Why do people appear not to extrapolate trajectories during multiple object tracking? A computational investigation.

Area of Science:

• Cognitive Science
• Computer Vision
• Psychology

Background:

• Extrapolation of object trajectories is intuitively expected to enhance visual tracking accuracy.
• However, human observers often neglect motion direction in multiple object tracking (MOT), favoring spatial memory.

Purpose of the Study:

• To investigate why extrapolation may not benefit multiple object tracking (MOT).
• To model human-like perceptual limitations, including noisy spatial perception, in MOT.

Main Methods:

• Developed probabilistic models incorporating perceptual limitations (e.g., noisy perception).
• Compared models with varying extrapolation weighting against a no-extrapolation baseline.
• Evaluated model performance across different weighting strategies for observations and predictions.

Main Results:

• No significant performance difference was found between models that weighted extrapolations and those that did not.
• Some models using extrapolation even performed worse than non-extrapolating models.
• Optimal models either avoided extrapolation or used it very conservatively, prioritizing direct observations.

Conclusions:

• Extrapolation is not always beneficial for visual tracking, especially in multiple object tracking (MOT) scenarios.
• Noisy perceptual inputs and featurally confusable targets complicate accurate trajectory extrapolation.
• Conservative or no extrapolation, relying on current observations, is often superior for robust MOT.