How plain-language search changes video investigations

An investigation usually begins with a vague report. A customer says a bag went missing sometime after lunch. A supervisor says a pallet was damaged on the late shift. Somebody saw a man in a green jacket near the staff entrance. The footage exists, and the work lies in finding the right thirty seconds of it.

With a conventional recorder the method is to pick a camera, pick a time and scrub. When the time is uncertain, that means hours of fast-forward for each camera. Motion search narrows the footage to moments when something moved, which in a busy building is most of the day. Attribute filters help when the incident happens to fit a fixed list of options.

Describing what you are looking for

Plain-language search works on a different principle. The vision model has already read the footage and understands the people, objects and activity in it, and it understands written descriptions in the same terms. An investigator types what they would say to a colleague: “man in a green jacket carrying a black backpack,” “white van at the loading bay,” “person lying on the floor in the corridor.” The system returns the matching moments from every camera, each with its time and location.

There are two other ways to start. A photo of a person or an item finds other appearances of the same thing across the site. Where face search is enabled, a face image finds each time that person passed a camera.

Writing a query that works

The most effective queries describe what a camera would see. A report of theft becomes “person putting bottles into a backpack.” A report of a trespasser becomes “person climbing over a fence.” A few habits make the difference.

• Describe appearance and objects: clothing color, bags, helmets, uniforms, vehicles.
• Describe visible actions: running, climbing, carrying a box, lying on the floor.
• Add a time window and a set of cameras when you know them, and search the whole site when you do not.
• Start broad, then add detail. “Person with a red suitcase” first, “near the elevators after 6 p.m.” second.

From one sighting to the whole story

The first good match is the start of the investigation. Use that sighting as the next query: search by the person’s image to find where they entered, which route they took, who they met and which vehicle they left in. Each result adds a point to a timeline that crosses cameras without anyone needing to know the camera layout by heart.

This changes who can investigate. A store manager or a shift supervisor can answer a routine question in a few minutes, and the security team spends its time on the cases that need judgment.

Keeping the result usable as evidence

A fast search is valuable only if the result stands up afterward. Export the original clip with its timestamp and camera name, and keep it unedited. Record who ran the search, when and for what reason. Limit search access to named roles, and enable face search where policy and local law provide for it.

Store exported clips in one case folder with a short note of the query that found each one, so a colleague, an insurer or the police can follow the same steps. Maiyn Investigate searches footage by description, by photo and, where enabled, by face.