Examples, route choice
Which way do people go after the entrance?
A route is a path through gates. Here, people who pass the entrance gate in the middle of a corridor leave by an exit on the left or on the right. CV-Scope records each of them as exactly one route, or as Unknown, Abandoned or Lost track, and never forces a route when the evidence is incomplete.
Switch the video off to see only what the run stored.
The scene
The repository’s sample scene, samples/scenes/corridor_route_choice.json. Coordinates are fractions of the frame, so the scene keeps working at any resolution.
| Object | Type | Position (x, y) | Records |
|---|---|---|---|
| Entrance | gate | (0.5, 0.45) to (0.5, 1) | crossings, both directions |
| Exit Left | gate | (0.15, 0.45) to (0.15, 1) | crossings, both directions |
| Exit Right | gate | (0.85, 0.45) to (0.85, 1) | crossings, both directions |
| Centre floor | zone | (0.3, 0.5) to (0.7, 0.5) to (0.7, 0.98) to (0.3, 0.98) | entry, exit, occupancy, dwell |
| Route | Starts at | Ends at | Gives up after |
|---|---|---|---|
| Route A | Entrance | Exit Left | 30 s |
| Route B | Entrance | Exit Right | 30 s |
What the run recorded
From the run’s stored summary: YOLO11m on cuda:0, ByteTrack, every frame of the 50 s clip.
- Passed the entrance
- 3
- Completed a route
- 3
- Unknown, abandoned or lost
- 0
- Mean decision time
- 1.03 s
- Route A 2
- Route B 1
| Route | Count | Share of valid | Mean duration | Mean speed |
|---|---|---|---|---|
| Route A | 2 | 66.7 % | 2.71 s | 0.247 frame widths/s |
| Route B | 1 | 33.3 % | 2.25 s | 0.184 frame widths/s |
| Mean decision time (entrance to the next line or zone) | 1.03 s |
|---|---|
| Route switches (a checkpoint of another route first) | 0 |
| Same route as the previous person | 1 of 2 |
Three people are far too few for a conclusion. The sample shows the method. The analysis page says the same: these are observed movements, and correlations with crowding or previous choices describe the recorded scene, not the reasons for a choice.
Change one condition
Guide 14 duplicates the experiment and changes only the detector. These are both runs, on the same scene version and clip.
| Measure | Recommended detector (YOLO11m on cuda:0) | Detector YOLO11n (YOLO11n on cuda:0) |
|---|---|---|
| Events | 27 | 29 |
| Crossings | 10 | 10 |
| Zone entries | 7 | 8 |
| Zone exits | 7 | 8 |
| Route events | 3 | 3 |
| Anonymous tracks | 7 | 7 |
| Routes (share of valid) | Route A 2 (66.7 %), Route B 1 (33.3 %) | Route A 2 (66.7 %), Route B 1 (33.3 %) |
| Mean decision time | 1.03 s | 0.92 s |
| Crossings per gate | Entrance 3, Exit Left 4, Exit Right 3 | Entrance 3, Exit Left 4, Exit Right 3 |
The route result is the same. The smaller detector recorded one extra zone visit of 0.25 s by track 5, which entered again 0.75 s later, and a slightly different decision time. Small differences like these are why a comparison changes one condition at a time.
Reproduce it
- Install CV-Scope; the setup downloads the corridor clip.
- Follow guide 11: import the sample scene and start a run.
- Duplicate the experiment and change the detector, as in guide 14.
- Check the result against the video with guide 17.
The recommended detector depends on the hardware; on a machine without a GPU it is YOLO11n, the second column above. Data exported 2026-09-29 from CV-Scope at commit 14ae109.