New paper in IEEE RA-L on Visual servoing in the Frequency Domain

A paper accepted in IEEE Robotics and Automation Letters.

Eric Marchand. Visual Servoing in the Frequency DomainIEEE Robotics and Automation Letters,  2020. details Hal : Hyper Archive en ligne pdf

In this paper, we propose an original approach to extend direct visual servoing to the frequency domain. Whereas most of visual servoing approaches  relied on the geometric features, recent works have highlighted the importance of taking into account the photometric information of the entire images. This leads to direct visual servoing (DVS) approaches. In this paper we propose no longer to consider the  image itself in the spatial domain but its transformation in the frequency domain. The idea is to consider the Discrete Cosine Transform (DCT) which allows to represent the image  in the frequency domain in terms of a sum of cosine functions that oscillate at various frequencies. This leads to a new set of coordinates in a new precomputed orthogonal basis, the coefficients of the DCT. We propose to use these coefficients as the visual features that are then considered in a visual servoing control law.  We then exhibit the analytical formulation of the interaction matrix related to these coefficients. Experimental results validate our approach.

Comments are closed.