Abstract

Incorporating smoothness constraints into feature matching is known to enable ultra-robust matching. However, such formulations are both complex and slow, making them unsuitable for video applications. This paper proposes GMS (Grid-based Motion Statistics), a simple means of encapsulating motion smoothness as the statistical likelihood of a certain number of matches in a region. GMS enables translation of high match numbers into high match quality. This provides a real-time, ultra-robust correspondence system. Evaluation on videos, with low textures, blurs and wide-baselines show GMS consistently out-performs other real-time matchers and can achieve parity with more sophisticated, much slower techniques.

Author

JiaWang Bian, Wen-Yan Lin, Yasuyuki Matsushita, Sai-Kit Yeung, Tan Dat Nguyen, Ming-Ming Cheng

Publication:

GMS: Grid-based Motion Statistics for Fast, Ultra-robust Feature Correspondence JiaWang Bian, Daniel Lin, Yasuyuki Matsushita, Sai-Kit Yeung, Tan Dat Nguyen, Ming-Ming Cheng IEEE CVPR, 2017 [Project Page] [pdf] [Code] [Video Demo]

Citation

If you use the code in your publication, please cite our paper.

Video Matching Demo

Environment:

The code can run on Windows, Linux, and Mac.

Requirement:

1.OpenCV 3.0 or later (for IO and ORB features, necessary)

2.cudafeatures2d module(for gpu nearest neighbor, optional)

How to run:

Image pair demo and video demo in demo.cpp.

(Note:	Please use gpu match when you run video demo.)

Tune Parameters:

In Main.cpp
	1.#define USE_GPU" will need gpu cudafeatures2d module for nearest neighbor match, 
		using cpu match by commenting it.

In gms_matcher.h
			
	2.	#define THRESH_FACTOR 6			// factor for calculating threshold
			The higher, the less matches, vice verse
			
	3. 	int GetInlierMask(vector<bool> &vbInliers, bool WithScale = false, bool WithRotation = false)
			You can open multi-scale and rotation if your image pair contains that.