Application Note: Video – ANV109
Enhanced Video Search in Brivo Eagle Eye Video
2026-10-12 Revision 1.0
This Application Note is for end customers and resellers using Natural Language Video Search on Brivo Eagle Eye Video, including security managers, investigators, and operations teams who search recorded video to locate people, vehicles, and scene activity during incident investigations. It explains how to write effective natural language queries, the query types supported in this release, and best practices for getting relevant, reliable results.
Search your video the way you'd describe it. "Man in a grey hoodie carrying a duffle bag"
"White pickup truck beside orange traffic cones." Type what you saw, and get the video back in seconds.
When an incident happens, finding the right clip is a race. Until now, that meant working within a fixed set of attributes: gender, a few clothing and vehicle colors, a handful of object types, and hoping your incident fit the list. When it didn't, the only fallback was scrubbing through hours of video across multiple cameras to find a single moment.
Natural Language Video Search removes the list. Describe people, vehicles, objects, and scene context in plain English, and the system finds matching frames, ranked so the strongest results come first, drawn from every camera on the site. A hoodie, a duffle bag, a ladder left by a shelf: if you can describe it, you can search for it.
The experience is a single search bar. You type a description, and results return as a grid of thumbnails, each showing exactly where the match appears.
Queries it understands (Dogfood) :
Person in a reflective vest carrying a toolbox
Woman on the phone holding a coffee cup
White pickup truck beside orange traffic cones
Two people standing at a bicycle rack
Sign reading "Innovation Lab" on a wall
The prerequisites for Natural Language Video Search are below.
Video Search during dogfood has two modes, switched with the toggle at the top of the page: Standard Search, the structured attribute search, and Natural Language, where you describe what you're looking for in plain English. Flip the toggle to Natural Language, and the search bar lights up, ready for a description.
Natural Language Video Search is built for precision. When it returns a result, it's a strong match for what you described - so you spend your time reviewing relevant video, not filtering out noise.
Before you've searched, the page shows a set of example queries under "See what's possible" - a quick model for how to phrase a description:
Person in a reflective vest carrying a toolbox
Woman on the phone holding a coffee cup
White pickup truck beside orange traffic cones
Two people standing at a bicycle rack
Man pushing a luggage cart with two suitcases
Enter your description and run the search. The search bar animates while results are retrieved, then returns a grid of video thumbnails ranked by relevance, pulled from across every camera in scope - no need to pick a camera first. Each result shows its camera name and timestamp, and a results count sits above the grid.
Click any thumbnail to open it in a larger pop-up view, where you can review the moment in detail and confirm it's what you're after - then continue into your normal review or your workflow.
Sort results by Most relevant (the default) or by time. Narrow what you get back with the scope controls beside the search bar - time range, Layouts, Tags, Sites, and additional options under Filters. Use Clear filters to reset. Describe what you saw in the query; use the filters to bound where and when to look.
Video Search understands what's visible in a single frame - people, vehicles, objects, and the scene around them. This is the biggest shift from Standard Search, which was limited to a fixed set of attributes: gender, a few clothing and vehicle colors, and a handful of object types. Here, you describe what you saw.
People: Find people by appearance rather than identity - clothing type and color, carried items, and simple posture.
"Woman in a red jacket carrying a backpack."
"Person in a reflective vest carrying a toolbox."
“Domino's delivery guy”
Vehicles: Search by color, type, make, and distinguishing features.
"White pickup truck beside orange traffic cones."
"FedEx truck"
Objects and scenes: Find everyday objects and the context around them - what's in the frame and where.
"Ladder near shelves."
"Two people standing at a bicycle rack."
Simple activity: Search for what a person is visibly doing in the moment: standing, sitting, walking, carrying.
"Person sitting on a bench."
"Man pushing a luggage cart with two suitcases."
Searching a specific time or place: Typing a time or location into the query - "yesterday," "entrance", “day before” - isn't supported in this release. To scope a search by time or place, use the time-range, Sites, and Layouts filters beside the search bar instead.
(Understanding time and location directly from the query is planned for a future release.)
Knowing the edges keeps your searches effective:
It doesn't identify people. Search matches appearance, not identity - no facial recognition, no naming individuals.
It reads moments, not intent. It can find a person carrying a bag, but not judge whether they're stealing it. Describe what's visible, not what you think is happening.
It sees single frames, not sequences. Actions that unfold over time - walking from a car to a door, swiping a badge - aren't searchable. Describe the one moment you'd see in a still.
It doesn't count or add up. "How many people entered today" isn't a search it can answer.
It can't search for the absence of something. "Workers without helmets" won't work - search for what's present instead ("workers wearing helmets"), then review the results.
Video Search returns strong matches when you describe what a camera actually captured. A few habits get you there.
4.1 Describe what's visible, not what happened
Search matches appearance in a single frame, so anchor your query to things you could point to in a still image - clothing, color, objects, posture. "Person in a grey hoodie carrying a duffle bag" gives the system something concrete to match. "Person acting suspiciously" doesn't - there's nothing visible to search on.
4.2 Lead with the details that stand out
The distinctive parts of a description do the most work. A color, a carried item, a vehicle type, a specific object - these narrow results fast. "White pickup truck beside orange traffic cones" is far more precise than "white truck"
4.3 Stack attributes to sharpen the ranking
The more distinct attributes you give, the better the system can push the right match to the top. Instead of "person with long hair", try "person with long hair in a black t-shirt wearing headphones." Each added detail—appearance, clothing, carried items—helps the strongest match rank higher, rather than leaving you to scan a broad set. The same applies to vehicles and scenes: "red sedan with a roof rack" beats "red car".
4.4 Be specific first, then broaden for more
Specific queries return tighter results. If you're not seeing enough, loosen one detail and search again - "blue jacket" instead of "navy blue windbreaker", or drop a secondary attribute. Describing the same subject differently often surfaces matches the first phrasing missed. This is the main lever you have when a result you expect isn't showing up.
4.5 Keep it to one moment
Describe a single frame, not a sequence of actions. "Person carrying a box" works; "person walking from the car to the entrance" describes something that unfolds over time. Search for the one moment you'd actually see in a still.
You don't need perfect phrasing. Describe what you saw in your own words - Video Search may fine-tune the wording to improve results, and always lets you run your original query instead. The habits above simply help you get there faster.
Results are ranked so the strongest matches for your description appear first. Because Video Search favors precision, the results you see are ones the system is confident about - low-confidence matches are held back rather than padding the grid with noise.
Reading the grid
Each result is a video thumbnail showing the frame where the match appears, labeled with its camera name and timestamp. A results count sits above the grid. Click any thumbnail to open it in a larger pop-up view.
When your query is adjusted
To return better matches, Video Search may rephrase your query - for example, interpreting "cleaning person" as "person cleaning." It runs the adjusted version automatically and shows you what it searched for, with a one-click option to search your original wording instead. If the adjusted results aren't what you meant, switch back and compare.
Sorting
Results are sorted by Most relevant by default. Switch to Most recent first or Oldest first to work through matches chronologically - useful when reconstructing a sequence of events across cameras.
When results are incomplete
Video Search is in Beta and still improving. It's strong at returning relevant matches but doesn't yet surface every possible one, so treat a set of results as your best strong matches, not a guarantee of every clip. If you're not seeing what you expect, broaden your description or adjust the time range and search again. The results view carries a thumbs up/down control - your feedback helps tune relevance over time.
When there are no results
If nothing matches, you'll see an empty state. Try describing the subject differently, loosen a specific detail, or widen the time range.
When a camera detects motion, the Bridge or CMVR sends key images from the camera’s preview stream to the Brivo cloud. Cloud AI analyzes each image and indexes the people, vehicles, objects, and scenes it contains, and Video Search queries that index. If preview images can’t reach the cloud, that camera can’t be searched.
6.1 Editions. On Professional and Enterprise, video is indexed as it’s recorded, so results return in seconds. On Standard (later release), you select the cameras and time window to search. Video Search indexes that video and notifies you when results are ready, typically within a few minutes depending on how many cameras and how much time you select. Searches on that window then return in seconds.
6.2 CMVRs. Video Search works with CMVRs the same way it works with Bridges. Full-resolution video stays on the CMVR, so playing a result streams the clip from the CMVR. The CMVR must be online.
6.3 M10 and PR1
Sending previews continuously adds about 0.1–0.2 Mbps of upload per camera (see ANV106).
6.4 Low data mode. Video Search doesn’t work on a Bridge or CMVR with Low data mode on (Bridge Settings → General; called Minimum Bandwidth Mode in the Classic interface). This mode sends previews only while someone is viewing them, so no key images reach the cloud for indexing. To make those cameras searchable, turn Low data mode off.
Natural Language Video Search changes how investigations start. Instead of translating what you saw into a fixed set of attributes, or scrubbing hours of video when it doesn't fit, you describe the person, vehicle, object, or moment in plain English and get strong, relevant matches in seconds. Clear, specific descriptions and the scope filters put the right video in front of you fast, so Video Search becomes the first place you look, not the last resort. As a Beta, it's still improving with every search and every piece of feedback you send.