Abstract:

To improve the reliability and real time of feature detection, a fast adaptive robust invariant scalable feature detector (FARISFD) is proposed. Firstly, the adaptive selection of octave is proposed to improve reliability of different images. Secondly, the creation of scale space based on intermediate layer is proposed to improve robustness and speed for scale space. Finally, the speed for computation of feature scores and sub-pixel refinement are improved by computing the feature scores via features from accelerated segment test (FAST) and simplification of traditional sub-pixel refinement methods. The experimental results of repeatability and time show that comparing with 5 widely available feature detectors, the FARISFD reveals stronger robustness, and the speed is faster.