Ellipse-CenterNet:Ellipse Detection in Complex Environments

The occluded ellipse in the complex background is hard to be detected and localized accurately. In this case, traditional methods for ellipse detection are vulnerable to both under-detection and over-detection, primarily caused by the confusion of the edges. This study proposes an anchor-free detector for ellipse detection, which, unlike existing CNN-based methods, can accurately and efficiently detect and localize ellipses in complex environments. Specifically, we implement two mechanisms of attention to enhance the model's ability to localize regions more precisely where ellipses are present and place greater emphasis on the useful channel information. Second, we modify the center bias loss for optimizing the ellipse regression accuracy. Results on two datasets show that our method achieves optimal performance.