An RGB-T Object Tracking Method for Solving Camera Motion Based on Correlation Filter

The RGB-T tracking based on correlation filter frame is widely studied because of its high efficiency in most complex scenes. However, the performance of these trackers is limited when facing some specific challenges, such as camera motion and background clutter. This paper focuses on how to solve the camera motion in the framework of correlation filter. First, given the input infrared and RGB images, we extract different features and use multi-expert systems to select the experts, and then conduct decision fusion tracking. Secondly, we first design a feature matching algorithm to locate the target that shows excellent performance. Comprehensive experimental results show that the proposed tracker has better performance in both accuracy and robustness. Our results on VOT-RGBT2019 dataset also demonstrate that it solves the common camera motion challenges in RGB-T tracking.

I. Introduction

As a fundamental problem in computer vision, object tracking has been widely used in various fields, such as intelligent monitoring [1], intelligent transportation and military fields [2]. Intelligent monitoring achieves the function of automatic alarm and monitoring by identifying and tracking the target in the video, and further realizes other advanced functions in the application level. The thermal infrared tracking is insensitive to lightning conditions and have a strong ability to penetrate haze and smog. RGB tracking possess plenty of fine-grained features, like colors and textures. Therefore, it is a good solution to choose RGB-T tracking. In certain situations such as low resolution, thermal crossover and low light as shown in Fig. 1, visible images can provide texture information to discriminate target which is challenging to infrared images.