[Paper Review] Comparative Study of ECO and CFNet Trackers in Noisy Environment

Mustansar Fiaz, Sajid Javed+2 more|arXiv (Cornell University)|Jan 29, 2018

Video Surveillance and Tracking Methods14 references3 citations

TL;DR

This paper evaluates the robustness of two state-of-the-art visual trackers, ECO and CFNet, under increasing noise levels in surveillance-like conditions. Using controlled noise injection on benchmark sequences, the study demonstrates that both trackers degrade significantly with higher noise, highlighting the critical need for noise-robust tracking algorithms in real-world applications where image quality is often compromised.

ABSTRACT

Object tracking is one of the most challenging task and has secured significant attention of computer vision researchers in the past two decades. Recent deep learning based trackers have shown good performance on various tracking challenges. A tracking method should track objects in sequential frames accurately in challenges such as deformation, low resolution, occlusion, scale and light variations. Most trackers achieve good performance on specific challenges instead of all tracking problems, hence there is a lack of general purpose tracking algorithms that can perform well in all conditions. Moreover, performance of tracking techniques has not been evaluated in noisy environments. Visual object tracking has real world applications and there is good chance that noise may get added during image acquisition in surveillance cameras. We aim to study the robustness of two state of the art trackers in the presence of noise including Efficient Convolutional Operators (ECO) and Correlation Filter Network (CFNet). Our study demonstrates that the performance of these trackers degrades as the noise level increases, which demonstrate the need to design more robust tracking algorithms.

Motivation & Objective

• To assess the robustness of ECO and CFNet trackers in noisy visual environments.
• To investigate how increasing noise levels affect tracking accuracy and stability.
• To identify limitations of current deep learning-based trackers under real-world image degradation.
• To provide empirical evidence for the need to develop more noise-resilient tracking algorithms.

Proposed method

• The study applies synthetic noise (Gaussian, Poisson, and speckle) at varying intensity levels to video sequences from standard tracking benchmarks.
• ECO and CFNet trackers are evaluated on standard datasets under increasing noise levels using standard tracking metrics such as precision and success rate.
• Noise is injected at multiple levels (low, medium, high) to simulate real-world surveillance camera degradation.
• Performance is quantified using standard evaluation protocols, including precision plots and success curves.
• The experiments are conducted under controlled conditions to isolate the effect of noise on tracker performance.

Experimental results

Research questions

• RQ1 How does increasing noise intensity affect the tracking accuracy of ECO and CFNet?
• RQ2 Which tracker—ECO or CFNet—exhibits greater robustness under noisy conditions?
• RQ3 To what extent do common visual tracking challenges like occlusion and deformation interact with noise-induced degradation?
• RQ4 Does noise significantly reduce the performance of state-of-the-art deep learning-based trackers in real-world scenarios?

Key findings

• Both ECO and CFNet show significant performance degradation as noise levels increase, particularly in terms of success rate and precision.
• The degradation is more pronounced in high-noise conditions, with success rates dropping substantially across multiple benchmarks.
• CFNet maintains slightly better performance than ECO under low to medium noise, but both suffer under high noise.
• The results indicate that current state-of-the-art trackers are not robust to common image noise found in surveillance systems.
• The study confirms that noise is a critical factor in real-world tracking failure, often overlooked in standard benchmark evaluations.