Exploring the Effects of Blur and Deblurring to Visual Object Tracking

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Motion blur is a critical challenge in visual object tracking, yet its quantitative impact remains understudied.
• The utility of image deblurring for enhancing visual object tracking performance is not well understood.

Purpose of the Study:

• To establish a benchmark for evaluating visual object tracking under varying motion blur conditions.
• To investigate the effects of different motion blur levels and deblurring techniques on tracking accuracy.

Main Methods:

• Development of the Blurred Video Tracking (BVT) benchmark with diverse motion blur levels and ground truth.
• Extensive evaluation of 25 state-of-the-art visual trackers on the BVT benchmark.
• Proposal and implementation of a Generative Adversarial Network (GAN)-based scheme for adaptive frame deblurring.

Main Results:

• Light motion blur can improve tracker accuracy, whereas heavy blur significantly degrades performance.
• Image deblurring enhances tracking in heavily blurred videos but can hinder performance in lightly blurred scenarios.
• The proposed GAN-based scheme effectively improves the robustness and accuracy of multiple trackers on motion-blurred videos.

Conclusions:

• Motion blur's effect on visual tracking is level-dependent, necessitating adaptive strategies.
• Deblurring is a viable but context-specific solution for improving tracking in blurred videos.
• The developed BVT benchmark and GAN-based deblurring scheme offer valuable contributions to robust visual object tracking research.