BIT: Biologically Inspired Tracker.

Area of Science:

• Computer Vision
• Computational Neuroscience
• Biologically Inspired Computing

Background:

• Visual tracking faces challenges from object deformation, scale, illumination changes, and occlusion.
• The human visual system (HVS) exhibits superior tracking capabilities, motivating biologically inspired models.
• Understanding neural mechanisms in the HVS is crucial for developing advanced computer vision trackers.

Purpose of the Study:

• To develop a biologically inspired visual tracking model by analyzing the ventral stream's visual cognitive mechanisms.
• To simulate neural units for extracting low-level features and employing advanced learning for target localization.
• To achieve real-time performance in visual tracking through efficient algorithms.

Main Methods:

• Simulated shallow neurons (S1 and C1 units) to extract biologically inspired features.
• Imitated advanced learning mechanisms (S2 and C2 units) combining generative and discriminative models.
• Utilized fast Gabor approximation and fast Fourier transform for real-time processing.

Main Results:

• The proposed biologically inspired tracker demonstrated superior performance compared to state-of-the-art methods.
• Achieved high accuracy, robustness, and efficiency in extensive experiments on large-scale datasets.
• An acceleration technique enabled the tracker to maintain a speed of approximately 45 frames/s.

Conclusions:

• The biologically inspired tracker effectively addresses challenges in visual tracking.
• The model's design, inspired by the human visual cortex, offers significant improvements in tracking performance.
• Real-time processing capabilities make the tracker suitable for practical applications.