A gabor based fast interest point detector for image-based robot visual servo control

No matter the end-effector control of joint robots or the pose control of mobile robots, the image feature detection is the critical component of these robot technologies. Moreover, the interest point is the most widely used feedback signal in visual servo control, which performances are directly effected by the actual execution time of the feature detection methods. However, in the robot visual servo control the stringent requirement of real-time restricts the using of existing approaches, which complex computing processes take much more time. To get enough stable points with less computing time, this paper proposes a fast interest point detector based on the Hessian matrix and Gabor filters. Firstly, the approach obtains the positions of candidate interest points by using the scale adapted Hessian matrix. Secondly, a Gabor based scale selection is adopted to select the corresponding characteristic scales. Thirdly, an approximate algorithm combines the integral image and box type filters is used to accelerate the point detection process. The experiments show the proposed detector has a good performance and robustness under rotation and scale change with less executive time compared to some other detectors.