In our info-obsessed culture, hackers are increasingly interested in ways to quantify the world around them. One popular project is to collect data about their home energy or water consumption to try and identify any trends or potential inefficiencies. For safety and potentially legal reasons, this usually has to be done in a minimally invasive way that doesn’t compromise the metering done by the utility provider. As you might expect, that often leads to some creative methods of data collection.

The latest solution comes courtesy of [Keilin Bickar], who’s using the ESP8266 and a serial TTL camera module to read the characters from the LCD of his water meter. With a 3D printed enclosure that doubles as a light source for the camera, the finished device perches on top of the water meter and sends the current reading to HomeAssistant via MQTT without any permanent wiring or mounting.

Of course, the ESP8266 is not a platform we generally see performing optical character recognition. Some clever programming was required to get the Wemos D1 Mini Lite to reliably read the numbers from the meter without having to push the task to a more computationally powerful device such as a Raspberry Pi. The process starts with a 160×120 JPEG image provided by a VC0706 camera module, which is then processed with the JPEGDecoder library. The top and bottom of the image are discarded, and the center band is isolated into blocks that correspond with the position of each digit on the display.

Within each block, the code checks an array of predetermined points to see if the corresponding pixel is black or not. In theory this allows detecting all the digits between 0 and 9, though [Keilin] says there were still the occasional false readings due to inherent instabilities in the camera and mounting. But with a few iterations to the code and the aid of a Python testing program that allowed him to validate the impact of changes to the algorithm, he was able to greatly improve the detection accuracy. He says it also helps that the nature of the data allows for some basic sanity checks;  for example the number only ever goes up, and only by a relatively small amount each time.

This method might not allow the per-second sampling required to pull off the impressive (if slightly creepy) water usage data mining we saw recently, but as long as you’re not after very high resolution data this is an elegant and creative way to pull useful data from your existing utility meter.