Energy management system in real time by image processing and deep learning

As complexity and the size of electrical systems perpetually enlarges, heavy loading will require extra systems to manage electrical power and regulate energy consumption in the demand and customers side. To overcome these problems, in this thesis, a smart power management system has been developed. The system has been realized using minicomputer, sensors, image processing and deep learning algorithms. The developed power management system can handle a lot of electrical management crucial aspects such as reducing power consumption to lower possible limits, controlling ambient temperature, turn on/off some devices depending on human activities. The system we developed in the thesis and named as smart building management has the advantage of making the decisions such as automatic control of the air conditioning system to adjust the ambient temperature according to the comfort of the user, lighting and turning the television on and off depending on the human mobility in the interior. In this study, a camera compatible with minicomputer as Raspberry Pi has been utilized for taking image or recording real-time streaming video for detection the existence of human and his activity. The deep learning based on Convolutional Neural Network (CNN) structure has been utilized as an efficient technique in order to classify the objects properly.Furthermore, the saliency object detection algorithm in addition to image processing tools has been exploited to extract the discriminative features and detect the valuable objects so that speed up both taking the true decision and network learning. The Python computer program language has been utilized because of having huge library functions and its compatibility in programming the real-time embedded system. Experimental results show that the proposed system could be administered electrical consumption with smart and accurate behaviour. Moreover, the recognition algorithm has rapidly produced good results in distinguishing human from other objects, where the accuracy of correct detection can reach to 97.9% and convergence time reach to 0.9 seconds.