SIFT and SURF performance evaluation and the effect of FREAK descriptor in the context of visual odometry for unmanned aerial vehicles

Abstract

Feature points detection and description play very important role in many of computer vision applications. Specifically in robot visual navigation systems (i.e. visual odometry or visual simultaneous localization and mapping), which need reliable high speed processing algorithms with low memory load. This paper presents a performance evaluation of the two robust feature detection/description algorithms (SIFT and SURF) with the effect of combining the FREAK descriptor. The performance of these algorithms was compared for the changes in noise, scale and rotation.