3D pose estimation of needle based on single-view images

Abstract

Needle insertion surgery through accurate augmented reality (AR) is effective because it allows surgeons to see inside the body accurately without using other imaging techniques. There are various image-based needle estimation techniques, but it is difficult to estimate 3D pose of the needle based on a single image without any bulky attachments for markers. This study is about a highly accurate needle position automatic estimation technology using simple markers for needle AR navigation. This research points out the problems of existing algorithms that are not high in accuracy, and introduces improved algorithms that improves the pointed problems. In order to confirm the legitimacy of the modeling, the accuracy was calculated through real-world experiments. The average error was 1.76mm; compared to the existing single-camera image-based automatic estimation method, the average accuracy increased by 62%. By enabling needle real-time tracking based on this algorithm, it is expected to be of great help in enabling needle AR navigation to be used in actual medical environments.