Computer vision
Sheet 12/12 · High · 8–24 weeks (audit and pilot first)Counting, grading and gate checks are done by eye; incidents are found on the CCTV afterwards, if at all.
- Tobacco, produce or ore is graded or sorted by eye, and the grade depends on who is on shift.
- Vehicles, people and stock pass gates on paper logs; disputes are settled by watching CCTV for hours.
- Safety rules (helmets, exclusion zones, PPE) are checked by supervisors when they are present.
- Stock counts on shelves or in yards take a day and are wrong by the next morning.
- Cameras exist but only for after-the-fact review; nobody watches forty screens.
- Count, measure and classify what a camera can see consistently — vehicles at a gate, items on a line, plants in a row, grades of a leaf — and raise an alert or a record.
- Detect defined events (a person in a zone, a truck without a booking, a missing helmet) and notify a person in seconds.
- Read plates, labels, tags and meters into your systems.
- Turn existing CCTV into a source of counts and events rather than an archive.
- See through bad lighting, dust, rain on the lens or a camera pointed at the wrong angle; the site conditions decide most of the accuracy.
- Work without power and a network at the camera — batteries, solar and local processing are part of the design here.
- Replace the inspector's final call on high-value grading; it makes the first pass consistent and flags the doubtful.
- Be bought as software only — every vision project is cameras, mounts, cabling, power and a model.
| Effort band | High · 8–24 weeks after audit and pilot |
|---|---|
| Data you need | Site drawings; what to detect and how often; existing camera inventory; sample footage; grading standards |
| Integration points | Cameras (IP/RTSP), edge devices, weighbridge/access control, ERP, alarm and messaging |
| People | Site engineer; the supervisor who receives alerts; a grading expert for labelling |
| Measure | Detection rate; false alarms per day; time to alert; count accuracy against manual; grade agreement with expert |
| Zimbabwe constraint | Power and network at the camera (solar, batteries, edge devices); dust and heat; footage of staff is personal data under the Cyber and Data Protection Act |
| Code | Shape | Fit | Why |
|---|---|---|---|
| S-01 | Audit | Primary | A site survey — cameras, angles, lighting, power, network, what exactly must be detected — decides feasibility before a cent is spent on models. |
| S-02 | Pilot | Primary | One camera, one line or gate, four to eight weeks, with a person checking every event. Accuracy on your site is the only accuracy that matters. |
| S-03 | Implementation | Secondary | Camera roll-out, edge processing, the event rules and the alert routes. |
| S-04 | Integration | Secondary | Weighbridge, access control, ERP, grading records and alarm systems are the endpoints. |
| S-06 | Managed support | Later | Lenses get dirty, seasons change the light, models drift; a monthly accuracy report is essential. |
- Which cameras, at which resolution, angle and lighting, and who supplies, mounts and powers them?
- Where does the model run — on a device at the site, in a Zimbabwe data centre, or abroad — and what happens when the link drops?
- On our site, in a pilot, what detection and false-alarm rates did you measure, and over how many days and hours of the day?
- How is a false alarm or a missed event reviewed, and how does the system improve from it?
- Who owns the footage, where is it stored, for how long, and how do we handle staff and visitor privacy?
- What is the total cost per camera per year including hardware, power, network, licence and support?
The full bank of 60 questions, grouped by risk, is at /vendor-questions/.
It is a construction project with a model in it
Every honest vision quote in Zimbabwe has more lines for cameras, mounts, cabling, solar, batteries and an edge computer than for the model. Sites lose grid power and network; a camera that must stream to a cloud abroad to count trucks will stop counting when the fibre does. Local processing at the site with periodic sync is the usual answer, and it changes the cost, the vendor and the maintenance.
Site-conditions sheet (bring this to the audit)
| Condition | Why it matters | What to record |
|---|---|---|
| Lighting by hour | Models trained at noon fail at dusk | Photos at 06:00, 12:00, 18:00 |
| Dust, rain, spray | Lens fouling drives false alarms | Cleaning access and frequency |
| Power at camera | Outages stop detection | Grid, solar, battery hours |
| Network to camera | Streaming vs edge | Bandwidth, dropouts per day |
| Field of view | Occlusion and angle | Mounting height and distance |
| Privacy | Staff and visitors on footage | Signage, retention, access list |
Where it is already used here
Vision is deployed in Zimbabwe for industrial monitoring, gate and fleet visibility and agricultural grading — tobacco grading vision is a listed service from a Harare provider in our directory. The category authority for computer-vision systems in this network is the sister company Eigenstate Systems (external site); this sheet covers only the buying questions. For access-control and mine-security designs, ask providers for site references and go and look at them; a vision system is judged standing next to the camera.