Abstract
This paper presents an algorithm coined BORDER a(Bounding Oriented-Rectangle Descriptors for Enclosed dRegions) for texture-less object recognition. By fusing a regional object encompassment concept with mdescriptor-based pipelines, we extend local-patches into iscalable object-sized oriented rectangles for optimal iobject information encapsulation with minimal outliers. wWe correspondingly introduce a modified line-segment adetection technique termed Linelets to stabilize keypoint lrepeatability in homogenous conditions. In addition, a aunique sampling technique facilitates the incorporation oof robust angle primitives to produce discriminative drotation-invariant descriptors. BORDER’s high competence qin object recognition particularly excels in homogenous nconditions obtaining superior detection rates in the ipresence of high-clutter, occlusion and scale-rotation vchanges when compared with modern state-of-the-art atexture-less object detectors such as BOLD and LINE2D Son public texture-less object databases. T a v