Layered data association using graph-theoretic formulation with applications to tennis ball tracking in monocular...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Object tracking in cluttered environments with switching dynamics is a significant challenge.
• Existing data association methods often struggle with complex scenarios, leading to track fragmentation or loss.

Purpose of the Study:

• To develop a robust multilayered data association scheme for multi-object tracking.
• To improve the accuracy and reliability of object trajectory identification in dynamic and cluttered scenes.

Main Methods:

• A graph-theoretic formulation is used to model object candidates and tracklets.
• Tracklets are grown from high-probability object candidates.
• Data association is solved as an all-pairs shortest path (APSP) problem on a weighted graph of tracklets.
• An efficient APSP algorithm exploiting graph properties is developed.

Main Results:

• The proposed scheme successfully identifies object trajectories, handling track initiation and termination automatically.
• Experiments on tennis sequences demonstrate superior performance compared to existing methods, especially in difficult conditions.
• The method shows robustness in scenarios where other approaches fail.

Conclusions:

• The multilayered data association scheme provides an effective solution for multi-object tracking with switching dynamics.
• The graph-theoretic approach and optimized APSP algorithm offer improved performance and efficiency.
• This method has practical applications in sports analytics and other complex tracking tasks.