Automatic Human Detection Using Reinforced Faster-RCNN for Electricity Conservation System

Electricity conservation systems are designed to conserve electricity to manage the bridge between the high raising demand and the production. Such systems have been so far using sensors to detect the necessity which adds an additional cost to the setup. Closed-circuit Television (CCTV) has been installed in almost everywhere around us especially in commercial places. Interpretation of these CCTV images is being carried out for various reasons to elicit the information from it. Hence a framework for electricity conservation that enables the electricity supply only when required, using existing resources would be a cost effective conservation system. Such a framework using a deep learning model based on Faster-RCNN is developed, which makes use of these CCTV images to detect the presence or absence of a human in a place. An Arduino micro controller is embedded to this framework which automatically turns on/off the electricity based on human's presence/absence respectively. The proposed approach is demonstrated on CHOKE POINT dataset and two real time datasets which images from CCTV footages. F-measure, Accuracy scores (AUC score) and training time are the metrics for which the model is evaluated. An average accuracy rate of 82% is obtained by hyper-parameter tuning and using Adam optimization technique. This lays the underpinning for designing automatic frameworks for electricity conservation systems using existing resources.