Markerless human motion tracking using hierarchical multi-swarm cooperative particle swarm optimization.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Markerless full-body articulated human motion tracking presents a high-dimensional search space challenge.
• Conventional Particle Swarm Optimization (PSO) methods often suffer from premature convergence and local optima, impacting tracking accuracy.

Purpose of the Study:

• To develop an improved metaheuristic approach for accurate markerless human motion tracking.
• To address the limitations of classical PSO in complex, high-dimensional optimization problems.

Main Methods:

• Formulation of human motion tracking as a non-linear 34-dimensional function optimization problem.
• Development and application of Hierarchical Multi-Swarm Cooperative Particle Swarm Optimization (H-MCPSO).
• Utilization of silhouette and edge likelihoods within the fitness function to quantify model-image discrepancies.

Main Results:

• H-MCPSO demonstrated superior performance compared to Annealed Particle Filter (APF) and Hierarchical Particle Swarm Optimization (HPSO) on the Brown and HumanEva-II datasets.
• The proposed method achieved automatic initialization and robust self-recovery from temporary tracking failures.
• Experimental results validate the effectiveness and reliability of the H-MCPSO approach.

Conclusions:

• H-MCPSO offers a significant advancement in markerless human motion tracking accuracy and robustness.
• The developed method effectively overcomes the limitations of traditional PSO algorithms in this domain.
• The system's ability for automatic initialization and self-recovery enhances its practical applicability.