A return visit.
A new window mark.
One car left and returned with a temporary mark on its side window. The prototype compared the recorded visits and flagged possible changes.
Original evidence · Cropped
Look at the same part of the window.
These enlarged crops retain the original image detail. The temporary mark appears on the return visit.
Fleet Snout at your depot
Act before the next hire.
In this trial, a new window mark became a specific item for review.
- Focus inspection effort
- Go straight to the flagged area and compare the two visits before a closer inspection.
- Back damage discussions
- Use the departure and return images when you discuss a new mark with a customer or driver.
- Plan the next hire
- Use the reviewed finding to decide whether further inspection or repair is needed before the asset goes out again.
Method and limits
A fixed camera recorded one departure and one return. Vehicle outlines guided image selection. The comparison matched similar views and searched for new visual structure across repeated return frames.
The original capture sampled five frames per second. The software compared 57 departure images with 42 return images. It aligned 41 return views and rejected one.
Three candidate views showed the temporary window mark. One was a wheel-edge false alarm. These are views of possible changes, not four separate defects. The strongest mark candidate repeated across six return frames.
Separate controls compared 14 departure pairs and 16 return pairs. These comparisons within each visit produced no persistent candidates.
This recording guided development. It does not measure accuracy on unseen footage. The mark was temporary. The trial does not prove that glass was cracked or establish reliable damage detection.
The depot uses above describe the intended workflow. This controlled trial did not measure inspection time saved, damage costs recovered, or reduced downtime.
Reflections, occupants, shadows, and moving parts can cause false alarms. Hidden surfaces remain unassessed. Human review is required.
The driving sequence is AI-edited. Its scene, vehicle appearance, and timing differ from the source. Added frames smooth the edited motion for 60 fps playback. Vehicle outlines come from the edited footage, and window brackets guide the presentation. The close-up and comparison use original cropped images. The results here come from the original trial.