A Robust Optical Flow Method Based on Deep Learning

Target tracking is an active research area in computer vision, in which optical flow estimation has been widely used due to its computational efficiency, simplicity and applicability. However, the traditional optical flow method relies on the assumptions of locally constant optical flow and luminance uniformity, which may not be fully valid in real dynamic lighting scenes, limiting its scalability. Therefore, in this paper, we use a deep convolutional network (CNN) to extract feature information from images, construct the learned feature map into a feature pyramid to enhance the adaptability to objects of different sizes or motions, and integrate it into the LK optical flow algorithm, so that the Hybrid optical flow method can break through the limitation of the single image luminance information. In order to verify the effectiveness of our method, we compare the proposed algorithm with the traditional optical flow method, and the simulation results demonstrate that our method shows better robustness and estimation accuracy in complex scenes such as dynamic lighting.