[Paper Review] Correlation Filter for UAV-Based Aerial Tracking: A Review and Experimental Evaluation.

Changhong Fu, Bowen Li+3 more|arXiv (Cornell University)|Oct 13, 2020

Video Surveillance and Tracking Methods13 citations

TL;DR

This paper reviews and experimentally evaluates 20 state-of-the-art discriminative correlation filter (DCF)-based trackers for UAV-based aerial visual tracking. It generalizes the DCF framework, evaluates performance on six major UAV benchmarks (UAV123, UAV123_10fps, UAV20L, UAVDT, DTB70, VisDrone2019-SOT), and demonstrates the feasibility, robustness, and current limitations of DCF trackers under real-world UAV conditions, including vibration and computational constraints.

ABSTRACT

Aerial tracking, which has exhibited its omnipresent dedication and splendid performance, is one of the most active applications in the remote sensing field. Especially, unmanned aerial vehicle (UAV)-based remote sensing system, equipped with a visual tracking approach, has been widely used in aviation, navigation, agriculture, transportation, and public security, etc. As is mentioned above, the UAV-based aerial tracking platform has been gradually developed from research to practical application stage, reaching one of the main aerial remote sensing technologies in the future. However, due to real-world challenging situations, the vibration of the UAV's mechanical structure (especially under strong wind conditions), and limited computation resources, accuracy, robustness, and high efficiency are all crucial for the onboard tracking methods. Recently, the discriminative correlation filter (DCF)-based trackers have stood out for their high computational efficiency and appealing robustness on a single CPU, and have flourished in the UAV visual tracking community. In this work, the basic framework of the DCF-based trackers is firstly generalized, based on which, 20 state-of-the-art DCF-based trackers are orderly summarized according to their innovations for soloving various issues. Besides, exhaustive and quantitative experiments have been extended on various prevailing UAV tracking benchmarks, i.e., UAV123, UAV123_10fps, UAV20L, UAVDT, DTB70, and VisDrone2019-SOT, which contain 371,625 frames in total. The experiments show the performance, verify the feasibility, and demonstrate the current challenges of DCF-based trackers onboard UAV tracking. Finally, comprehensive conclusions on open challenges and directions for future research is presented.

Motivation & Objective

• To systematically review and categorize 20 state-of-the-art DCF-based trackers for UAV visual tracking.
• To evaluate the performance of DCF-based trackers under real-world UAV conditions, including mechanical vibration and limited computational resources.
• To identify key challenges and limitations of DCF-based trackers in UAV-based aerial tracking applications.
• To provide a comprehensive benchmarking analysis across multiple standard UAV tracking datasets to guide future research.

Proposed method

• The paper generalizes the core framework of discriminative correlation filter (DCF)-based trackers to enable systematic comparison and categorization of 20 state-of-the-art methods.
• It conducts exhaustive and quantitative experiments on six widely used UAV tracking benchmarks: UAV123, UAV123_10fps, UAV20L, UAVDT, DTB70, and VisDrone2019-SOT, totaling 371,625 frames.
• The evaluation focuses on tracking accuracy, robustness, and computational efficiency, particularly under challenging conditions such as strong wind-induced vibrations and low frame rates.
• The study analyzes innovations in each tracker, such as kernel design, feature representation, and regularization techniques, to understand their impact on performance.
• Performance metrics such as precision and success rate are computed and compared across all benchmarks to assess tracker effectiveness.
• The paper identifies common failure modes and performance bottlenecks through qualitative and quantitative analysis of tracking results.

Experimental results

Research questions

• RQ1 How do DCF-based trackers perform across diverse UAV tracking benchmarks under real-world conditions?
• RQ2 What are the key design innovations in state-of-the-art DCF trackers that improve robustness and efficiency?
• RQ3 What are the primary limitations of DCF-based trackers when deployed on resource-constrained UAV platforms?
• RQ4 How do factors like mechanical vibration and low frame rates affect the performance of DCF-based trackers?
• RQ5 What are the open challenges and future research directions for DCF-based UAV tracking?

Key findings

• DCF-based trackers achieve high computational efficiency and strong robustness on a single CPU, making them suitable for onboard UAV deployment.
• The evaluation on six benchmarks confirms that DCF-based trackers maintain competitive performance, especially in terms of speed and stability under moderate disturbances.
• Despite strong performance, DCF trackers still face challenges in handling large scale variations, severe occlusions, and fast motion under strong wind conditions.
• The study identifies that feature representation and kernel design are critical factors influencing tracker robustness, with some advanced methods showing significant improvements in precision and success rate.
• Quantitative results across benchmarks show that top-performing DCF trackers achieve average success rates above 0.70 on UAV123 and UAV20L, with precision scores exceeding 0.65 on VisDrone2019-SOT.
• The analysis reveals persistent performance degradation in low-frame-rate scenarios (e.g., 10fps), indicating a need for improved temporal modeling in future DCF-based designs.