AI video analytics on the video wall server GPU

Detect and track people and vehicles

A neural network detects people and vehicles in RTSP camera streams, and a tracker follows each object from frame to frame.

Processing runs on the same NVIDIA GPU that composes the wall. For each camera you choose the mode: always on, on motion or off.

Twelve cameras, one top-down map

On the product stand, twelve warehouse cameras go through detection in one pipeline on the server GPU. A multi-view tracker joins the detections into one track per person.

Each person moves across a top-down plan of the site as a numbered marker with a trail, in real time. The camera mosaic and the map are ordinary sources: they sit on the wall next to everything else.

Draw rules on the live frame

Pick a camera and draw on its live picture: a line to cross, a zone to enter, an area where an object must not be left.

Bind the rule to a zone and an object class, then set its schedule and re-arm time.

Every incident gets a verdict

A rule hit opens an incident in the operator queue, sorted by severity and age.

The operator closes it with one of three verdicts: confirmed, false alarm, or other with a comment. Keyboard shortcuts keep the work fast.

The overview shows totals, confirmed and false shares and response times for the last day, week or month.

A site plan with zones and camera sectors

Upload a site plan, mark the zones and place the cameras with their fields of view.

An incident appears on the plan in its zone. Links between neighbouring zones tell the module which cameras to show next.

The wall shows the incident by itself

An incident raises an alarm in Craft Wall. The wall switches to the incident scene and returns when the alarm clears.

• How alarm scenes work →

What the module does not do

The module does not recognise faces or number plates, and it does not replace your video management system or recorder.

Use it to bring attention to the wall. Keep recording and forensic search in the video management system.

Check before sizing analytics

Camera count alone does not size analytics. Collect these facts first.

• Resolution and frame rate of every analysed camera.
• Mounting height and how large a person appears in the frame; small, distant targets are harder to detect.
• Lighting at night and in bad weather.
• Which rules each camera needs.

• Size the server →
• AI on control room video walls →
• Test analytics on your cameras →
• All product functions →