Stable Line and Circle Detection Method in Noise Image for Machine Vision

It is not easy to stably and precisely detect geometric features in noise image in machine vision as it could be used in calibration center detection for robot monocular vision or robotic grasping operation. A new stable line and circle detection method in noise image is proposed in this paper. First, a region of interest (ROI) of line or circle is detected using a template matching method of the stable feature. Then, a 2D feature detection problem can be converted to 1D by projection lines sampled vertical to the detection line, and the sampled pixels along the projection line are projected to the sample line. Second, the local noise scale and filtering variance are computed from pixels along the sample line, and the edge point is detected from the first order derivative and the outliers are rejected. Finally, the line or circle is fitted by using a linear least square method. To verify the proposed method, lots of synthetic images are tested and compared with the state-of-the-art method. The experimental results show that our method is better than the advanced commercial method.