An Improvement of SIFT Feature Points Detector for Localization of Mobile Robot

J. Liu, Q. Chen, X. Ma, Z. Sun, and W. Zhang (PRC)


Computer Vision; Features Detector; SIFT; SURF; APL


Visual odometry is a new method for mobile robot localization. Detecting features among pictures is one of the most important steps during it works. SIFT (Scale Invariant Feature Transform) features detector and descriptor are considered to be one of the best ones recently. However, the process costs so much time that it is hard to apply it into a real time localization of mobile robot. This paper presents a new algorithm: APL (Approximation of Laplacian) inspired by SURF (Speed Up Robust Features), which is a kind of improvement of SIFT. The APL algorithm rectifys some theoretical flaws of SURF and reaches higher speed than SIFT by means of simplifying the Laplacian of Gaussians during the computation of the first “octave”, which costs most of the time in SIFT. The paper presents experimental results on some pictures in a data set presented by Mikolajczyk. The results show that the new detector performs almost the same as SIFT on repeatability and costs less time, at the same time it costs almost the same time as SURF, and gain a higher repeatbility than SURF.

Important Links:

Go Back