Evaluation of Interest Point Detectors and Feature Descriptors for Visual SLAM

Abstract

ln this paper we present comparison of feature detectors and descriptors in the visual feature-based simultaneous localization and mapping (SLAM) context. Feature mraction concept is widely used in computer vision and imaee processing algori1hms which are SLAM, panorama stitching, object detection, etc. Since features are used as the starting point and main primitives for subsequent algorithms, the overall algori1hm will often only be as good as its feature detector. Identifying of detected features is provided with the help of descriptor that distinguish it from the rest features. In tum, descriptor should provide invariance when finding the matches between the specific points relative to the image transformation. In this study, we evaluate the performance of well-known detectors and descriptors under the effects of JPEG compression, mom and rotation, blur, viewpoint and illumination variation. Performance parameters of the descriptors have investigated in terms of precision and recall values.

Keywords

• computer vision
• image processing
• feature detection
• simultaneous localization and mapping