Analysis and comparison of feature detection and matching algorithms for rovers vision navigation

In rovers' vision navigation, feature detection and matching algorithm is an important factor affecting navigation precision and speed. Harris, SIFT (Scale Invariant Feature Transform) and SURF (Speeded-Up Robust Features) are three commonly used feature detection and matching algorithms. Harris has been widely used in engineering application with high stability. SIFT is an efficient way to solve large scale changes of images in rovers' movement. It has high robustness and location precision. SURF is a speed-up algorithm of SIFT. In this paper, the cost of time, amount of features, amount of matching points and ratio of false match of these three methods mentioned above are studied and compared by experiments. Simulation shows that, Harris has the highest execution efficiency, while its false match rate is higher in large scale changes. SIFT can extract a great deal features and has the highest correct matching rate, but also has the longest computing time. SURF is much faster than SIFT, simultaneously having the same performance, which is the best method considering comprehensive performance.