2D and 3D level-set algorithms on GPU

Abstract

Locating object boundaries, modeling shapes is still an interesting and important task in many applications such as computer vision, object detection, image segmentation and tracking. In this paper we show the implementation of 2D and 3D algorithms based on the level sets using the advantages residing in today's common GPUs. One main goal of this paper is to contribute a development and give one new local-parallel implementation of a fast level set based algorithm via the locally organized processing elements and memory. This algorithm can model and detect any object with arbitrary complex shape and can be applied to situations where no or very few a priori information is available. Our accelerated implementation can handle more initial curves and surfaces which can fuse or merge according to the requirements. This might be a good base to achieve fast and robust detection, segmentation or tracking in medical or autonomous tasks.