Multi-feature Based Estimation of Microstructure Position and Orientation

Abstract

Various micromanipulations require the assessment of the target's position and orientation (PO) under microscopic vision. Here a universal method based on multi-feature fusion and discrimination correlation filtering (DCF) is suggested to improve the accuracy of PO estimation. A continuous convolution operator is first introduced to achieve multi-feature fusion and sub-pixel localization for high-precision location estimation. Then, by decoupling the translation and rotation with Fourier-Mellin transformation, the orientation information is expressed using rotation features retrieved. Finally, the mutual correction mechanism is used to merge position estimation and orientation estimation into a framework. In experiments, 4 pixels of position estimation error and 0.3 degrees of orientation estimation error were achieved.