Follow people and vehicles across cameras

One camera tells you what happened in its view. With several cameras you want to know where things happened and how an entity moved between cameras:

Employee-017
09:12  Gate cam       Entered Gate North
09:14  Corridor cam   Entered Corridor B        moved in 40 s (expected 30 s-1m 30s)
09:16  Parking cam    Entered Parking Zone 4    moved in 2 min (expected 1-2 min)

CV-Scope does not follow anyone continuously between cameras. It connects two sightings when the same recognized person or plate appears on two cameras and your location model says the move is plausible, and it gives every move a confidence and its evidence. A likely move is shown as likely, never as a fact.

Step 1. Describe the site

1. Open Topology (sidebar, Locations › Topology).
2. Click + Site and name it, for example North Campus. In the properties on the right, open Layout: choose Floor plan, Schematic or Map, give the width and height and the unit. Use metres when the plan is to scale: travel times can then be estimated from distances.
3. Optional: Upload floor plan / map picture. CV-Scope stretches it over the layout; it loads no map tiles from the internet.
4. Select the site and add what is inside it with + Add inside: a building with floors (give each floor its own layout), areas, rooms, corridors, gates, parking areas - or a road network with roads and junctions.
5. Drag each place to where it is on the plan. Give areas a Width and Depth to draw them as rectangles.
6. Tick Entry point on gates and entrances, and Restricted on places people should only reach through an entry point.

GPS is optional. A node with a latitude and longitude on a Map layout with a geographic box is placed from its GPS.

Step 2. Place the cameras

1. Under Cameras not placed yet, select the place a camera is in (in the hierarchy), then click Place in … next to the camera.
2. Drag the camera on the plan and set Facing (0 = up, 90 = right), Field of view and Reach. The plan draws the view as a cone.
3. Under Scene zones → locations map each zone of the camera’s scene to the place it is, for example zone Gate A → Gate North, and click Save mapping. A camera covers the places its zones are mapped to.

Step 3. Connect the cameras

1. Click Link above the plan, choose CONNECTED_TO (or ADJACENT_TO, LEADS_TO), then click the first camera and the second camera.
2. In the link’s properties set Shortest travel and Longest travel, for example 30 and 90 s for a corridor walk.
3. Choose the place between them under Via, for example Corridor B. Moves then also record that the entity passed there (MOVED_THROUGH, marked as inferred from the model, not observed).
4. Tick One way for passages used in one direction only, Views overlap when both cameras see the same spot.
5. Repeat for every pair of neighbouring cameras. Links between places (for example Gate North LEADS_TO Corridor B) help where cameras are not directly linked.
6. Open the Check a move tab, choose two cameras, and read what the correlation will conclude: Directly connected, Connected through 2 links, Not connected, the way, and the expected travel time.

Step 4. Run the cameras

1. In Experiments, turn on Relationships for the experiment of every camera (guide 25). Rules are optional; place relationships and identity links need none.
2. With the licensed modules, turn on face or plate recognition for the experiments (guide 22).
3. Start the runs. Correlation across cameras runs by itself a few seconds after each batch of results and again when a run ends.

Step 5. Follow a journey

1. Open Journeys (sidebar, Relationships › Journeys) and choose a recognized person, a registered vehicle, a plate or a track.
2. The Timeline lists every sighting (camera, places entered and left) and between them each move: moved from Gate cam in 40 s (expected 30 s-1m 30s) · confirmed by rule 91 %.
3. Click a move. The Move card gives its provenance: from and to (camera and place), the time it took, the expected time and where it comes from, the identity confidence, the topology, sensors on the way, and the final confidence with its components. Supporting observations opens the evidence; the From / To buttons open the video of each camera at that moment.
4. The Location path draws the same journey on the plan with numbered steps; the Relationship graph below shows the entity and everything it is related to. Click a step, a marker or a graph edge: the other two views highlight the same thing.

Step 6. Explore the relationship graph

1. Open Relationship graph (sidebar, Relationships › Graph) and choose what to show: around an entity, a run, an experiment, a site or location, or a time range.
2. Read the drawing: circles are tracked objects, diamonds recognized identities, rectangles places, framed rectangles locations, camera icons cameras. Grey edges are observed facts, blue ones inferred by a rule, purple ones identity links, orange ones moves across cameras, dotted grey ones location context. Solid means confirmed or likely, dashed possible, dotted insufficient; S marks sensor-supported edges.
3. Drag the background to pan, use the wheel to zoom, drag a node to pin it. Double-click a node to expand it in place; select it and click Collapse to fold it again.
4. Use Show on the left to filter by kind of edge, relationship type, confidence and camera.
5. In the time bar choose Replay (how it grew) and press Play or ▶: the graph builds up in the order things happened. Snapshot at time T shows only what held at that moment; Time range a period.
6. Select a node and click Path from here, select another and click Path to here: the shortest chain of relationships between them is highlighted.
7. Click an edge to list the stored relationships it stands for; Evidence explains each one, with a jump to the video.

Step 7. Watch the site and read deviations

1. Open Site view (sidebar, Locations › Site view). Cameras with an active run are highlighted, a blue badge counts the objects in view, a red dot marks alerts of the last 24 hours, orange lines are recent moves between cameras (red when flagged).
2. Click a camera for its run, its Live feed and its experiments.
3. Under Where was it last seen? choose an entity: the place is ringed on the plan.
4. Alerts lists topology deviations: a move faster than possible (seen 10 s after Parking cam; the expected travel time is 60-125 s), a move between cameras the model does not connect, a one-way passage used backwards, a restricted place reached without passing an entry point, and a camera sequence unusual for this identity. They are also recorded as events of the run. They describe what was observed, never intent - a wrong identification is often the reason.
5. In Topology › Cross-camera settings you can change the tolerances, report fewer deviations, or tick Also link anonymous tracks by timing alone. After editing the location model, click Correlate again to apply it to recent runs.

The full reference is docs/locations.md.