Vehicle Tracking in Wide Area Motion Imagery via Stochastic Progressive Association Across Multiple Frames (SPAAM)

Area of Science:

• Computer Vision
• Artificial Intelligence
• Data Science

Background:

• Vehicle tracking in Wide Area Motion Imagery (WAMI) requires associating detections across frames.
• Temporal window length (M) presents a trade-off between tracking accuracy and computational cost.
• Larger M improves context for motion modeling and handling occlusions, but increases computational complexity exponentially.

Purpose of the Study:

• To introduce SPAAM, an iterative approach for vehicle tracking in WAMI.
• To enhance tracking accuracy by progressively increasing the temporal context (M).
• To maintain computational manageability through novel hypothesis pruning strategies.

Main Methods:

• SPAAM progressively increases the temporal window (M) over iterations.
• Hypothesis pruning uses a co-registered road network to discard unlikely associations.
• A stochastic disassociation process limits new hypotheses by revisiting only open possibilities from the previous iteration.
• Associations are estimated globally by solving a binary integer programming problem.

Main Results:

• SPAAM demonstrates significant performance improvements over existing state-of-the-art methods.
• The approach effectively leverages enlarged temporal context for better track estimation.
• Computational complexity is managed effectively despite increasing temporal windows.

Conclusions:

• SPAAM offers a computationally efficient and accurate solution for vehicle tracking in WAMI.
• The proposed pruning methods successfully balance accuracy and complexity.
• This iterative approach advances the state-of-the-art in WAMI-based vehicle tracking.