Advanced computer vision technology for reading utility meters, including digital panels, serial numbers, and analog dials.
Reading panels are isolated on the source image
Isolated panels are extracted for separate detection processes. This isolates the required information from the rest of the noise in the image and speeds up detection.
Individual alphanumeric characters are detected within the extracted segments. This two stage detection model results in far more accurate results than OCR which fails often in real world conditions.
{
"serial": {
"text": "5168XW",
"confidence": 0.89
},
"units": {
"text": "03649",
"confidence": 0.92
}
}
Real-world utility meter detection across diverse conditions
Helios solves the challenge of obtaining large datasets for model training with the ability to create an unlimited number of unique, high-quality images using a combination of raytraced 3D models and 2D image composites.
This allows the model to be trained on a diverse dataset that simulates real-world conditions—varying lighting, angles, materials, and environmental factors. This approach ensures the model is exposed to a wide range of scenarios, enhancing its accuracy and robustness when reading dials and gauges in any setting.