Endoscopy is a medical procedure that provides a non-invasive view of a person's internal organs using a camera. Endoscopic videos can be processed in real time using computer vision methods, allowing surgeons to navigate more accurately during the procedure and make informed decisions based on the analysed data. Visual tracking plays an important role in capturing data from video, which is particularly challenging in endoscopy due to tissue deformations and occlusions. We tested the tracking accuracy of different algorithms implemented in the OpenCV library on the Cholec80 dataset. We developed an annotation software and used it to specify the optimal tracking of a selected tissue segment in each video. Finally, we compared the results of the trackers with the annotations to evaluate their performance and determine the best tracker.
|
