Edge Detection of Motion-Blurred Images Aided by Inertial Sensors

Abstract

Edge detection serves as the foundation for advanced image processing tasks. The accuracy of edge detection is significantly reduced when applied to motion-blurred images. In this paper, we propose an effective deblurring method adapted to the edge detection task, utilizing inertial sensors to aid in the deblurring process. To account for measurement errors of the inertial sensors, we transform them into blur kernel errors and apply a total-least-squares (TLS) based iterative optimization scheme to handle the image deblurring problem involving blur kernel errors, whose relating priors are learned by neural networks. We apply the Canny edge detection algorithm to each intermediate output of the iterative process and use all the edge detection results to calculate the network’s total loss function, enabling a closer coupling between the edge detection task and the deblurring iterative process. Based on the BSDS500 edge detection dataset and an independent inertial sensor dataset, we have constructed a synthetic dataset for training and evaluating the network. Results on the synthetic dataset indicate that, compared to existing representative deblurring methods, our proposed approach demonstrates higher accuracy and robustness in edge detection of motion-blurred images.