A dense matching algorithm for complex street scenes reconstruction
Guofeng Tong, Xiaoping Sun, Xiangnan Meng · 2015
In this paper, we propose a novel dense matching algorithm based on mixing characteristics for complex street scenes reconstruction. Firstly, we choose adaptive weighted stereo matching algorithm for local matching. Then a three neighboring region partition algorithm based on SIFT feature matching is proposed, the algorithm obtains the target matching seeds through the three nearest neighbor SIFT feature points' matching relations that is in the same or similar depth plane with the original match points. Finally, we make use of polar constraint to improve algorithm accuracy. Experiment proves our algorithm can achieve high accuracy and quickness. Moreover, it has a good performance for complex street scenes reconstruction.