What types of metadata do video analytics generate?
Video analytics metadata is structured data that describes what appears in the video rather than the video itself. It typically includes object classifications, bounding box coordinates, timestamps, tracking identifiers, behavioural events, and attributes such as colour, direction or licence plate text, all linked to a camera and a point in time. The exact content depends on the analytics engine and the VMS. Common examples include:- Object detection data: person, vehicle or other class labels, with bounding boxes and confidence scores for each frame or sampled frame.
- Tracking data: persistent track IDs, paths and dwell times as objects move through a scene.
- Event data: line crossing, intrusion, loitering, abandoned object or crowd density alerts.
- Recognition data: ANPR plate reads, face match results and associated images or thumbnails.
- Counting and heat map data: people counting, occupancy and retail analytics outputs.
How much storage does analytics metadata actually need?
Analytics metadata usually needs far less storage than the video it describes, but the amount varies widely. A quiet corridor generates little, while a busy transport hub with many tracked objects per frame can generate substantial, continuous metadata. There is no universal figure, so it should be estimated per project from the analytics configuration. Four main factors drive the volume:- Scene activity: more objects per frame means more records.
- Metadata frequency: per-frame output produces much more data than sampled or event-only output.
- Attribute richness: detailed attributes, thumbnails and snapshots add to each record.
- Camera count and retention period: both multiply the total.
Where should metadata be stored: with video or separately?
In most analytics-heavy systems, metadata should be stored separately from the primary video archive, typically in a database on faster storage. Video is written as large sequential streams, whereas metadata involves many small, indexed reads and writes. Mixing the two on the same volume can degrade performance and increase fragmentation. The workloads differ in important ways:- Video archive: sustained sequential writes, suited to high-capacity hard drives in RAID.
- Metadata database: small random reads and writes plus search queries, which benefit from SSD or cached storage and adequate memory.
Does analytics processing change video storage requirements too?
Analytics processing can indirectly change video storage requirements, even though it does not alter the video itself. The most common effects are a shift from continuous to event-based recording, higher bitrates to support accurate detection, and extra streams or retained clips generated for analytics purposes. Consider these effects when sizing:- Recording mode: analytics-triggered recording can reduce stored video in quiet scenes, but reliable detection often needs continuous recording, so savings should not be assumed.
- Resolution and frame rate: accurate analytics may require higher resolution or frame rates than a basic monitoring requirement, increasing bitrate.
- Additional streams: some systems record a second stream for analytics or retain clips of detected events.
- Compression settings: aggressive compression or smart codecs can reduce accuracy in some cases, so test before committing.
How long should analytics metadata be retained?
Analytics metadata should be retained for as long as it serves an operational, investigative or compliance purpose, which is often shorter than the retention period for the underlying video. There is no single correct period. The right duration depends on the use case, applicable legal obligations and the cost of keeping searchable data online. Practical points to consider:- Operational use: counting and occupancy data may be valuable for months or years as aggregated trends, even when the original video is gone.
- Investigative use: metadata that supports forensic search is most useful while the matching video still exists. Keeping metadata longer than its video can leave orphaned records that point to nothing.
- Data protection: metadata that identifies individuals, such as face recognition or plate reads, may fall under UK GDPR. Retention should follow your documented purpose and policy.
- Tiering: keep recent metadata on fast storage and move or summarise older records to reduce cost.
How do you plan server storage for analytics-heavy systems?
Plan analytics-heavy surveillance storage by sizing video and metadata separately, then validating the combined workload against server throughput. Calculate video capacity from bitrate, camera count and retention, add a metadata estimate with headroom, and design redundancy and performance for the complete architecture rather than for drive capacity alone. A practical sequence:- Define the analytics requirements: which analytics run, on which cameras, and whether processing is at the edge, on a server or on a dedicated analytics platform.
- Size the video archive: use camera bitrates, recording mode and retention days.
- Estimate metadata: use documented assumptions for record rate, size and retention, and allow growth headroom.
- Separate the workloads: place metadata and databases on appropriate fast storage, and keep video on high-capacity arrays.
- Add resilience: RAID protects against drive failure but is not a backup, so consider failover and recovery for the metadata database as well.
- Validate before deployment: test the design under realistic load.
Frequently Asked Questions
How do I calculate a basic metadata storage estimate for my project?
Start by documenting four assumptions: average records per second per camera, average record size in bytes, number of cameras, and retention days. Multiply them together (records per second × record size × 86,400 seconds × cameras × retention days) to get a raw figure, then add headroom for database indexes and growth. Count any stored snapshots or clips separately, and validate the result against your analytics vendor’s documentation, since record sizes vary between platforms.
What happens if my metadata database fills up or becomes corrupted?
Depending on the VMS, a full or corrupted metadata database can stop forensic search, event logging or analytics indexing, even if video continues to record. Reduce the risk by monitoring free space with alerts, keeping the database on a dedicated volume with headroom, and including it in your backup or failover plan. RAID protects against drive failure but not against corruption or accidental deletion, so scheduled database backups are still worthwhile.
Can I move an existing metadata database to separate storage after the system is live?
Often yes, but it depends on the VMS and analytics platform, as some allow the database path to be changed in configuration while others require a migration or vendor support. Plan a maintenance window, take a verified backup first, and test search and playback afterwards. Check your vendor’s documentation before starting, and where possible decide on database location at design stage to avoid migrating later.
Should I use SSDs for the metadata database, and how much do I need?
SSDs or cached storage are generally recommended because metadata involves many small random reads and writes plus search queries, which suit flash better than large hard drive arrays. Size the SSD volume from your metadata estimate plus headroom for indexes and growth, and choose drives with suitable endurance ratings for continuous writes. Adequate server memory also improves search performance, so consider RAM alongside storage.
What are the most common mistakes when planning storage for analytics-heavy systems?
Common mistakes include sizing only for video capacity, assuming analytics-triggered recording will always save space, placing metadata on the same volume as the video archive, and ignoring snapshot or clip storage. Another frequent error is failing to test under realistic load, so performance problems only appear after deployment. Documenting assumptions and validating the design with a benchmarking tool before installation helps avoid these issues.
How do I stop metadata growing out of control once the system is running?
Review analytics settings regularly, as high-frequency per-frame output, unnecessary attributes and analytics running on low-value cameras all inflate data volumes. Switch to sampled or event-only output where per-frame detail is not needed, and apply retention policies that align metadata with video retention. For long-term trend data such as counting or occupancy, summarise or aggregate older records instead of keeping every raw entry.
How do I get started if I am adding analytics to an existing CCTV installation?
Begin by auditing the current setup: camera resolutions and bitrates, recording server load, available storage and where the VMS keeps its databases. Then define which analytics you need on which cameras, estimate metadata and any extra video demand, and check that the server has spare compute (including GPU if required) and fast storage for the database. Running a pilot on a small number of cameras lets you measure real metadata growth before scaling up.
Related Articles
- How can system integrators estimate the total cost of ownership of a CCTV SAN?
- How does RAID protect footage on a surveillance server?
- Can a standard IT server handle HD surveillance workloads?
- What RAID level is best for HD surveillance servers?
- What is the difference between RAID 5 and RAID 6 for surveillance?