Attribute search lets an investigator find footage by the characteristics of an object rather than by time alone — "red vehicles", "a person wearing a backpack", "a truck heading north". As video is recorded, analytics tag detected objects with attributes (class, colour, size, direction, and more) and write them to a metadata index; attribute search queries that index and returns matching clips in seconds instead of hours of manual review. It sits a rung above simple class search and below full appearance or natural-language search.
Its power is speed over a large archive: because it reads compact metadata, not the video pixels, a query across days of footage returns a ranked set of candidates almost immediately. This is what makes "find the silver van that passed between 3 and 4" a practical investigative step rather than an afternoon of scrubbing. It is the everyday workhorse of forensic search in a modern VMS.
The defining pitfall is that you can only search for attributes that were extracted at record time. If the analytics never captured colour, you cannot search by colour; if an object was never detected, no attribute query will surface it — recall is capped by the upstream detection and tagging. Reprocessing the raw video to extract more attributes is possible but slow. Results are ranked candidates to confirm, not exact matches, so attribute search is triage, never 100% — and what you index today determines what you can find tomorrow.
