How to reduce the time spent searching CCTV footage

Ask a security manager how long an incident investigation takes and you will usually get a number for the whole job. Ask how that time splits, and the answer is nearly always the same: a few minutes watching the relevant clip, and hours finding it.

That ratio is the thing worth attacking. It is also mostly a configuration and process problem rather than a technology one.

Why the search is slow

Four things account for most of it.

Nobody knows which camera covers the spot. On an estate of forty cameras named cam_14 and CH07, working out which one sees the loading apron takes someone who has been there three years.

The time is approximate. “Some time Tuesday afternoon” is a four-hour scrub. If the reporting person is a member of staff recalling it on Thursday, that is the usual precision.

The subject moves between cameras. Following one person across a site means finding them again on each camera, in sequence, by eye.

The clocks disagree. Access control, the recorder and the alarm panel drifting by even a few seconds turns cross-referencing into guesswork.

What actually helps

Attach the evidence at the moment of the event

The single biggest reduction comes from not searching at all. If an event is raised with the frame that caused it and the clip around it already attached, the retrieval step disappears for every event the system raised itself.

This is worth saying plainly because it changes what you should instrument. The footage you will need in six months is mostly footage of events that were detectable when they happened. Capturing it then costs nothing extra; finding it later costs hours.

Name cameras after places

Goods door — inside beats cam_14 every time. It sounds trivial. It removes a step from every single investigation anyone ever does on that estate, and it takes an afternoon.

Synchronise the clocks

Put every system — recorder, access control, analytics appliance — on the same time source. Cross-referencing two systems that disagree by eight seconds is the difference between a two-minute job and a thirty-minute one.

Narrow by person, not by time

Where person re-identification is available and the cameras support it, following the same person between cameras is faster than scrubbing each camera for the period. It is not magic: it depends on resolution, on the person staying visible, and on cameras whose views overlap or adjoin sensibly. Treat it as a way to narrow the search, not to replace it.

Keep the decision, not just the footage

If an event was reviewed and cleared, record who cleared it and when. Half of investigation time is spent re-establishing what was already known.

What to be sceptical of

“Just describe what you’re looking for.” Natural-language search over video is an active area and it demos extremely well. PARVEK does not offer it, and we would rather say so than let you assume it. Ask any vendor who claims it to run it live on your footage, not a curated set — the gap between those two is usually where the product is.

Claims about search time with no baseline. A vendor quoting a reduction percentage has measured it on someone else’s estate, with someone else’s camera naming and someone else’s clock discipline. The number does not transfer. Measure yours before and after.

Anything that needs footage uploaded. If the search runs in a cloud, the footage has to get there. For a thirty-camera site that is a bandwidth and governance conversation long before it is a feature conversation.

How to measure the improvement

Four numbers, taken before you change anything and again a month later:

• Median time from “an incident is reported” to “the relevant footage is on screen.” This is the headline number.
• Proportion of investigations where usable footage was found at all. Sometimes the honest answer is that the camera could not see it; you want to know how often.
• Time to assemble an evidence pack — the clip, the stills, the times, the context — in a form someone else can read.
• Number of events already carrying their own evidence, so no search was needed. This should climb as more of the site is instrumented.

A realistic sequence

1. Rename the cameras after the places they watch. One afternoon.
2. Put everything on one time source. One morning.
3. Instrument the two or three areas where incidents actually happen, so those events arrive with evidence attached.
4. Measure the four numbers above.

Steps one and two cost nothing and will shorten every investigation you ever do. Do them whether or not you buy anything.