How natural language search works in Security Center SaaS

2026-03-13Last updated

Natural language search within the Investigation task allows users to find relevant video clips by typing search terms in plain language.

You can use everyday language search terms to describe what you are looking for. For example:
  • "person wearing a red top near the loading bay around 3 pm".
  • "person wearing a red top and blue pants yesterday afternoon".
  • "person wearing a white dress".
  • “white truck entering the car park yesterday afternoon”.
  • “blue sedan parked outside HQ entrance last night”.
  • "vehicle outside two days ago".
    Note:
    Location results are only shown if the location fully or part matches one or more camera names.

You can also input specific descriptors such as brands and details about what people are wearing, carrying, or doing. For example:

Vehicle make, model, and color
"Black Volkswagen Tiguan", "White Honda Civic", or "Red Tesla".
Text or images found on a vehicle
"FedEx", "UPS", "Taxi", and so on.
Person holding an object
"Woman holding a black bag", "Man holding a laptop", or "anyone carrying a backpack near the restricted area last weekend".
Vehicles with additional attributes
"Pick up with silver toolbox" or "Work truck with ladder on top".
Descriptors going beyond clothing color
"Man wearing a safety vest", "Woman wearing a toque", "Man in a black North Face jacket and grey hat", or "A person doing graffiti in November".

This enables you to concentrate on the most relevant search results and cut down the time to find evidence, helping you locate what you're looking for more quickly.

There are two ways to sort search results:
Relevance
Displays 500 most relevant results in order of relevance.
Most recent
Displays 500 most relevant results in reverse chronological order.
Note:
You can use filters to include, exclude, or apply specific attributes related to camera metadata, time, and location. Filters can also be used independently without relying on natural language search.
Investigation does not support searching for terms that contain offensive language or aim to expose sensitive information:
Subjective terms
Anything that cannot be determined by using an image of a person or vehicle. For example, “stealing”, “stupid person”, “speeding SUV”, “man holding a laptop”, or “woman next to a lamp post”.
Hate speech
Search term phrasings based on race, ethnicity, nationality, language, gender, sexual orientation, age, disability, health status, economic or social origin, social status, religion, or political views.
Unsupported terms
Searches that include phrases related to emotions, culture, or religion are not supported.

Time references

When time references appear in search results from natural language queries or the filter widget, they indicate the user's local time.

Each result is displayed using the camera's local time, which includes the seek time in the video player. For instance, if you're an operator on the East Coast managing sites on both the East and West Coasts, searching for 4 pm to 5 pm will show results timestamped 4 pm to 5 pm for East Coast cameras and 1 pm to 2 pm for West Coast cameras. Your descending list may also appear out of order since you'll see results at 4:59 pm, 1:58 pm, 4:47 pm, and so on.
Note:
For results outside your time zone, a globe icon () appears next to the time stamps. Hovering over the icon shows details about the relevant continent and country.