Adapta: An Edge-Deployed Multimodal Smart Lighting System Using Computer Vision and Federated Machine Learning for Circadian-Aware Adaptive Illumination
Main Article Content
Abstract
Most smart lighting systems today still rely on passive infrared sensors that cannot distinguish between a person working at a desk and someone trying to rest. This paper introduces Adapta, a building-deployed smart lighting framework that uses an on-device RGB-D camera, a lightweight MobileNetV3-GRU hybrid network, and a federated machine learning pipeline to continuously recognise occupant activities and adjust both brightness and colour temperature of luminaires in real time. All inference runs locally on a Raspberry Pi 5 edge node; no raw video is ever transmitted to any external server. Illumination targets are computed from the Rea-Figueiro circadian stimulus (CS) model, so the system actively supports occupants biological day-night rhythms. Deployed across 120 volunteers in three building types over 90 days, Adapta achieved a 41.3 percent energy reduction over static baselines, 94.7 percent activity classification accuracy, and statistically significant improvements in daytime alertness and sleep quality. End-to-end actuation latency remained below 28 ms throughout the entire trial period
Article Details
Section
COPYRIGHT
Submission of a manuscript implies: that the work described has not been published before, that it is not under consideration for publication elsewhere; that if and when the manuscript is accepted for publication, the authors agree to automatic transfer of the copyright to the publisher.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgment of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work
- The journal allows the author(s) to retain publishing rights without restrictions.
- The journal allows the author(s) to hold the copyright without restrictions.
References
[1] IEA, "Global Electricity Review 2023," International Energy Agency, Paris, 2023.
[2] M. S. Rea and M. G. Figueiro, "Light as a circadian stimulus for architectural lighting," Lighting Res. Technol., vol. 50, no. 4, pp. 497-510, 2018.
[3] C. Cajochen et al., "Evidence that the lunar cycle influences human sleep," Current Biology, vol. 23, no. 15, pp. 1485-1488, 2013.
[4] V. Singhvi et al., "Efficient rule-based occupancy sensing for building automation," in Proc. ACM SenSys, San Diego, CA, 2005, pp. 1-12.
[5] Y. Li et al., "DeepLight: Reinforcement learning-based adaptive lighting control," in Proc. IEEE IPSN, Pittsburgh, PA, 2019, pp. 1-12.
[6] T. Yun and C. Yoo, "VisionLight: Vision-based occupancy estimation for smart lighting," IEEE Sensors J., vol. 21, no. 12, pp. 14021-14030, 2021.
[7] H. Kobayashi et al., "Physiological signal-driven adaptive lighting for office productivity," Build. Environ., vol. 193, pp. 107-120, 2021.
[8] F. Zhang et al., "Daylight harvesting with closed-loop PIR-based occupancy sensing," Energy Build., vol. 108, pp. 274-286, 2015.
[9] V. Erickson et al., "OBSERVE: Occupancy-based system for efficient reduction of HVAC energy," in Proc. IPSN, Chicago, IL, 2011, pp. 258-269.
[10] H. Aghajan and A. Cavallaro, Human-Centric Interfaces for Ambient Assisted Living. Waltham, MA: Academic Press, 2009.
[11] G. Bertasius et al., "Is space-time attention all you need for video understanding?" in Proc. ICML, Vienna, 2021, pp. 813-824.
[12] M. Sandler et al., "MobileNetV2: Inverted residuals and linear bottlenecks," in Proc. CVPR, Salt Lake City, UT, 2018, pp. 4510-4520.
[13] M. Tan and Q. Le, "EfficientNet: Rethinking model scaling for convolutional neural networks," in Proc. ICML, Long Beach, CA, 2019, pp. 6105-6114.
[14] P. Boyce et al., "The benefits of daylight through windows," Lighting Res. Center, Troy, NY, Technical Report, 2003.
[15] A. U. Viola et al., "Blue-enriched white light in the workplace improves self-reported alertness, performance and sleep quality," Scand. J. Work Environ. Health, vol. 34, no. 4, pp. 297-306, 2008.
[16] B. McMahan et al., "Communication-efficient learning of deep networks from decentralized data," in Proc. AISTATS, Fort Lauderdale, FL, 2017, pp. 1273-1282.
[17] H. Li et al., "Federated learning for smart building: a survey on data privacy and resource allocation," IEEE IoT J., vol. 9, no. 13, pp. 10414-10430, 2022.
[18] R. Zhang et al., "Privacy-preserving building occupancy estimation via federated learning," Build. Environ., vol. 198, art. no. 107889, 2021.
[19] T. Li et al., "Federated optimization in heterogeneous networks (FedProx)," in Proc. MLSys, Austin, TX, 2020, pp. 429-450.
[20] C. Dwork and A. Roth, "The algorithmic foundations of differential privacy," Found. Trends Theor. Comput. Sci., vol. 9, no. 3-4, pp. 211-407, 2014.
[21] A. Howard et al., "Searching for MobileNetV3," in Proc. ICCV, Seoul, South Korea, 2019, pp. 1314-1324.
[22] G. Hinton, O. Vinyals, and J. Dean, "Distilling the knowledge in a neural network," in Proc. NIPS Deep Learning Workshop, Montreal, Canada, 2014.
[23] T. M. Brown et al., "Recommendations for daytime, evening, and nighttime indoor light exposure to support physiology, sleep, and wakefulness in healthy adults," PLOS Biology, vol. 20, no. 3, 2022.