Water, gas, and industrial meters can be read remotely without touching the meter — by mounting a low-power IoT camera in front of the dial. NE101 can send scheduled images directly to your system; when numeric readings are needed, NeoMind’s OCR extension can run on an NG4500 or suitable PC/Linux host and deliver the result to your SCADA or BMS without a cloud dependency when deployed on-site.
Most legacy water, gas, and industrial meters have no data port, no pulse output, and no wireless interface. They were designed to be read by a person standing in front of them. Replacing them with smart meters solves the data problem — but at a cost that is often prohibitive at scale: each meter swap requires service interruption, possible regulatory sign-off, and hardware costs of $150–300+ per unit. For a facility manager with 200 sub-meters, or a utility running tens of thousands of meters in the field, full replacement is a multi-year capital project, not a quick integration task.
The camera-based approach addresses this differently: the mechanical meter stays in place and continues operating normally. A compact, battery-powered imaging camera is mounted in front of the dial face. It captures a photo on a configured schedule, and transmits the image to your system. If you need a numerical reading rather than the photo itself, NeoMind can run OCR on a host and forward the result to your data system. No pipework. No service interruption. No meter certification re-process.
NE101 handles image capture and transmission on its own. When a project needs automatic digit recognition, the OCR function can be added on a separate host. This guide covers both routes: direct image delivery and NeoMind OCR on an NG4500 or suitable PC/Linux host, which can run on-site without a third-party cloud.
The NE101’s job is clear: capture a sharp image and deliver it reliably to the destination you choose. Image-only projects can stop there. When you need digit recognition, NeoMind adds a separate OCR layer that can be updated without changing the camera installation.
NeoEyes NE101 — compact form factor designed for tight meter enclosures
Typical deployment scenario: underground water meter pit, battery-powered camera with LTE uplink
A camera-based meter reading system starts with capture and image delivery. The OCR and structured-data steps below apply when your project needs automatic readings rather than the photos alone.
The NE101 wakes from deep sleep on a time schedule (e.g., every 4–6 hours) or on an external signal (reed switch, PIR, or MQTT command). Scheduled low-frequency capture can support multi-year battery operation; size the battery for the actual capture frequency and network conditions.
The 5MP OV5640 sensor captures a full-resolution image of the meter face. Lens selection matters: a narrow-FOV module (≈60°) suits close-mounted meters; a wider module (≈120°) works where there’s more stand-off distance. The NE101’s modular lens design allows field swaps without depot return.
The image is compressed and pushed over Wi-Fi 4, or — for meters without reliable WiFi coverage (basements, outdoor pits, remote sites) — optional LTE Cat.1 or WiFi HaLow. Upload time depends on image size and link quality, after which the device returns to deep sleep.
For automatic numerical readings, configure NE101 to publish images to the MQTT broker used by NeoMind on a NeoEdge NG4500 or suitable PC/Linux host. NeoMind’s OCR extension extracts the reading and can run on-premise without a third-party cloud service. If you only need photos, the same NE101 reporting settings can target a reachable broker in your own system or cloud; no OCR host is needed.
For image-only use, NE101 can publish via MQTT (a lightweight IoT pub/sub messaging protocol) to a broker used by your system or cloud, without NeoMind. If OCR is required, NE101 sends the image to NeoMind’s configured broker; after recognition and validation, NeoMind delivers the structured reading to SCADA, BMS, or a dashboard through OpenAPI, Data Push, or Webhook integration — see the CamThink Wiki for device-side MQTT reporting.
NE101 can deliver meter images directly to your MQTT broker when no OCR is needed. For automated readings, NeoMind handles OCR and data aggregation on an NG4500 or suitable PC/Linux host.
OCR recognition accuracy depends on three factors you control: image sharpness (lens alignment, standoff distance), lighting conditions (the NE101 has a built-in fill light configurable for automatic, scheduled, or always-on modes), and model quality. On clean, high-contrast mechanical meter faces under consistent lighting, a validated digit recognition model can achieve high read accuracy. On worn, reflective, or partially obscured dials, accuracy drops and supplemental lighting or a custom-trained model is needed. CamThink’s algorithm customisation service can train and validate a model specifically for your meter type — contact us if you need a turnkey inference layer rather than building it yourself.
For the NE101 hardware side: install batteries, connect to the device’s built-in Wi-Fi AP (SSID: NE101_XXXXXX, no password), open 192.168.1.1 in a browser, configure your MQTT broker address and capture schedule — typically under 15 minutes. The first test image uploads immediately on button press. The OCR layer (extracting the numeric value from the image) is separate: if you are connecting to an existing server-side OCR pipeline, add-on time depends on the integration. For a new deployment, install NeoMind and its OCR extension on the host, then validate readings against your own meter images. See the NE101 Quick Start guide for the device setup flow.
The NE101 ships with pre-built open-source firmware including MQTT and HTTP support — no custom firmware needed for day-one image upload. Trigger logic, image compression, and wake schedule are configurable via Web UI or the open SDK. On the NeoMind host, install the paddle-ocr-v6 OCR extension and connect the camera’s image reports to the platform. The NeoMind OCR use case documents the image-ingestion and recognition workflow. If your meter faces need a customised model, validate it with representative images before deployment.
A bare ESP32-S3 camera board can capture images and upload them over WiFi — the core function is achievable with commodity hardware. The practical question for any integrator deploying across 10, 50, or 200+ meter points is not whether it’s technically possible, but what the total cost looks like when you factor in enclosure engineering, field failure rate, battery longevity, and ongoing maintenance overhead.
| Dimension | DIY ESP32 Camera Build | NeoEyes NE101 |
|---|---|---|
| Time to first image | Hours to days (PCB assembly, firmware flash, case fabrication) | Minutes (pre-flashed firmware, Web UI configuration) |
| Weather protection | DIY enclosure — waterproofing quality varies | IP67-rated — dust-tight, built for -20°C to 50°C environments |
| Battery life | Weeks to months depending on hardware selection and sleep configuration | 2–3 years estimated at low capture rates; use the battery calculator for your schedule and network mode |
| Connectivity options | WiFi (typically); LTE or sub-GHz requires additional module work | WiFi 4 · BT 5.0 · LTE Cat.1 · WiFi HaLow — modular, field-swappable |
| Lens flexibility | Fixed to chosen module; lens swap requires new hardware | Modular, interchangeable lenses — 3D-printable mounts available |
| Scale management | No fleet management — each unit configured manually | Compatible with NeoMind edge AI platform for device fleet management |
| Hardware cost (per unit) | $20–60 (BOM only; excludes engineering time, enclosure, accessories) | $69.90–$112.00 — fully assembled with accessories |
| Open source / hackable | Fully open | Open SDK + GitHub — firmware is modifiable |
The DIY path is not wrong — it is a legitimate starting point for developers learning the stack or validating a concept for a single site. The packaged hardware path becomes more cost-effective once you account for engineering hours, enclosure sourcing, and the ongoing cost of a component that fails in the field because moisture got in. At 10+ units, the per-unit cost delta typically inverts.
NE101 can deliver meter photos directly to your system, whether you have a handful of points or a larger fleet. Add NeoMind when numerical readings are needed. NE301 is an option when the application also calls for its supported on-device vision functions; meter-reading OCR can use the same NeoMind host workflow after validating the camera’s reporting path.
The majority of legacy meters are not in server rooms. They are in basement utility closets with no power outlet, in outdoor street-side cabinets, in underground pits, and at agricultural pump stations kilometres from the nearest router. Any camera-based solution that requires mains power or reliable WiFi coverage will fail on this hardware before it starts.
In scheduled mode, the NE101 wakes, captures, transmits, and returns to deep sleep. The 0.8 W standby figure in the product overview describes a PIR/radar outdoor-monitoring example, not the battery budget for scheduled meter reads. At a capture frequency of once every 4–6 hours (4–6 reads per day), multi-year operation is possible under suitable conditions. Use the CamThink Battery Life Calculator for the selected battery, network mode, and capture schedule; actual life also depends on signal quality, temperature, and image upload time.
The NE101’s modular communication design supports an optional LTE Cat.1 module. LTE Cat.1 provides adequate throughput for image upload, with payload size depending on image settings. It operates on existing 4G infrastructure. For meters in outdoor utility cabinets or rural water supply networks with cellular coverage, LTE Cat.1 removes the WiFi dependency entirely.
The NE101 includes a built-in fill light controllable directly from the Web UI — no external LED or GPIO wiring required. Four modes are available: automatic (light activates when ambient lux falls below a configurable threshold), scheduled (active only during a defined time window), always-on, and always-off. For underground meter pits or enclosed cabinets with no ambient light, automatic mode with a low lux threshold is a practical configuration: the light fires in sync with each capture event and shuts off immediately after. Include its energy use in the per-capture battery budget. Light intensity is adjustable from 1 to 100. Full configuration walkthrough is in the NE101 Quick Start guide.
For sites where cellular is unavailable or cost-prohibitive but WiFi range is the limiting factor, the NE101 also supports an optional WiFi HaLow (802.11ah) module. WiFi HaLow operates in the sub-GHz band, giving it significantly better wall and concrete penetration than standard 2.4 GHz WiFi — making it well suited for basement meter rooms, dense building infrastructure, and campus-scale utility networks where a HaLow access point can serve multiple meters within its tested coverage. Unlike cellular, HaLow operates on private infrastructure with no SIM card or recurring data cost.
Underground meter pits and dense building environments often combine two constraints: no standard WiFi penetration and no power outlet. The NE101 addresses this with three field-swappable communication modules: LTE Cat.1 for cellular coverage, WiFi HaLow (802.11ah, sub-GHz) for long-range through-wall coverage on private infrastructure, and standard WiFi 4 for connected indoor environments. Battery operation + IP67 casing handles the power and weatherproofing side.
The NE101 is a general-purpose imaging device — it captures whatever is in front of the lens. The constraint is optical, not functional: the meter face must be legible in the image, which means adequate lighting, minimal reflective glare, and sufficient resolution to distinguish individual digits.
Mechanical totalizing water meters (drum counters, odometer-style readouts) are the most straightforward case. The digit window is typically high-contrast, printed on a light background, and does not change orientation. Standard OCR models trained on digit sequences handle these reliably. Wet-dial meters (where the counter is submerged under a glass cover) may require a supplemental LED ring light in low-ambient-light pits — this can be triggered in sync with the camera’s capture event.
Residential and commercial gas meters with dial-type or digital readout windows are also within scope. Dial meters (pointer-on-numbered-dial format) require a slightly different OCR model than digit-window meters — dial position inference rather than character recognition. This is a solvable problem but does add model complexity on the server side.
Pressure gauges, flow meters, and other analogue industrial instruments with circular dial faces can be read with pointer-detection models. The NE101 has been deployed in machine malfunction monitoring contexts where similar visual inspection logic applies. For complex multi-variable panels, the NeoEdge NG4500 provides the compute headroom to run heavier vision models locally.
Fully electronic meters with backlit LCD displays in direct sunlight environments can produce overexposed images that challenge OCR. Meters in deep pits with no ambient light require supplemental illumination. Meters where the dial face is not physically accessible for camera mounting are better served by pulse-output pulse counters or RS485 output modules — camera-based reading is not universally applicable.
If your system needs photos rather than extracted readings, point the NE101 at your reachable MQTT broker, whether it is on-site or in your cloud, and consume the image reports directly; NeoMind and an OCR host are optional. When numerical data is needed, MQTT carries the image to NeoMind, which runs OCR and delivers the processed reading to SCADA, BMS, or building automation systems through OpenAPI, Data Push, or Webhook.
In the OCR workflow, each capture event publishes an image and device metadata to NeoMind’s configured MQTT broker. The camera payload does not contain the OCR value. NeoMind performs recognition and validation, then exposes a structured reading and photo evidence to your business system. The JSON below illustrates a downstream record after OCR; it is not the raw camera MQTT payload.
{
"device_id": "ne101-a3f2",
"timestamp": "2026-03-20T08:30:00Z",
"site": "building-b-level2",
"meter_id": "WM-045",
"image_url": "https://your-server/captures/ne101-a3f2-20260320-083000.jpg",
"ocr_value": "01027.8",
"unit": "m3",
"battery_pct": 84
}
Device-side MQTT reporting and NeoMind’s camera/OCR workflow are documented in the CamThink Wiki. Configure the camera to reach the NeoMind host, then map the processed reading to the fields required by your system.
With an on-site NeoMind host and no cloud forwarding, image capture, digit recognition, and data delivery can stay on your own network. NE101 captures and transmits; NeoMind on a NeoEdge NG4500 or suitable PC/Linux host runs the OCR extension; your SCADA or BMS receives the structured reading through the configured integration.
The right hardware depends on what you need from the image: a photo in your own system, or an automatic reading. The cards below show where each CamThink product fits.
