GPU Accelerated Cascade Hashing Image Matching for Large Scale 3D Reconstruction

TL;DR

This paper introduces a GPU-accelerated cascade hashing method for large-scale image matching in 3D reconstruction, significantly reducing computation time by optimizing data exchange, parallelizing algorithms, and incorporating epipolar constraints.

Contribution

It presents a novel GPU-based acceleration framework for cascade hashing in image matching, including data management strategies and geometric constraints, enabling faster large-scale 3D reconstruction.

Findings

Image matching is about 20 times faster than SiftGPU.

Nearly 100 times faster than CPU Cascade Hashing.

Hundreds of times faster than CPU Kd-Tree matching.

Abstract

Image feature point matching is a key step in Structure from Motion(SFM). However, it is becoming more and more time consuming because the number of images is getting larger and larger. In this paper, we proposed a GPU accelerated image matching method with improved Cascade Hashing. Firstly, we propose a Disk-Memory-GPU data exchange strategy and optimize the load order of data, so that the proposed method can deal with big data. Next, we parallelize the Cascade Hashing method on GPU. An improved parallel reduction and an improved parallel hashing ranking are proposed to fulfill this task. Finally, extensive experiments show that our image matching is about 20 times faster than SiftGPU on the same graphics card, nearly 100 times faster than the CPU CasHash method and hundreds of times faster than the CPU Kd-Tree based matching method. Further more, we introduce the epipolar constraint to the proposed method, and use the epipolar geometry to guide the feature matching procedure, which further reduces the matching cost.