Giraffe AI Overview

We often get asked about the capabilities and features of our AI offerings so we thought we would create a page all about it.

Our focus

At Giraffe our focus is to detect and deter would be intruders at the perimeter.

We say this because we believe it is important to be honest about what we are good at.

Detect

If a person or vehicle comes into the field of view of a camera at night, our aim is detect this person as soon as possible. We will then send an alert to the keyholder or alarm receiving centre as quickly as possible.

This is the primary objective.

Deter

If we detect a person, we need to try to deter them from coming any closer. We call this feature Active Threat Deterence (ATD) combined with Virtual Security Guard (VSG).

ATD is the first line of response - typically switch on strobe lights, flood lights or a siren sound. It's designed to startle the intruder as quickly as possible and make them aware that the system is working. It buys us some time.

VSG is the second line of response - it delivers a customised audio talkdown directly to the intruder. If they are wearing a yellow jacket and pushing a wheel barrow, VSG will say that when addressing them.

Perimeter focussed

We say at the perimeter because we understand our shortcomings. Giraffe AI is designed to work on a locked, alarmed site where there should be no human activity

We do not have features such as loitering detection or object counting and we have no immediate plans to build them.

If you are interested in features such as loitering, you can still use them with Giraffe! These type of analytics can be run on the cameras or third party applications which can forward alarms to Giraffe.

Capabilities

This section explans a bit about how 'good' Giraffe AI is, including it's downsides.

Detection Distance

Giraffe AI is really sensitive. We will often detect vehicles just based on their headlights.

Likewise, we can often detect people based on just a faint outline.

The below table gives an idea of the sort of detection distances we are aiming for. Of course this varies a bit depending on the exact camera model and environmental conditions.

Detection Distance	Width of view	Probability of detection
Less than 30m	Less than 30m	High
30m to 50m	Less than 15m	High
50m+

Classification

Accuracy vs False Alarms

Edge vs Cloud

False positives and misses