Controlled pilot

A pilot should end in a decision, not a debate

In short

A controlled pilot establishes what video analytics actually does on your cameras, in your conditions. It runs in six phases — camera readiness, log-only baseline, ground truth, tuning, measurement, decision — against acceptance criteria agreed with you before it starts, so the result is evidence rather than an argument about whose numbers to believe.

Method

Six phases

Preparation

What a pilot needs from you

Most pilot delays come from missing camera information rather than from the analytics. Gathering these first shortens everything that follows.

Pilot inputs

Objective — what the pilot must establish, stated as a question with a yes or no answer
Location and the specific cameras in scope
Camera resolution, codec and delivered frame rate (measured, not configured)
Field of view, mounting height and downward angle per camera
Lighting across the operating schedule, including the worst realistic condition
Weather exposure for outdoor cameras
Expected object size at the furthest point of interest
Occlusion sources — racking, vegetation, parked vehicles, structure
The behaviour or event to detect, described operationally
Alert workflow: who responds, through which console, and how quickly
Operating schedule and when rules should be active
Ground-truth method agreed before the pilot starts
Acceptance metrics and their thresholds
Who reviews the results and makes the decision

What gets measured

Event recall against scripted ground truth
False alerts per camera-hour, counted by the operator
Alert latency from event to operator queue
Duplicate alerts per genuine event
System and stream availability across the pilot period
Operator handling time per alert
Search time where investigation workflows are in scope
Counting error against manual counts
System resource usage against the deployed hardware

No universal acceptance numbers are published on this site. A threshold only means something once it is attached to your use case, your cameras and your environment.

Frequently asked questions

What acceptance thresholds should we set?

Ones that reflect your environment and your use case. This site publishes no universal acceptance numbers, because a threshold that is achievable at a well-lit indoor doorway is not achievable on a 200-metre fence line at night. Thresholds are agreed with you, from your baseline, before the pilot starts.

How long should a pilot run?

Long enough to cover the variation that matters. Six weeks is typical: roughly two weeks log-only to establish a baseline, then tuning, then a measured period covering at least one poor-weather spell and one full operating cycle including weekends.

What happens if cameras fail the readiness review?

You are told which ones, and why, before anything is configured. Some will need re-aiming, some a lens change, and some are genuinely unsuitable for the rule intended. Discovering that at the start is the single biggest predictor of whether the deployment succeeds.

Do we have to buy anything to run a pilot?

Scope, commercial terms and any hardware requirement are agreed with Ayonix before a pilot starts. This site does not publish pricing or commercial terms, because none have been supplied for publication.