Smooth tracked positions with Kalman filter

Use This When

• Tracking positions jitter or jump between frames due to detection noise
• Building analytics that depend on smooth trajectories or velocity estimates
• Preparing tracked objects for visualization where stable positions improve clarity

What It Does

• Maintains per-object Kalman filters that smooth noisy position measurements over time
• Estimates velocity and orientation vectors from position history for motion analysis
• Automatically forgets objects that disappear for extended periods to prevent memory bloat
• Outputs refined positions and dominant motion direction for each tracked object

Works Best With

• Object trackers → this component → distance calculators, zone analytics, or visualizers
• Any pipeline where tracked position stability matters more than raw detection precision
• Motion-based logic that needs velocity vectors for direction or speed thresholds

Caveats

• Process noise must be tuned to motion characteristics; too low under-smooths, too high over-smooths
• Objects not observed for forget_threshold frames are permanently removed from state
• Assumes relatively constant velocity; rapid acceleration or erratic motion reduces accuracy