Occupant's Behavior and Emotion Based Indoor Environment's Illumination Regulation

TL;DR

This paper introduces a novel framework combining posture and emotion detection from Kinect sensor data with sensor networks to optimize indoor illumination, reducing energy consumption effectively.

Contribution

It develops a new occupancy modeling approach based on posture and emotion detection, integrating multiple sensors and machine learning techniques for energy-efficient indoor lighting regulation.

Findings

Random forest achieved highest accuracy in posture classification.

Deep learning with Adadelta optimizer excelled in emotion detection.

Integrated sensor network enables unmanned illumination regulation.

Abstract

This paper presents an efficient approach for building occupancy modeling to reduce energy consumption. In this work, a novel approach to occupancy modeling based on the posture and comfort level of the occupant is developed, and subsequently, we report a new and efficient framework for detecting posture and emotion from skeleton joints and face points data respectively obtained from the Kinect sensor. The proposed approach is tested in terms of accuracy, region of convergence, and confusion matrix using several machine learning techniques. Out of all the techniques, random forest classifier gave the maximum blind test accuracy for multi-class classification of posture detection. Deep learning is used for emotion detection using several optimizers out of which Adadelta gave the maximum blind test accuracy for multi-class classification. Along with the Kinect sensor, several other sensors such as the magnetic door sensor, pyroelectric sensors, and illumination sensors are connected through a wireless network using Raspberry Pi Zero W. Thus creating an unmanned technique for illumination regulation.