tracking and outlier rejection. An update transformation is derived and applied to the original video sequence and a new stiller video is created. Minimally Invasive Heart surgery is a modern approach to heart surgery whereby the surgeon operates on the patient remotely using robotic arms to perform delicate operations within the patient. Benefits of such endoscopic surgery include less scarring, lower rates of wound infection and a faster recovery time. As patient demand for such surgery increases, the technique still poses problems for the surgeon such as having only an indirect view of the field of surgery via a small camera. When operating on the heart, traditionally a Heart Bypass machine is employed to still the heart to ease the job of the surgeon, however sometimes serious complications can arise as a result of using such machines. Beating Heart surgery requires the surgeon to operate on the heart whilst it is still beating which requires additional effort and concentration. Motion compensation of the beating heart has been investigated by several international research groups over recent years. In this project, we investigate the application of a recent 2D motion stabilisation method to the problem of motion compensation in minimally invasive beating heart surgery. The investigation served to assess how suitable this method is when applied to a 3D deforming surface such as the heart. For evaluation purposes, we apply the procedure to a video of a heart phantom as well as footage of a real moving heart and quantitatively and qualitatively measure the reduction in apparent motion in each video. The project found that 2D methods are capable of compensating non-linear 3D deforming motions to an extent. In our real heart video we found that our global motion estimate method was able to derive a decent global motion estimate despite the high level of specular reflection and pre-existing motion blur. Consequently, the method was able to compensate a substantial amount of respiratory motion in addition to a reduction in observable cardiac motion.