Conservation of Energy using Object Detection Model

Abstract

Energy conservation has become a vital responsibility for every citizen. Considering classroom environment, electric appliances like fans and lights are usually unmonitored while students leave. It leads to the wastage of electricity. To save electricity, conventionally, sensors can be deployed to detect the presence / absence of person in the classroom and control electric appliances based on its trigger. Since (low-cost) sensors have reliability issues with shorter life span, it can't be used effectively. On the other hand, if costly (high precision and reliability) sensors were used to detect persons, deploying it in each and every classroom is not practicable due to very high initial investment. Here, this paper's approach is to use a medium quality, low cost night vision web-camera to detect persons inside classroom using YOLOv3 Object detection model built on top on TensorFlow framework. Computational capabilities for processing webcam footage is provided by PCs inside each and every classroom. (Assumption: Each and every classroom has a dedicated PC for sharing power-point slides) Switch board is configured with relays, which are connected in parallel to normal switches to allow manual intervention. Relays are controlled by Wi-Fi enabled micro-controllers like NodeMCU. Communication is made possible between NodeMCU and PC via LAN. By this means, a huge amount of electricity can be saved with least deployment cost.