Energy · AI and image processing

Electricity Meter Reading

Turn a smartphone camera into a meter-reading device: the index and serial-number barcode of analogue and digital electricity meters are read instantly from the camera stream with AI.

Field worker reading a meter in an electrical panel with a tablet
Local and cost-effective

Smartphones instead of dedicated reading devices

We turn our scientific research in artificial intelligence into products that meet the needs of the sector. With our algorithms, any Android or iOS device becomes a meter-reading device that reads both the index and the serial-number barcode of analogue and digital electricity meters.

Foreign alternatives charge high per-use fees, which makes wide-scale use unfeasible. Our locally developed, cost-effective application is suitable for wide-scale use and can be customised for each company.

Reading an electricity meter with a smartphone camera
Four steps

How are meters read with AI?

Point the camera

The user points the mobile device's camera at the meter.

Image is captured

A digital image of the meter is captured.

Image is analysed

The image is processed and analysed with computer vision.

Data is returned

The meter serial number and index values are returned as output.

Rationale and key features

Faster reading, fewer errors, lower cost

The values to be read from a meter are split into time-of-use periods; with weekend consumption and measurement types added, the reader's workload and the chance of error grow. Field staff read meters quickly with the app and share the data with the company server without extra devices. Smartphones reduce the cost of dedicated reading devices.

Subscribers can also read their meters with the company app, request their bills and pay online if they wish. Liveness detection and cross-validation against previous indexes ensure that subscribers are photographing their own meters.

Electricity meter and the mobile app screen performing the reading
Technical approach

Object detection and OCR with deep learning

Index data is detected in the camera stream with a deep-learning object detection algorithm and read simultaneously with our own OCR. We developed dedicated image processing algorithms for single-phase analogue and multi-phase digital meters; the model detects the T, T1, T2… tariff indexes on the digital display in real time.

The automatic barcode reader reads the serial number from the same frame and checks whether the meter belongs to the subscriber in question.

Digital displays of different electricity meters
Architecture

On the device or on the server

Whether the application works through an API or natively on the device is up to the customer.

Native on-device model

  • The trained model is optimised for mobile processors
  • Runs fast on the device without an internet connection
  • Only the result is sent to the server

API-based

  • The model runs on the server
  • Can be used on iOS, Android, Windows and the web
  • Camera and interface modules are added to the customer's requirements
Frequently asked questions

Common questions

Who takes the meter photo: field staff or the subscriber?

Both are possible; the choice is the company's.

Field staff: today field workers type index values into handheld terminals by hand. With this system, values are read quickly with standard smartphones.

Subscribers: if the company chooses, subscribers can send a photo of their meter at any time during the billing period to see their current bill. Where regulations allow, subscribers can also be billed on their own declaration when access is difficult, for example during a pandemic or quarantine.

Why use image processing when the index can be typed in by hand?
  • Speed: the barcode number and the index are captured in a few seconds.
  • Accuracy: manual reading errors are eliminated.
  • Security: intentional or accidental errors can be audited against the photo archive.
  • Convenience: the subscriber only needs to frame the meter in the camera.
How are erroneous or deliberate false readings prevented?

An accidental error can only come from a misreading by the app, and that risk is low. After reading, the app opens a confirmation screen and the record is created only after approval. Alarms can be raised for values that deviate from previous indexes and seasonal averages. If an error still occurs, the current photo in the database settles it.

A deliberate error would require tampering with the photo. Because the barcode and the index are read from the same frame, they cannot come from different meters; image processing also checks that the photo was taken of a real object.

How does reading work in the dark or with glare?

The app uses the camera light with adjustable intensity. Images with glare are used to retrain the model. Indexes that cannot be read can be entered by hand; every manually entered value is stored together with its photo.

Demo or quote for Electricity Meter Reading

Fill in the form and we will show you the product at a time that suits you, or prepare a price quote for your use case.