Computer vision · Video synopsis
A whole night, in minutes.Only what moved.
Video synopsis, also called video summarisation, is the computer vision technique that extracts only the moving objects from long recordings and shows them together, on the same background, each with the time it appeared. We build systems that condense hours of footage from fixed cameras into a short summary, searchable by object type, zone and time range, from which one click takes you to the original clip.
Events
- —4 h of footage → a 2-minute synopsis
- —Filter: vehicles and people
- —Each object keeps its real time
- 13:47Jump to the original footage
The footage exists, but nobody has time to watch it
Cameras record around the clock, yet an archive of dozens of hours is almost impossible to review in good time. When someone looks for an event, they scrub backwards and forwards by hand, at the risk of missing exactly the moment that matters.
- Reviewing one night of footage from a single camera can take hours
- Long stretches with no activity take up most of the viewing time
- The archive can only be searched by date and camera, not by what happened in the frame
- With many cameras, checks are selective and some areas are never reviewed
Who it is for
- Security teams who review footage after an incident or at the start of a shift
- Warehouses, logistics yards and industrial sites with many cameras and little time to watch them
- Building managers who need to find out quickly what happened in an area during a given period
- Operations teams who want to understand how vehicles and machinery move through a space
Step by step
How a video synopsis is built, step by step
The five steps on the animation show how a long recording becomes a short clip: from the fixed camera's archive, through detecting and tracking objects, to composing the synopsis and the operator who jumps to the original footage.
Events
Events
- —4 h of footage → a 2-minute synopsis
Events
- —4 h of footage → a 2-minute synopsis
- —Filter: vehicles and people
Events
- —4 h of footage → a 2-minute synopsis
- —Filter: vehicles and people
- —Each object keeps its real time
- 13:47Jump to the original footage
Events
- —4 h of footage → a 2-minute synopsis
- —Filter: vehicles and people
- —Each object keeps its real time
- 13:47Jump to the original footage
Step 1 of 5
The fixed camera's archive
We start from the recording of a camera that does not move, exported from the NVR or read directly from the stream. From frames with no activity, the system learns what the empty scene looks like, the background, and updates it slowly as the light changes through the day. A camera that pans or changes zoom spoils that background, so we check this first.
Technical options and when each fits
On-demand synopsis from the archive
The synopsis is generated only when someone asks for a time range, from recordings that already exist. It suits sites where searches are rare, for example after an incident, and do not justify processing every camera continuously.
Continuous indexing
Detection and tracking run as the camera records, so the object index is always ready. Searches answer straight away, for example at the start of a shift, but this needs processing resources running all the time.
Path map
Instead of a clip, the paths from a time range are combined into one image showing where movement was heaviest and where objects stopped. It is useful for operations teams who want to understand flows through a yard or a hall rather than look for one specific event.
What you receive at hand-over
- The detection model for the agreed classes, fine-tuned on footage from your cameras
- The configured connection to your NVR, VMS or video streams, for the chosen cameras
- The search and synopsis interface, with filters by class, zone, direction and time range and a jump to the original footage
- Access permissions and retention rules for synopses and the index, aligned with your video surveillance policy
- A test report on clips the team has already reviewed, showing what appeared in the synopsis and what was missing
- Documentation and, typically, a short training session for operators
No strings attached: tell us what the cameras should see and we'll tell you honestly whether it can be done.
Let's talkWhat the video synopsis system can do
- 01
Condenses hours of footage
Objects that crossed the frame at different times are shown simultaneously, so a night's activity can be seen in a short clip.
- 02
Filters by object type
The synopsis can show only vehicles, only people or only machinery, based on the classes we agree together.
- 03
Searches by zone and direction
You can ask for only the objects that entered a particular zone, crossed a line or moved in a particular direction.
- 04
Marks the time of each object
Each object in the synopsis shows the real time it appeared, so you know exactly when it happened.
- 05
Links to the original clip
One click on an object in the synopsis opens the original recording at that moment, for a full review.
- 06
Blurs people when needed
The synopsis can be generated with people blurred, and the system does not analyse faces or identify anyone.
How it works
How hours of footage become a summary of a few minutes
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Human in the loop
- Result
- 01
Input
Camera recordings
The archive of a fixed camera, from an NVR, VMS or video files, for the chosen period.
- 02
AI
Detection and tracking
The model finds moving objects and tracks their path throughout the recording.
- 03
Action
Background extraction
The background image of the scene is built from frames with no activity.
- 04
Action
Synopsis composition
Object paths are placed on the background, compressed in time and marked with the real time.
- 05
Human in the loop
Review and verification
The operator goes through the synopsis and opens the original clips worth a closer look.
- 06
Result
Searchable synopsis
You are left with a short, filterable clip linked to the original recordings.
What a useful video synopsis needs
- Recordings from fixed cameras; cameras that pan or change zoom often produce synopses that are hard to use
- Access to the archive through an NVR or VMS export, video files or the RTSP stream, where the equipment allows
- The list of object classes you want to search by, for example people, cars, lorries or forklifts
- The access and retention rules for footage set out in your video surveillance policy
- Processing resources suited to the volume of footage: a local server with a GPU or the cloud
Possible integrations
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- Video exports from NVRs and VMS systems
- MP4 or MKV video files
- FFmpeg for video decoding and conversion
- Local servers with GPUs
- NVIDIA Jetson edge devices
- Cloud storage, for example Amazon S3 or Azure Blob Storage
- PostgreSQL databases for the object index
- REST API and webhooks
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- Video exports from NVRs and VMS systems
- MP4 or MKV video files
- FFmpeg for video decoding and conversion
- Local servers with GPUs
- NVIDIA Jetson edge devices
- Cloud storage, for example Amazon S3 or Azure Blob Storage
- PostgreSQL databases for the object index
- REST API and webhooks
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays with people
- Interpreting what the synopsis shows and deciding whether an event is an incident
- Checking any important moment on the original recording before drawing conclusions
- Deciding who has access to the synopses and the archive, in line with the video surveillance policy
- Decisions on keeping, handing over or deleting recordings, including to the authorities on request
Limits and operating conditions
- Synopsis works on fixed cameras; a camera that moves or changes angle spoils the background and the object paths
- In very busy scenes, objects overlap in the synopsis; filters by class, zone or time range then become necessary
- Objects that do not move, for example a parcel left on the floor, appear only when they are brought in or taken away, not for the whole period
- The synopsis is a review tool; the reference remains the original recording, which must be consulted for any conclusion
How we evaluate a video synopsis system
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Missed activity
- Events noted by the team when watching test clips in full, compared with those that appear in the synopsis
- Filter precision
- The share of objects shown for a search that really belong to the requested class or zone, checked manually on samples
- Review time
- The time operators need to find known events, with and without the synopsis, on the same recordings
- Synopsis readability
- Operator feedback on how cluttered the clip is, compared across quiet and busy periods and different compression levels
How we implement video synopsis, in 4 steps
- 01
We choose the cameras and the purpose
We agree which cameras matter, which situations are usually searched for and who will use the synopses.
- 02
We check access to the archive
We test the export or video stream from your equipment and the image quality in real conditions.
- 03
We configure and test on real footage
We generate synopses of clips the team has already reviewed and check what appears and what is missing.
- 04
We provide the interface
Operators get access to search and synopses, with set permissions, and quality is monitored over time.
Use cases
The scenarios below are conceptual examples, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Logistics
The night in the warehouse yard, reviewed in the morning
In the morning, the security guard receives the night's synopsis for the yard cameras. They see all the vehicles that came in within a short clip and open the original footage only for the one that stopped by the fence.
Conceptual example - Manufacturing
Looking for a missing pallet in the dispatch area
The team searches only for the people and forklifts that entered the dispatch area within a two-hour window, then checks on the original recording the moment the pallet was moved.
Conceptual example - Facilities management
Access to the underground car park after hours
The building manager generates a synopsis of vehicles that entered after closing time over a week, to see whether the barrier stays open longer than it should.
Conceptual example
Useful guides
Before you decide, read
Frequently asked questions
What is video synopsis?
It is a computer vision technique that extracts moving objects from long recordings and shows them together, on the background of the scene, each marked with the time it appeared. This way, the activity from several hours of footage can be viewed in a clip of a few minutes, linked to the original clips.
How does it differ from fast-forwarding the recording?
Fast-forwarding shows everything, including hours with no activity, and objects pass too quickly to follow. A synopsis keeps only what moved, at normal speed, and lets you filter by object type, zone or time range.
Does the system identify the people in the footage?
No. People are detected only as moving objects, with no facial analysis and no identification. The synopsis can be generated with people blurred, and access to the original recordings remains governed by your video surveillance policy.
Can the synopsis be used as evidence?
The synopsis is a review tool that helps you find the important moments faster. For any conclusion, or for handing footage to the authorities, the reference remains the original recording, kept according to your rules.
Does it work with my existing cameras and NVR?
It can, if the recordings can be exported or accessed through a video stream and come from fixed cameras. We check the export, image quality and angle of each camera in practice before confirming compatibility.
Are additional copies of the recordings created?
The synopsis and the object index are new data, so we agree together where they are stored, who sees them and how long they are kept, usually no longer than the original recordings. Processing can be done locally, without the images leaving your network.
Can’t find the answer? Write to us and we’ll reply directly.
Do you have hours of footage nobody watches?
Tell us how many cameras you have, how you keep your recordings and what situations you usually look for. We will tell you whether video synopsis is a good fit and what should be checked first.
Related solutions
- Video event detectionVisual rules that flag entries into restricted zones, missing protective equipment or fallen objects, for human review.
- Object detectionModels that recognise the objects that matter to your process in images and pass them on as structured data.
- Operational monitoringVisual tracking of process stages and of vehicle and pallet movement in yards and warehouses, so you can see where the flow gets stuck.