Extract per-object motion vectors from optical flow

Use This When

• Converting dense optical flow fields into per-object motion summaries for tracking or behavior analysis
• Determining which segmented objects are moving and in what direction from frame-to-frame flow
• Building motion-triggered alerts or filters that need object-level movement direction and magnitude

What It Does

• Aggregates dense optical flow within each segmentation mask to compute mean motion vector per object
• Filters flow by magnitude threshold to ignore camera noise and focus on actual movement
• Outputs object center position paired with dominant motion direction for each mask
• Can reverse direction or scale magnitude to match coordinate conventions or downstream needs

Works Best With

• Optical flow calculator + segmenter → this component → tracker or motion-based classifier
• Pipelines needing object motion context without full per-pixel flow analysis
• Scenarios where segmentation masks define regions of interest for movement extraction

Caveats

• Movement threshold must be tuned to flow magnitude scale; too high misses slow motion, too low includes noise
• Averages all flow within mask; complex internal motion patterns collapse to single vector
• Mask quality directly affects accuracy; poor segmentation yields unreliable motion estimates