Circle Area Detection Based on Convolutional Neural Networks

It is difficult for the traditional circle detection methods to locate the circle areas in an image. In this paper, we develop an end-to-end convolutional neural network (CNN) to detect circle areas for objects. In the CNN, an edge feature extraction module is designed to extract the edge feature. Then the edge and high-level features are fused in order to locate the circle areas accurately. A circle dataset with 4024 images is built for the CNN training. Several circles are embedded into the image, which will be used to train the CNN, in order to realize data augment simply and efficiently. The comparison experiments to the state-of-art CNN-based methods are well conducted. Our circle detector achieves the average precision 92.7% and recall rate 95.8% on datasets, and the time-cost is at the same level with others. The effectiveness of the proposed method is validated by a series of experiments.