GPU-accelerated feature tracking for 3D reconstruction

Abstract 3D reconstruction based on structure from motion is one of the most techniques to produce sparse point-cloud model and camera parameter. However, this technique heavily relies on feature tracking method to obtain feature correspondences, then resulting in a heavy computation burden. To speed up 3D reconstruction, in this paper, we design a novel GPU-accelerated feature tracking (GFT) method for large-scale structure from motion (SFM)-based 3D reconstruction. The proposed GFT method consists of GPU-based Gaussian of image (DOG) keypoint detector, RootSIFT descriptor extractor, k nearest matching, and outlier removing. Firstly, our GPU-based DOG implementation can detect thousands of keypoints in real-time, whose speed is 30 times faster than that of the CPU version. Secondly, our GPU-based RootSIFT descriptor can compute thousands of descriptors in real-time. Thirdly, our GPU-based descriptor matching is 10 times faster than that of the state-of-the-art methods. Finally, we conduct thorough experiments on different datasets to evaluate the proposed method. Experimental results demonstrate the effectiveness and efficiency of the proposed method.