Fuzzy-based feature matching for 3D reconstruction from multiple views

This paper presents new fuzzy based feature matching technique for 3D reconstruction from multiple views. Three fuzzy inputs are developed based on 1) traditional feature matching, 2) normalized RGB histogram, and 3) orientation similarity of the neighboring points. Fuzzy output represents a level of feature matching. The proposed technique is compared with the traditional correlation measure. Three 3D models are investigated, i.e., two chicken bones and a cup. The results showed that the proposed technique provided higher matching accuracy than the traditional one. Angle between two cameras can also be set wider. Therefore, the number of photos is greatly reduced.