Image Recognition Algorithm from the Electricity Metering Devices

Abstract

Works on image recognition with metering devices of electricity are conducted within the framework of the joint project of the Udmurtenergo (Branch of IDGC of Centre and Volga region) and Kalashnikov ISTU in the development of an automated system «Mobile Energy Billing». When considering the question of measuring the amount of energy consumed, when the full replacement of obsolete instruments of the automated information-measuring system of commercial accounting in a short time is impossible, there is a need for automated verification of readings. Automation of check of correctness of readings allows for increasing the measurement accuracy of electricity supply, which in turn leads to the reduction of losses. To achieve this purpose, you can use photos of metering devices made by the controllers to the terminal. Images are processed by an algorithm that recognizes the readings, and then they are stored in a database of electricity consumption. Before recognition, the image is analyzed for the location of objects and correcting distortions. The preprocessing procedures consist of several stages of finding the contours of objects and their modification. For this purpose, the Canny detector and Hough transformations are propose and also the affine transformation to correct the position of the objects in the image. This method allows for more accurate extracting the desired information. The recognition algorithm uses Viola-Jones object detection, which is effective when working with large amounts of data on low power computing machines. Also, this method is easy to use, allowing in short terms to adapt to the emergence of new types of metering devices, with previously unaccounted for elements in the training. The described process allows for quickly processing a large amount of information with high precision, thus reducing risks and losses at the account of electricity sold to consumers.