Most non-contact meter reading projects run into trouble before OCR is even involved — the field capture layer breaks down first. This article explains why NexAscent, a Singapore IoT integrator, chose CamThink NE101 as the image capture node for their MeterOCR workflow on PUB water meters: battery-powered, independent 4G uplink, no meter modification required.
When system integrators take on a meter digitization project, the OCR algorithm is rarely the hard part. Most teams either build their own or source it from a specialist — the recognition logic is a solved problem at this point. What consistently stalls deployments is the step before OCR: getting a reliable, consistent image out of the field in the first place.
The three constraints that appear in nearly every legacy meter installation create a specific hardware problem:
Meters in underground pits, outdoor enclosures, or sub-metered locations rarely have a convenient power outlet. Running a cable means added civil works and cost.
Relying on the end customer’s Wi-Fi or LAN creates a support dependency you don’t control. Signal quality, firewall rules, and password changes become your maintenance problem.
Regulatory requirements, utility provider rules, or lease agreements often prohibit any physical modification to existing meters. Any attachment or wiring directly to the meter body is out.
These three constraints together define a very specific hardware brief: the capture device must be self-powered, independently connected, and physically non-invasive. Generic IoT cameras and smart meter retrofit kits solve only one or two of these at a time. A purpose-built low-power capture node addresses all three.
NexAscent is a Singapore-based IoT solution provider focused on smart building and facility management. Their product MeterOCR is a proprietary OCR-based workflow that digitizes readings from existing water meters using image analysis — validation, anomaly flagging, and reporting are handled entirely on their platform.
The project brief came from a commercial building operator who needed accurate water consumption data to meet ESG reporting requirements under Singapore’s PUB framework. The core constraint was non-negotiable: PUB incoming water meters cannot be replaced, modified, or physically contacted under Singapore’s utility regulations.
NexAscent’s hardware requirement checklist mapped directly to the three field constraints described above:
Mount the camera in front of the meter face. Zero physical contact with the meter body. No brackets that attach to the meter itself.
Images must reach MeterOCR without depending on the building’s Wi-Fi or LAN. The hardware needed its own SIM-based uplink so NexAscent retained full control of the data path.
No power wiring to the meter location. The device needed to wake on a schedule, capture a high-quality image, upload it, and return to deep sleep — self-sufficiently.
NexAscent selected CamThink NeoEyes NE101 as the capture node. The NE101 handles scheduled image acquisition and 4G LTE upload. MeterOCR receives the images and handles all OCR processing, validation, and reporting on NexAscent’s side.
For PUB compliance, we couldn’t touch or modify the water meters. NE101 solved this — non-contact capture, independent 4G upload, no gateway wiring. It integrated cleanly with our MeterOCR workflow.
NE101 is a capture node, not an inference engine. Understanding this distinction is important before scoping a meter reading project around it.
NE101 is built on an ESP32-S3 — a microcontroller chosen for ultra-low power consumption, not AI compute. It does not have an NPU or the processing headroom to run OCR models locally. What it does well is the job that precedes inference: waking up reliably on a schedule, capturing a high-resolution 5MP image, and transmitting it to wherever the OCR processing lives.
In the NexAscent deployment, the workflow had three steps:
NE101 wakes from deep sleep at a configured interval, captures the meter face image, and uploads it via 4G LTE into NexAscent’s MeterOCR workflow. The device then returns to deep sleep.
NexAscent’s MeterOCR platform processes the incoming image, extracts the digit reading, runs validation checks, and flags any image quality issues for review.
Validated consumption data is aggregated into the building management reporting workflow for ESG compliance and operational tracking.
With standard NE101 firmware, configure an MQTT/MQTTS broker to receive the captured images. Your own OCR platform can subscribe to that broker; NeoMind is optional. If your system requires HTTP/HTTPS ingestion, use middleware between MQTT and your API or a validated custom firmware integration.
This separation of concerns matters for integrators who already have an OCR stack. NE101 is a configurable capture node that feeds images into your existing pipeline. In this case, NexAscent owns the downstream MeterOCR processing; the camera does not run water-meter OCR on-device.
Building a meter reading or visual inspection solution? NE101 can serve as the field capture layer for your existing OCR or monitoring workflow.
Discuss Your Project →NE101 fits the brief for non-contact meter reading — but a few physical and operational factors determine whether it fits your specific deployment. These are the constraints that come up in real integration projects.
| Constraint | Detail | Notes for Integrators |
|---|---|---|
| Capture distance | 8cm–400cm (depends on lens module) | Choose from 60° FOV (15cm/400cm) or 120° FOV (8cm/300cm) OV5640 modules based on meter face size and deployment distance. |
| Image resolution | 5MP OV5640 sensor · Interchangeable lenses | Sufficient for standard rolling-digit and LCD meter faces. Choose from 60° or 120° FOV OV5640 modules based on meter face size and distance. Test with your specific meter type during PoC. |
| Lighting | Built-in fill light · Four modes | Daytime outdoor: typically fine. For underground pits or enclosed boxes without ambient light, NE101’s built-in fill light with Automatic mode ensures reliable capture. Learn fill light setup → |
| Connectivity | 4G LTE Cat.1 (optional module) | Cat.1 handles 1–5 MB image uploads reliably. Check local carrier band support. Wi-Fi and Wi-Fi HaLow modules are also available if cellular is not preferred. |
| Battery life | Depends on capture frequency and connectivity | Use the CamThink battery life calculator to estimate runtime for your specific schedule. Account for battery chemistry and capacity, cellular signal and retries, as well as the configured capture interval. |
| Weather protection | IP67 rated, –20°C to +50°C | Suitable for outdoor installation and rain protection. IP67 means dust-tight and protected against temporary immersion (1m depth, 30min). Underground flood-prone pits or locations with prolonged submersion risk need an additional enclosure. |
| Not a fit | High-speed rotating dial meters | If the meter dial rotates faster than the capture interval can track (e.g. very high-flow industrial meters with rapid dial spin), scheduled imaging will miss intermediate readings. Pulse output or continuous streaming is a better fit. |
On battery life: NE101 enters deep sleep between captures. CamThink’s water-meter engineering guide estimates 0.83 years in its default Cat.1 configuration and 2.08 years in an optimized configuration at five captures per day; these are engineering estimates, not measured results from the NexAscent deployment. Battery chemistry and capacity, LTE signal strength, image size, and retries will change the result. NE101 supports four 1.5V AA alkaline or lithium iron disulfide (Li-FeS2) cells; 3.6/3.7V 14500 cells are not interchangeable. Model your schedule with the battery life calculator and validate it in your field conditions.
The NexAscent deployment is a water meter use case, but the underlying hardware pattern — battery-powered scheduled capture + independent 4G uplink + MQTT image delivery — applies wherever a physical gauge, counter, or dial needs to be read remotely without infrastructure changes.
Integrators have used this same NE101-based capture architecture for gas meters, electricity sub-meters, industrial pressure gauges, level indicators, and building automation panel displays. The common thread: the asset cannot be touched, there is no reliable power or network at the location, and the downstream analytics system already exists.
If your deployment grows beyond a handful of nodes and you need to aggregate image streams from multiple capture points before OCR processing, a gateway-based architecture using NeoEdge NG4500 can consolidate feeds from multiple NE101 units and run inference centrally — reducing per-node cost while keeping data on-premise.
No. NE101 is a capture and transmission node built on ESP32-S3, which does not have the compute headroom to run OCR models. Its role in a meter reading system is to produce a consistent, high-quality image and deliver it to wherever OCR processing occurs — your platform, a cloud endpoint, or an on-premise server.
In this case, NexAscent’s MeterOCR platform performs recognition after NE101 uploads the image. The OCR model is not deployed on the camera.
The standard NE101 setup publishes captured images through MQTT/MQTTS to a configured broker. Your OCR system can consume those messages directly or through middleware. NeoMind is optional; an HTTP/HTTPS API can be reached through middleware or a validated custom firmware integration, rather than a standard-firmware HTTP push setting.
Wi-Fi, 4G LTE Cat.1, and Wi-Fi HaLow connectivity modules are all available depending on your site’s network environment.
NE101 uses a 5MP OV5640 sensor. At typical meter reading distances (15–40 cm), this provides sufficient resolution for standard rolling-digit and LCD meter faces. Image quality is consistent across captures because the device wakes, stabilizes, and captures from a fixed mount position.
Factors that affect downstream OCR accuracy include ambient lighting (especially in enclosed or underground locations), meter face condition (dirt, condensation), and lens selection. A short PoC with your specific meter type and environment is the most reliable way to validate image quality before scaling.
Battery life depends on capture frequency, connectivity module, signal strength, and battery capacity. The energy used by each capture-and-transmit cycle depends on the network and image upload conditions.
Use the CamThink battery life calculator to model your capture frequency, battery chemistry and capacity, and network conditions. At five Cat.1 captures per day, CamThink’s engineering estimate is 0.83 years by default or 2.08 years with optimization. Confirm runtime on your actual meter sites before setting a service interval.
The standard documented workflow is to set the capture schedule and MQTT upload settings in NE101’s local Web UI. Plan for local access if those settings need to change after installation. A project-specific remote configuration workflow requires a separately validated integration; it is not a default NeoMind or NE101 firmware capability.
For an experienced IoT integrator, initial configuration (Wi-Fi or Cat.1 credentials, capture schedule, MQTT broker) via the NE101 web UI typically takes under 30 minutes per device. The first image upload can be achieved the same day the hardware arrives.
For a multi-node rollout, document each device’s SIM, MQTT broker, capture schedule, and mounting position during commissioning. The standard configuration and firmware update steps are in the NE101 Quick Start; account for local Web UI access unless your project has a validated remote management integration.
No CamThink cloud service is required. In the standard configuration, NE101 publishes images to the MQTT/MQTTS broker you configure; the broker and downstream OCR system can be hosted in your own environment. For Cat.1 deployments, verify the carrier network path and broker location against your data residency requirements. HTTP/HTTPS delivery to your API needs middleware or a validated custom firmware integration.
