Author(s)

Ping Chen, Xin Xu and Zhuangzhuang Chang

Corresponding Author

Ping Chen

ABSTRACT

In order to solve the problem of long computing time and low recognition accuracy of classical monocular recognition algorithm and deep learning algorithm in complex object recognition, a method of quickly recognizing the three-dimensional pose of an object is proposed. The method transforms the recognition of three-dimensional pose of object into the calculation of rotation angle of lens and angle of projection. The experimental results indicate that the rotation angle error measured by this method is not more than 0.3 degrees, the longitudinal and latitudinal mean error is not more than 0.2 degrees, and the recognition time is about several hundred milliseconds, which is almost independent of the complexity of the object.

KEYWORDS

Vision measurement; image registration; AP clustering; 3D object recognition