Visual Tracking via Sparse and Local Linear Coding.

Area of Science:

• Computer Vision
• Machine Learning
• Robotics

Background:

• Object tracking is crucial for many applications.
• Particle filters are effective but face challenges with discrete state spaces.
• Precise object localization is complicated by state space discretization.

Purpose of the Study:

• To propose a novel object tracking algorithm.
• To extend the state space of particle observations from discrete to continuous.
• To achieve accurate object localization via iterative linear coding.

Main Methods:

• The algorithm extends particle observation state space from discrete to continuous.
• Solution determined accurately via iterative linear coding between two convex hulls.
• Algorithm modeled by an optimal function solved by convex sparse coding or locality constrained linear coding.

Main Results:

• The proposed algorithm demonstrates accurate searching mechanisms using sparse representation.
• Flexibility is shown by implementing least soft-threshold squares and adaptive structural local sparse appearance models.
• Qualitative and quantitative results show favorable performance against state-of-the-art methods in dynamic scenes.

Conclusions:

• The novel tracking algorithm effectively addresses state space discretization issues.
• The continuous state space extension and iterative linear coding enhance tracking accuracy.
• The algorithm's flexibility and performance make it suitable for dynamic scene object tracking.