A motion parameters estimating method based on deep learning for visual blurred object tracking

Abstract

Tracking the specific object in the blurred scenes is one of the challenging problems in computer vision and image processing. The accuracy and performance of trackers within the blur frames usually demonstrate a severe decrease. Accordingly, this problem needs to be corrected for better tracking results. Thus, this study seeks to present the best solution. To this end, a novel deep learning approach is proposed for object tracking in the presence of motion blur and fast motion. The hidden information in the blurring kernel is useful for tracking a specific blurred object through a series of consecutive frames. In this study, this matter is evaluated from a new perspective to solve the problems of blurred object tracking and objects with highly fast motions using a convolutional neural network (CNN) and a particle filter. Therefore, the proposed framework has two phases. First, the kernel leading to blurring is estimated by CNN, and then by a particle filter and the probability distribution of motion information that has been achieved by the kernel estimation the object is tracked. The results demonstrate that the suggested method can enhance the accuracy of tracking compared with the state-of-the-art, especially when the amount of blur is high.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.