Learning Response-Consistent and Background-Suppressed Correlation Filters for Real-Time UAV Tracking

Abstract

With the advantages of discriminative correlation filter (DCF) in tracking accuracy and computational efficiency, the DCF-based methods have been widely used in the field of unmanned aerial vehicles (UAV) for target tracking. However, UAV tracking inevitably encounters various challenging scenarios, such as background clutter, similar target, partial/full occlusion, fast motion, etc. These challenges generally lead to multi-peak interferences in the response map that cause the target drift or even loss. To tackle this problem, a response-consistent and background-suppressed correlation filter is proposed for UAV tracking. First, a response-consistent module is developed, in which two response maps are generated by the filter and the features extracted from adjacent frames. Then, these two responses are kept to be consistent with the response from the previous frame. By utilizing the l2-norm constraint for consistency, this module not only can avoid sudden changes of the target response caused by background interferences but also enables the learned filter to preserve the discriminative ability of the previous filter. Second, a novel background-suppressed module is proposed, which makes the learned filter to be more aware of background information by using an attention mask matrix. With the introduction of this module into the DCF framework, the proposed method can further suppress the response interferences of distractors in the background. Finally, extensive comparative experiments have been conducted on three challenging UAV benchmarks, including UAV123@10fps, DTB70 and UAVDT. Experimental results have proved that our tracker has better tracking performance compared with 22 other state-of-the-art trackers. Moreover, our proposed tracker can run at ∼36 FPS on a single CPU for real-time UAV tracking.

Learning Background-Suppressed Dual-Regression Correlation Filters for Visual Tracking

“…Furthermore, it employs the dual regression model to train the filter so as to suppress background response fluctuations. Zhang et al [50] proposed a novel tracker with responseconsistent and background-suppression, which uses the known response of the previous frame as a consistency reference to guide the construction of the filter, and introduces an attention mask matrix to enhance the perception of background information. Fu et al [51] introduces the saliency detection method to construct the saliency perception regularity constraint of the target by perceiving the change of the appearance of the object, so as to achieve highlighting the appearance of the target while suppressing the irrelevant background noise.…”

FusionTrack: Multiple Object Tracking with Enhanced Information Utilization

“…The goal of multi-object tracking (MOT) is to track objects within a video sequence. Compared with single-object tracking [1][2][3], which aims to track a single designated target, multi-object tracking requires to detect and associate all targets within a sequence. The detection sub-task aims to identify and localize objects in the current frame, and the association sub-task aims to establish connections between targets across frames.…”