Circle detection in images: A deep learning approach

Abstract

Circle detection is a well-known application in computer vision. The Hough transform has been the traditional algorithm applied to detect circular objects in images. In this paper, we are concerned with detecting circular objects for an autonomous underwater robotics application using computer vision. One of the task of the autonomous robot is to identify circular objects underwater. However, circular objects in the image appear in different sizes depending on the depth of the robot. Our experimental studies show that using conventional algorithms have limitations as the environmental light and reflections significantly affect the performance. Furthermore, the algorithm performance depends on parameters used in preprocessing and it has a considerable amount of computational complexity and large memory space requirements. There are various techniques introduced in computer vision literature aiming to reduce computational complexity or to improve its accuracy. Heuristic optimization techniques such as genetic algorithms and simulated annealing are used for large or noisy images for the advantage of less computation time compared to Hough transform. Nevertheless, deep learning algorithms recently become very popular in computer vision due to their remarkable performance in object detection. In this paper, we experimented detecting circular objects in images using latest deep learning algorithms and studied their performance. It showed significant advantage in underwater images compared to conventional algorithms.