Abstract
Circular object detection is of great significance in wide areas, such as image processing, pattern recognition, computer vision, etc. Most present methods for circle detection are 2D-based algorithms, which means that only ideal 2D circles in the image plane can be detected. Circular objects in 3D physical world are usually ellipses in 2D images rather than standard 2D circles. Consequently, most common methods based on 2D images are unable to distinguish real space circles from ellipses. To solve the problem, in this paper, a 3D circular object detection method based on binocular stereo vision is proposed. It detects and fits ellipses/circles in stereo images firstly, then obtains sub-pixel-level disparity data between two images after stereo matching with the fitted mathematical model. Next, according to the binocular stereo vision model, the disparity data is reversely projected into 3D space. By the preset thresholds for parameters to evaluate the extent deviated from circular and coplanar objects, space circular objects can finally be detected based on those 3D data. Numerous experimental results demonstrate that the proposed method can detect circular objects successfully with high precision and efficiency.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call