oa Integration of Multisensor data and Deep Learning for realtime Occupancy Detection for Building Environment Control Strategies

One of the most prominent areas of energy consumption in residential units is for heating, ventilation and air-conditioning (HVAC) systems. The conventional systems for HVAC depend on the wired thermostats that are deployed at fixed locations and hence, are not convenient and do not respond to the dynamic nature of the thermal envelope of the buildings. Moreover, it is important to note that the distribution of the spatial temperature is not uniform. The current environment control strategies are based on the maximum occupancy numbers for the building. But there are always certain areas of a building which are used less frequently and are cooled needlessly. Having the real-time occupancy data and mining on it to predict the occupancy patterns of the building will help in energy effective strategy development for the regulation of HVAC systems through a central controller. In this work, we have deployed a network of multiple wireless sensors (humidity, temperature, CO2 sensors etc.), computational elements (in our case, a raspberry pi, to make it cost effective) and camera network with an aim to integrate the data from the multiple sensors in a large multifunction building. The sensors are deployed at multiple locations in such a way that the non-uniform spatial temperature distribution is overcome, and these sensors capture the various environmental conditions at a temporal and much finer spatial granularity. The pi camera is connected to a raspberry pi which is fixed at an elevation. The detection is performed using the OpenCV library and the python programming. This system can detect the occupancy with an accuracy of up to 90%. For occupancy detection and counter, a linear SVM is trained sampling on positive and negative images and the evaluation on test images or video feed makes use of non-maximum suppression (NMS Algorithm) to ignore redundant, overlapping HOG (Histogram Oriented Gradient) boxes. The data collected by the sensors is sent to the central controller on which the video processing algorithm is also running. Using the multiple environmental factors data available to us, models are developed to predict the usage in the building. These models help us to define the control parameters for the HVAC systems in adaptive manner in such a way that these parameters not only help in reducing the energy used in a building, but also help to maintain the thermal comfort. The control parameters are then sent as IR signals to AC systems that are controlled by IR remotes or as wireless signals to AC systems controlled by wireless thermostats. In comparison to the conventional temperature controller, our system will avoid overcooling of areas to save energy and predict the occupancy in the buildings so that the temperature is brought within the comfort zone of humans before over-occupancy takes place. Our system also has benefits of using wireless sensors that operate on low power, but the tradeoff between the power and the communication frequency should be well maintained. Our system additionally has two features: firstly, it can provide the live video streaming for remote monitoring using a web browser for the user interface and secondly, sending automatic notifications as messages in case of anomalies like abnormally high temperatures or high carbon dioxide concentration in a room. These two features can be used as cost-effective replacement for traditional systems in the applications of CCTV, burglary systems respectively. Keywords: wireless sensors, Air conditioning, opencv, NMS Algorithm, Histogram Oriented Gradient, thermal comfort.