Feature Detector Using Adaptive Accelerated Segment Test

Abstract

Features from Accelerated Segment Test (FAST) based on the circular mask segment test is recognized as a superior feature detector in terms of speed. The radius of the circular mask and the arc length of the segment are the two geometric parameters that affect the speed and the quality of the detector. Binary Robust Invariant Scalable Key points (BRISK) applied FAST directly across all the layers of the scale-space representation of an image without considering the effect of these parameters. In this paper, we propose a new feature detector, Features from Adaptive Accelerated Segment Test (FAAST), to further reduce the computation time, and to improve the quality of the corner response of FAST. The standard FAST detector is designed for fixed scale of the image. To apply this detector across all the layers of the scale-space, the proposed detector applies different mask sizes with their corresponding arch lengths across different layers of the pyramid. Experimental results on benchmark datasets reveal that the proposed method produces significant improvements in repeatability score by generating more discriminating features that are close to the real corner location. The proposed detector also shows better speed than FAST detector.