Are you aware of how difficult and annoying the job of energy auditor is? Hundreds of visits to strangers per day to ascertain how much electricity they’ve used, keeping all readings on paper. To save time performing this routine task and avoid possible mistakes, one of our customers came up with the idea of developing a special ElectriCity project based on Machine Learning. It’s an Android application which recognizes meters and captures readings and serial numbers via a mobile phone camera.
A private good
The project is protected by a Non-Disclosure Agreement
Machine Learning in ElectriCity project
One of the most innovative features Machine Learning allows is the recognition of images, voice, words, faces, and other objects. Applying complex ML algorithms, our mobile app developers built an app that recognizes meters with all their presented data via a camera.
After recognizing a certain meter, ElectriCity captures the readings and a serial number and, then, tries to find a pre-registered user in the database to add the record of the readings to the user's history.
Using this application, it’s possible to make a registry of subscriber readings in a few simple steps. All you need is a mobile phone with a good camera that can distinguish the details in poor lighting.
How does ElectriCity work?
The application recognizes a meter, capturing the readings and serial numbers of any type of meter by editing configuration files.
One of the main features is that ElectriCity performs the recognition process in real time. In other words, you don’t need to take photos, just run the application and point a camera at a meter (exactly like QR code). This ensures that during the recognition process, the application determines the same readings on the meter for each frame.
Try the ability to view a detailed history of registry of the readings for each user. According to the history, you can create a histogram of the use of electricity between the registry of the readings.
ElectriCity supports geolocation services for confirming a location of, "shooting" in order to avoid any falsification of results.
All cut out image frames with the actual photo of the readings and a serial number are stored.
The app has a unique design with a nice variety of custom components.
It was necessary to learn the networks to detect a recognizable object and serial number on the image, and select the character area. Then, the app should confirm which metric belongs to each particular person. The client also wanted to introduce a feature of app profiling to work with different meters.
We used the method of contour analysis that narrows the number of meters, which the application can potentially work with, to presence of a key contour ("anchor"). It is an object that has a unique contour (in most cases, a black box) around the numbers with the readings.
Based on the geometry of the anchor, we implemented a technology which calculates the location of the serial number and accurately recognizes it.
Registering the location of a particular meter in the database, tracking the shooting location, and comparing all known meter data, the app can easily determine the owner of a certain meter.
We transferred all the parameters, which the recognition process may depend on, to the configuration file, and determined the parameters for each supported meter with the possibility of choosing an operational meter. This procedure allowed us to introduce a useful app profiling feature.
As you can see, we have developed a very useful Android application based on Machine Learning, designed for power companies, which easily recognizes meters and captures readings and serial numbers through smartphones. Our client has presented the world with a unique software for gleaning actual data from various meters quickly and precisely. Thanks to our successful cooperation, the customer’s revenue has been increased by 45% and the workflow of thousands of energy auditors has become more effective and less time-consuming.