Live status for every desk, zone and doorway your cameras can see.

Sentinel Vision runs on your own hardware and turns the footage you already record into plain, timestamped statuses. People are counted, never identified.

The scenes are simulated. The outlines and statuses in them are model output. Open Raw model output to read the detections for the frame on screen.

What it reports

Three desk statuses. Each is a single rule applied to what the detector found in the frame.

A person typing at a desk, outlined in green and labelled Working.

Working

Someone is in the seat and there is no phone in their hands. It says the person is present. It says nothing about their output.

A person is detected inside the seat's outline.

An empty office chair outlined in red and labelled Absent, while the desks around it stay green.

Absent

The seat has been empty long enough to rule out a detector blink. It stays marked until someone sits back down.

No person in the seat for 0.75 seconds. Clears after 0.4 seconds back.

A person holding a phone, outlined in red and labelled Phone usage, with the phone itself boxed in amber.

Phone usage

A phone is in the person's hands. Phones lying on the desk are ignored, including the one that was just picked up.

A phone is detected inside the person's outline, away from where phones were resting, for 0.35 seconds.

How it works

Four steps from footage to a log you can review. Nothing is installed on the cameras and nothing is labelled by hand.

  1. Point it at footage

    Give it a recorded clip or a live camera feed from the cameras you already have.

  2. It maps the floor

    In the first second it finds everyone who is seated and marks each seat. For doors, shelves and driveways, you draw the zones.

  3. It watches for change

    A detector finds people, phones, vehicles and pets in every frame, and a pose model finds which way each head is turned. Fixed rules turn those detections into statuses.

  4. You review

    Each change lands in a timestamped log that you can export. Nothing is acted on automatically.

Where it works

One engine covers five kinds of floor. You draw a zone on the camera view, pick a rule and set its threshold.

Offices

Desk occupancy
Which seats are filled right now, and which have emptied.
Phone in hand
A phone being held, as opposed to one lying on the desk.
Walk-throughs
People crossing the floor are counted separately from people at desks.

Shop floors

Restricted areas
Someone stepping into a stockroom or staff-only doorway.
Dwell at a display
A person standing at a shelf or cabinet past a time you set.
Queue length
More people at the checkout than your limit, held for a moment.

Exam halls

Head direction
How far each candidate's head has turned from where it started, in degrees.
Whose desk
Which desk the head is pointing at, and for how long.
Phones
A phone at a desk raises a flag for an invigilator to check.

Industrial sites

Exclusion zones
A person inside the marked area around machinery.
Corridor occupancy
Too many people in a walkway at the same time.
Movement heatmaps In development
Where people spent time across the footage, without tracking who.

Homes

Front door
A person on the path to the door raises an alert within half a second.
Driveway
A vehicle pulling in is logged as a notice and raises no alarm.
Pets
Dogs and cats are recognised and left out of the alerts.

Under the hood

The models behind the demos are on this page. Pick one and run it on a sample frame, a photo of your own or your camera. It downloads to your browser and runs on your device, and nothing you give it is uploaded.

The model is 10 MB and loads when you press the button.

These are the nano models, chosen so they load quickly. The demo clips above were analysed with the large ones, which find smaller objects such as phones and place keypoints more steadily.

Models
YOLO26 from Ultralytics. Detection comes in three sizes: nano (9.9 MB) for live use and this page, small (38 MB) when accuracy matters more than speed, large for offline analysis. YOLO26 pose adds 17 body keypoints per person.
Runtime
ONNX Runtime. Native on your PC, on the graphics card through DirectX 12 when there is one, and WebAssembly inside this page.
Speed, measured
68 ms a frame for nano and 208 ms for small at 640 px, on a 2-core Xeon cloud CPU with no GPU. Large at 1280 px takes about 2 seconds a frame on the same machine.
The demo run
964 frames analysed with the large models. 7,068 detections kept above 25% confidence, 41,801 body keypoints and 13 status changes logged.
Rules
Zone entry, dwell, occupancy limits, seat absence, phone in hand, head turn and gaze target. 41 automated tests cover them.
Check it
Download the raw detections behind each clip: office, shop floor, front yard, exam hall.

The model files

These are ordinary ONNX files, the ones the Windows app loads. Download one, check its SHA-256 and open it in an ONNX viewer to see the network.

FileFindsWeightsSizeReturns per frameSHA-256 starts
yolo26n.onnxObjects, 80 classes2.4 million10.4 MB300 rows of box, confidence, classa772bebdb74762b8
yolo26s.onnxObjects, 80 classes9.5 million38.8 MB300 rows of box, confidence, class2a0b6ac6d7d441fb
yolo26n-pose.onnxPeople, 17 body keypoints each2.9 million12.5 MB300 rows of box, confidence, 17 pointsb690600113b28104
yolo26s-pose.onnxPeople, 17 body keypoints each10.4 million42.3 MB300 rows of box, confidence, 17 points209b1d2b54efc281

Input is one RGB image scaled to 0 to 1, at any width and height that are multiples of 32. Opset 17, exported from Ultralytics YOLO26 with its end-to-end head, so the rows are final detections and need no separate filtering step. The browser demo above uses 640-pixel exports of the two nano models. Full checksums are in the documentation.

The zone file for the shop-floor clip, as written
{ "name": "Aisle 1", "kind": "dwell", "seconds": 5,
  "poly": [[300,470],[470,455],[600,680],[330,720]] },

{ "name": "Checkout queue", "kind": "occupancy",
  "limit": 3, "hold": 0.6,
  "poly": [[980,760],[1215,700],[1920,1040],
           [1920,1100],[980,1100]] },

{ "name": "Stockroom door", "kind": "entry",
  "seconds": 0.35, "latch": 2.5,
  "poly": [[1352,235],[1622,262],[1612,640],[1350,625]] }
The run that produced that clip
$ python detect.py shop.mp4 detections.json --classes 0
done 241 frames in 504s

$ python zones_analyze.py detections.json scene.json states.json
{'t': 3.83, 'zone': 'Stockroom door', 'level': 'alert', 'text': 'ENTRY'}
{'t': 5.0,  'zone': 'Aisle 1', 'level': 'alert', 'text': 'DWELL 5.0s'}
{'t': 7.83, 'zone': 'Stockroom door', 'level': 'ok', 'text': 'cleared'}
{'t': 8.21, 'zone': 'Checkout queue', 'level': 'alert', 'text': '4 of 3 · OVER'}

$ python -m unittest discover -s tests
Ran 41 tests in 0.115s
OK
The exam-hall thresholds, as written in exam_analyze.py
YAW_GATE = 38.0        # degrees off baseline before a look leaves own paper
TARGET_TOL = 30.0      # most degrees allowed between head direction and a desk
TARGET_RANGE = 7.0     # a desk must be within this many shoulder-widths
SWITCH_MARGIN = 0.25   # cost gap another desk needs to take over as target
LOOK_HOLD = 0.35       # seconds a look must last before it scores
LOOK_RATE = 42.0       # risk per second while looking at another desk
DECAY_RATE = 7.0       # risk per second while not
PHONE_RISK = 85.0      # a phone at the desk lifts the score to this
REVIEW_AT, REVIEW_OFF = 25.0, 12.0
FLAG_AT, FLAG_OFF = 60.0, 45.0
The run that produced the exam-hall clip
$ python detect_pose.py exam.mp4 detections.json
done 241 frames in 970s

$ python exam_analyze.py detections.json states.json
10 seats; 3 status changes
   2.00s  S8  NORMAL -> REVIEW  (looking at desk S9, risk 26)
   2.83s  S8  REVIEW -> FLAGGED  (looking at desk S9, risk 61)
   4.71s  S9  NORMAL -> FLAGGED  (phone at the desk, risk 85)

Windows app

The same models and rules as a desktop program. It plays your webcam or any video and annotates it as it runs, on your own PC.

The office program on a raw clip of six desks. Four are outlined in green and labelled Working, an empty seat is red and labelled Absent, and a person holding a phone is labelled Phone usage. A box in the corner gives the model, input size, processor and time per frame.
SentinelOffice
The shop program: three zones drawn over a supermarket, with the checkout queue zone in red.
SentinelShop
The home program: a driveway zone with a parked car in amber and a front path zone with a person in red.
SentinelHome
The exam program: candidates outlined with head keypoints and desk zones, one of them flagged in red for looking at another desk.
SentinelExam

One frame from each program, running on a clip with nothing drawn in it. The box in the corner is the program's own readout: the model, its input size, GPU or CPU, and how long the frame took.

Download for Windows
Version
0.1.0, released 13 November 2026
Download
210 MB zip. Unzip it and double-click a program. Nothing is installed.
Needs
Windows 10 or 11, 64-bit. It runs on the CPU and uses a DirectX 12 graphics card when it can.
Inside
Four programs (Office, Shop, Home, Exam), the four ONNX models and a raw sample clip for each program.
SHA-256
4080591841529eb2948d3062764b212ad11abc34d08115569c29affcf4fc27ed

Every release is built on GitHub's Windows machines. Before it is published, each program is started on its sample clip and has to report the expected events. See the build log or read the source.

The documentation covers the keys, the rules and their thresholds, and the model files.

In version 0.1.0

  • Four programs: Office, Shop, Home and Exam
  • Webcam and video-file input
  • Zones you draw, with entry, dwell and occupancy rules
  • Event log as CSV, screenshots and annotated video
  • Runs on the graphics card through DirectX 12, or on the CPU

In development

  • A review screen: jump to an event, mark it checked, add a note
  • Movement heatmaps
  • IP cameras over RTSP
  • A retention timer for logs and clips
  • One program with an installer

Deployment

Sentinel Vision is software you run on your own machine. There is no cloud service behind it and no account to create.

Runs on
Your own computer or server. Footage is processed there and is not uploaded.
Identity
None. There is no face recognition, no names and no demographic guesses. Head direction comes from five keypoints on the head. Each person gets a number for the session, and it is dropped when they leave the frame.
Detects
People, phones, vehicles, dogs and cats. Where head direction matters, 17 body keypoints per person.
Input
Recorded video files and webcams. In development IP cameras over RTSP.
Output
Timestamped events as CSV and JSON, and annotated video. In development Movement heatmaps.
Retention
Logs and saved clips stay on your machine until you delete them. In development A timer that deletes them after 24 hours unless you change it.
Decisions
None are automated. Every status is a prompt for a person to look.
Rules
Fixed and inspectable. Each alert states the zone, the threshold and the time it fired.

Responsible use

Sentinel Vision reports what a camera can see: presence, position and objects. It cannot see intent or effort.

"Working" means someone is at their desk. "Phone usage" means a phone is in their hands. Neither says why, and neither should be acted on without a person looking at the footage first.

In an exam hall, a flag means a head stayed turned toward another desk, or a phone appeared at a desk. It gives an invigilator a timestamp to check. Whether anything improper happened is for them to decide.

Tell the people in frame that the system is running and what it records, and follow the workplace-monitoring and privacy rules that apply where you operate.

Request a demo

Tell us what you want to watch and where. We will reply to arrange a walkthrough.

We use these details only to reply to you. Privacy