Can a Baseline Catch It? What Comparison-Based Traffic Detection Sees and Misses

Open with a split view: a highway under normal conditions, framed as the kind of scene a traffic camera monitors. Pose the driving question and frame the core idea — the system works by comparing 'now' to 'then', not by understanding the road.

• Comparison-based detection compares current imagery to a stored baseline
• The baseline defines what 'normal' looks like for that camera
• The driving question: which road changes does this catch, and which slip through?

Show three short road scenarios and let the learner predict, for each, whether a pixel-difference comparison against the baseline will catch it. Commit before seeing the detector's output.

• Scenario A: a fallen tree blocking a lane
• Scenario B: a long-standing construction zone painted into the baseline
• Scenario C: a pothole on a road surface that looks unchanged from above
• Learner predicts catch / miss for each before running the comparison

Reveal the detector's output for the three scenarios. Show highlighted difference regions over the road image and label each as 'flagged' or 'not flagged'.

• Fallen tree: large new pixel differences across the lane → flagged
• Painted-in construction zone: matches the baseline → not flagged
• Pothole: small, low-contrast change on a textured surface → missed
• Visible evidence: only changes that produce pixel-level differences above threshold trigger alerts

Let the learner manipulate the comparison — adjust the pixel-difference threshold, switch between raw pixel diff and edge-aware diff, and toggle 'ignore baseline regions'. Watch how flagged regions shrink, grow, or vanish.

• Lower threshold → more noise, more false flags
• Higher threshold → small real changes disappear (misses)
• Ignoring baseline-allowed regions hides legitimate but allowed changes
• Edge-aware diff is more robust to lighting changes than raw pixel diff

Lay out the structural limits that no threshold tuning fixes: changes the baseline already permits, changes outside the camera's view, and changes invisible to optics.

• Permitted in baseline: scheduled work zones, seasonal lane markings, routine detours
• Outside the field of view: incidents a block away, on cross streets, or behind the camera
• Invisible to the camera: subsurface damage (pothole roots, sinkholes, bridge scour), internal signal faults
• These limits are properties of the comparison method, not bugs to patch

A single transfer question: given a brand-new intersection with no stored baseline yet, predict whether a comparison-based detector can flag a stalled truck, and justify whether the 'no baseline' condition makes the system stronger or weaker.

• Compare: established baseline site vs. brand-new site with no baseline
• Decide: will the stalled truck be caught?
• Justify: how the absence of a baseline changes what 'catch or miss' even means

Run a second comparison on a different scene: a quiet rural road at night. Let the learner see whether a new pothole, a deer on the shoulder, and a faded re-painted lane line are caught or missed, then compare results to the daytime highway case.

• Rural night scene has lower contrast and more noise
• Deer on shoulder: large but transient — caught while present
• Faded re-paint: small contrast change — easily missed
• Lighting changes can swamp real differences unless edge-aware diff is used

Close with a direct answer to the driving question, summarizing the catch/miss categories and tying the boundary cases back to the comparison method itself.

• Catches: new, visible changes that differ from the stored baseline above threshold
• Misses: baseline-permitted changes, out-of-view events, and sub-visible conditions
• Threshold tuning trades false flags for missed detections but cannot remove the structural limits
• Bottom line: comparison-based detection is a 'difference' alarm, not an 'understanding' alarm