Intelligent Smart Lighting Systems with Human Detection and AI-Based Anomaly Detection: A Comprehensive Framework for Energy-Efficient Building Automation

Available online www.ejaet.com European Journal of Advances in Engineering and Technology, 2019, 6(8):105-117 Research Article ISSN: 2394 - 658X 105 Intelligent Smart Lighting Systems with Human Detection and AIBased Anomaly Detection: A Comprehensive Framework for EnergyEfficient Building Automation Tharakesavulu VANGALAPAT Department of Computer Science in Data Science, Indiana University Bloomington, IN, USA [email protected] _____________________________________________________________________________________________ ABSTRACT Building automation systems have evolved significantly with the integration of Internet of Things (IoT) technologies and artificial intelligence algorithms. This paper presents a comprehensive Smart Lighting System (SLS) framework that incorporates multi-sensor human detection, adaptive illumination control, and machine learning-based anomaly detection to achieve enhanced energy efficiency and system reliability. The experimental deployment across three distinct building environments demonstrated substantial improvements: a 42.3% reduction in energy consumption, 94.7% accuracy in human presence detection, and 89.2% precision in anomaly identification. The system architecture integrates passive infrared (PIR) sensors, computer vision modules, and ensemble machine-learning algorithms to create an autonomous lighting infrastructure capable of real-time adaptation and predictive maintenance. Through extensive 90-day field trials involving 156 participants across office, residential, and laboratory environments, this study validates the effectiveness of the proposed approach in diverse operational scenarios. This study contributes to the original algorithms for sensor fusion, adaptive control strategies, and novel anomaly detection techniques specifically designed for smart lighting applications. This study demonstrates that intelligent lighting systems can achieve substantial energy savings while maintaining superior user comfort and system reliability through the application of novel AI-driven techniques. The improvements build upon proven commercial foundations, suggesting opportunities for technology transfer and integration with the existing lighting infrastructure. Keywords: Smart Lighting Systems, Human Detection, Anomaly Detection, Machine Learning, IoT, Building Automation, Energy Efficiency, Sensor Fusion _____________________________________________________________________________________________ INTRODUCTION The proliferation of Internet of Things (IoT) technologies and the advancement of machine learning algorithms have fundamentally transformed traditional building automation systems into intelligent and adaptive environments capable of autonomous decision-making and predictive maintenance [1]. Among the various building systems, lighting is a critical component for both energy consumption and occupant comfort, typically accounting for 15– 20% of the total building energy usage in commercial environments [2]. Traditional lighting control systems have several fundamental limitations that hinder their effectiveness in modern building applications. These systems suffer from excessive energy consumption owing to static operational modes, lack of occupancy awareness leading to unnecessary illumination during unoccupied periods, absence of predictive maintenance capabilities resulting in unexpected system failures, and limited adaptability to user preferences and varying environmental conditions [3]. These shortcomings have motivated researchers and practitioners to develop intelligent Smart Lighting Systems (SLS) that can address these challenges through advanced sensor integration and artificial intelligence techniques. The emergence of affordable sensor technologies, including passive infrared (PIR) detectors, computer vision cameras, and environmental monitoring sensors combined with powerful edge computing capabilities, has enabled the creation of sophisticated lighting systems capable of real-time human detection, behavioral pattern learning, and autonomous adaptation to changing conditions [4]. Furthermore, the integration of machine-learning algorithms for Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 106 anomaly detection provides unprecedented capabilities for pre-dictive maintenance and system optimization, potentially reducing maintenance costs and improving system reliability. This paper presents a comprehensive Smart Lighting System framework that addresses three critical aspects of intelligent building automation. First, the system implements robust human detection through multi-sensor fusion techniques that combine the strengths of different sensor modalities while mitigating their individual weaknesses. Second, it incorporates adaptive lighting control based on occupancy patterns, user preferences, and circadian rhythm considerations to optimize both energy efficiency and user comfort. Third, the system features an AI-based anomaly detection for predictive maintenance and system reliability enhancement. The primary contributions of this research include the development of a novel multi-modal sensor fusion algorithm that combines PIR sensors with computer vision to enhance human detection accuracy, design of an adaptive lighting control system utilizing reinforcement learning principles for personalized illumination management, implementation of an ensemble machine learning approach for comprehensive anomaly detection in smart lighting infrastructure, extensive experimental validation across multiple building environments with quantitative performance analysis, and provision of an opensource hardware and software framework to enable reproducible research and practical deployment. The experimental validation demonstrates significant improvements over conventional lighting systems, with an average energy savings of 42.3%, human detection accuracy of 94.7%, and anomaly detection precision of 89.2%. These results were obtained through rigorous testing across three different building environments over a 90-day period involving 156 participants in real-world operational scenarios. RELATED WORK Smart Lighting System Technologies The evolution of smart lighting systems has progressed over several distinct technological generations, beginning with simple timer-based controls and advancing to sophisticated AI-driven adaptive systems. Early research efforts focused on basic occupancy detection using single-sensor approaches, which achieved moderate success but suffered from high false-positive rates and limited adaptability to varying environmental conditions [5]. Recent technological advances have emphasized multi-sensor integration and machine learning approaches to improve system performance and reliability. Wang and Liu [6] demonstrated the effectiveness of reinforcement learning techniques in adaptive lighting control applications, achieving 23% energy savings compared with conventional systems. However, their approach was limited to simple occupancy patterns, and did not address the critical requirements for anomaly detection and predictive maintenance. Commercial smart lighting solutions, including popular systems such as Philips Hue and LIFX, have gained significant market adoption and have established important foundations for user-controlled color and intensity adjustments [7]. Building on these commercial successes, this research explores autonomous intelligence and adaptive behavior capabilities. Industrial solutions developed by companies such as Acuity Brands and Cree have successfully incorporated basic occupancy-sensing capabilities, thereby providing a solid foundation for more sophisticated AI-driven anomaly detection and predictive maintenance features. Human Detection in Smart Environments Human detection technologies for smart building applications have evolved from simple motion sensors to sophisticated computer vision systems with advanced pattern-recognition capabilities. Yun and Lee [8] provided comprehensive analysis of PIR sensor-based detection systems, demonstrating detection ranges of 3–12 meters with power consumption below 0.1W. However, PIR sensors suffer from inherent limitations, including the inability to distinguish between humans and other heat sources, sensitivity to environmental temperature variations, and false triggers from air conditioning systems and other building equipment. Computer-vision approaches have demonstrated superior accuracy in human detection and classification tasks. Liu et al. [9] achieved 96.2% accuracy using convolutional neural networks (CNNs) with RGB-D cameras for human presence detection. Deep learning models, particularly You Only Look Once (YOLO) and R-CNN variants, have demonstrated real-time human detection capabilities suitable for smart building applications, with processing latencies below 100ms [10]. Sensor fusion techniques that combine multiple detection modalities have emerged as the most promising approach for robust detection of human pres-ence in challenging environments. Kumar and Singh [4] demonstrated that combining PIR sensors with computer vision systems reduces false positive rates by up to 60% while maintaining detection accuracy above 90%, making such systems suitable for practical deployment in diverse building environments. Anomaly Detection in IoT Systems Anomaly detection in IoT environments presents unique challenges owing to the heterogeneous nature of sensor data, varying operational conditions, and need for real-time processing with limited computational resources. Traditional statistical approaches, including Gaussian Mixture Models (GMM) and Principal Component Analysis (PCA), have been applied to IoT anomaly detection with moderate success in controlled environments [11]. Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 107 Machine learning approaches have shown superior performance in detecting complex anomalies in IoT systems compared with traditional statistical methods. Isolation Forest algorithms have proven particularly effective for detecting point anomalies in high-dimensional sensor data, whereas deep learning approaches, especially autoencoders and Long Short-Term Memory (LSTM) networks, have demonstrated excellent performance in detecting temporal anomalies in time-series IoT data streams. Ensemble methods that combine multiple anomaly detection algorithms have shown improved robustness and reduced false-positive rates compared to single-algorithm approaches. Research by Chen and Liu [2] demonstrated that ensemble approaches can achieve up to a 15% improvement in anomaly detection precision compared to individual algorithms, making them suitable for practical deployment in critical building automation systems. SYSTEM ARCHITECTURE AND METHODOLOGY Overall System Design The proposed Smart Lighting System (SLS) architecture follows a distributed computing model with edgeprocessing capabilities and cloud connectivity for data aggregation and model updates. The system comprises four primary subsystems that work collaboratively to achieve optimal lighting control: Sensor Data Acquisition, Human Detection Engine, Adaptive Lighting Controller, and Anomaly Detection Module. Figure 1: Smart Lighting System Architecture Overview As illustrated in Figure 1, the Sensor Data Acquisition subsystem is responsible for collecting data from multiple sensor modalities including PIR motion sensors, RGB-D cameras, environmental sensors, and ambient light sensors. This subsystem implements data preprocessing, filtering, and synchronization to ensure consistent and reliable sensor data streams for subsequent processing stages. The Human Detection Engine processes the acquired sensor data through a multi-modal fusion approach that combines the PIR sensor outputs with computer vision analysis to achieve robust human presence detection and tracking. This subsystem implements advanced signal processing techniques and machine learning algorithms to minimize false positives while maintaining high detection sensitivity. The Adaptive Lighting Controller utilizes human detection results along with the environmental context, and learns user preferences to make intelligent lighting control decisions. This subsystem implements circadian rhythm modeling, brightness adaptation algorithms, and user preference learning to optimize both energy efficiency and user comfort. The Anomaly Detection Module continuously monitors system performance and sensor data patterns to identify potential failures, security breaches, or unusual operational conditions. This subsystem employs ensemble machine learning techniques to provide robust anomaly detection with low false-alarm rates. Multi-Modal Human Detection Algorithm The human detection system employs a sophisticated multi-stage approach that combines PIR sensor data with computer vision analysis to achieve robust presence detection and tracking capabilities. The PIR signal-processing stage implements digital filtering and adaptive threshold detection to minimize noise and environmental interference. PIR sensors generate analog signals that undergo digital filtering using optimized coefficients to reduce highfrequency noise while preserving the motion-related signal components. The filtering process was implemented using a finite impulse response (FIR) filter with seven coefficients optimized for building automation applications. (1) Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 108 where wi represents the optimized filter coefficients [0.05, 0.1, 0.15, 0.4, 0.15, 0.1, 0.05]. The motion detection threshold is adaptive based on environmental noise characteristics, automatically adjusting to varying background conditions, such as HVAC system vibrations and temperature fluctuations: T adaptive (t) = T base + α · σ noise (t) (2) where Tbase = 0.15V, α = 2.5, and σnoise denotes the estimated noise standard deviation. The computer vision processing module implements a comprehensive detection pipeline that includes background subtraction to identify moving objects, depth filtering to remove objects outside the relevant detection range (0.5m to 5m), contour detection to identify connected components representing potential human subjects, feature extraction using Histogram of Oriented Gradients (HOG) and geometric features, classification using pre-trained Support Vector Machine (SVM) classifiers, and tracking using Kalman filters for motion prediction and object continuity. The sensor fusion stage combines the PIR and vision confidence scores using a weighted voting approach with weights optimized through historical performance data analysis. C final = w PIR · C PIR + w vision · C vision (3) where the weights are set to w PIR = 0.35 and w vision = 0.65 based on extensive empirical evaluation. Adaptive Lighting Control System The adaptive lighting controller adjusts illumination parameters based on occupancy detection results, time-of-day considerations, and learned user preferences to optimize both energy efficiency and user comfort. The circadian color temperature model follows a cosine function that varies between 2700K (warm white) and 6500K (cool white), with the peak cool temperature occurring at 2 PM: (4) where CCTmin = 2700K, CCTmax = 6500K, and tpeak = 14 hours (2 PM). Brightness adaptation is implemented based on ambient light levels and detected activity patterns. L target = L base · (1 − k · L ambient ) · A factor (5) where Lbase = 500 lux, k = 0.7 is the daylight factor, and Afactor represents activity-based adjustment. Ensemble Anomaly Detection Framework Anomaly detection systems monitor multiple data streams, including sensor readings, power consumption patterns, communication network status, and user interaction patterns. The system employs three complementary algorithms: the Isolation Forest for point anomalies, LSTM Autoencoder for temporal anomalies, and One-Class SVM for boundary-based detection. The ensemble decision process combines the individual algorithm outputs using the following optimized weights: Sensemble = w1SIF + w2SLSTM + w3SSVM (6) where the weights are determined through cross-validation: w 1 = 0.30, w 2 = 0.45, w 3 = 0.25. EXPERIMENTAL SETUP AND IMPLEMENTATION Experimental validation was conducted across three distinct building environments to ensure comprehensive system evaluation under diverse operational conditions. The selection of these environments was designed to represent typical smart lighting deployment scenarios, while providing sufficient diversity in occupancy patterns, environmental conditions, and usage requirements. Deployment Environments Office environment testing was conducted in a public co-working space at the Austin Technology Center, featuring dimensions of 8m × 6m × 3m ceiling height with a typical occupancy of 6–12 people during business hours (8 AM– 6 PM). The lighting installation consists of 12 LED fixtures arranged in a 3×4 grid configuration to provide uniform illumination coverage. The testing duration was extended to 90 days from January 15 to April 15, 2019, capturing seasonal variations in daylight availability and occupancy patterns. The residential environment testing involved three volunteer households within the Austin metropolitan area, focusing on living rooms (5m × 4m) and bedrooms (4m × 3m) with a typical occupancy of 2–4 people on variable schedules. The lighting installation featured eight LED fixtures distributed strategically across rooms to provide comprehensive coverage while maintaining the aesthetic appeal. The testing period covered 90 days, from February 1 to May 1, 2019, encompassing diverse residential usage patterns, including weekday and weekend variations. Laboratory environment testing was conducted at the Austin Community Lab, a public maker space with dimensions of 10m × 8m × 3.5m ceiling height supporting 3–8 researchers with potential 24/7 operation schedules. The advanced lighting installation includes 16 LED fixtures with enhanced control capabilities to support various research activities. The testing duration spanned 90 days, from March 1 to May 30, 2019, capturing diverse usage patterns typical of research and development environments. Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 109 Data Collection Methodology The data collection methodology implemented comprehensive monitoring across multiple dimensions to capture the system performance under realistic operational conditions. Sensor data streams were continuously collected at high temporal resolution, including PIR sensor motion events timestamped at 100 Hz, RGB-D camera frames captured at 30 FPS with human detection annotations, environmental sensor readings (temperature, humidity, and light levels) sampled at 1 Hz, power consumption measurements from LED drivers recorded at 10 Hz, and user interaction events, including manual overrides and preference adjustments. Ground truth collection was maintained through trained observer logs during peak occupancy hours (9 AM–5 PM) to provide validation data for the algorithm performance evaluation. Observer annotations included precise timestamps for human presence and absence, counts and positions of occupants, documentation of unusual events or system behaviors, and daily user satisfaction ratings on a 1–5 scale across multiple comfort dimensions. Controlled anomaly injection testing was performed to evaluate the performance of the detection system under various failure conditions. The testing protocol included 47 hardware failure events (sensor disconnections and LED failures), 23 software anomaly events (communication timeouts and processing delays), 18 security-related events (unusual access patterns and unauthorized system changes), and 12 environmental anomaly conditions (extreme temperature/humidity situations). The complete dataset comprises 2,160 hours of continuous operation generating 1.2TB sensor data, providing comprehensive coverage for algorithm development, validation, and performance evaluation. RESULTS AND ANALYSIS Human Detection Performance Evaluation The multi-modal human detection system demonstrated superior performance compared to individual sensor approaches across all three deployment environments, validating the effectiveness of the sensor fusion approach. The comprehensive evaluation included precision, recall, F1-score, and falsepositive rate metrics calculated against manually annotated ground-truth data. As shown in Table 1, the sensor fusion approach achieved an average F1-score of 0.948 across all environments, representing a 12.3% improvement over vision-only detection and a 23.7% improvement over PIR-only detection. Table 1: Human Detection Performance Comparison Method Environment Precision Recall F1-Score FPR PIR Only Office 0.823 0.867 0.844 0.142 Residential 0.798 0.889 0.841 0.167 Laboratory 0.856 0.834 0.845 0.119 Vision Only Office 0.887 0.901 0.894 0.089 Residential 0.912 0.878 0.895 0.076 Laboratory 0.923 0.889 0.906 0.071 Sensor Fusion Office 0.943 0.951 0.947 0.048 Residential 0.934 0.967 0.950 0.039 Laboratory 0.952 0.943 0.947 0.041 False-positive rates were reduced by an average of 66.2% compared to the PIR sensors alone. Figure 2: Human Detection Accuracy by Time of Day (Office Environment) Temporal analysis of the detection performance, illustrated in Figure 2, revealed significant variations throughout the day, with peak performance during normal occupancy hours (10 AM–4 PM) when lighting conditions are Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 110 optimal for computer vision processing. The sensor fusion approach consistently maintained superior performance across all time periods, demonstrating particular advantages during challenging lighting conditions, such as dawn and dusk transitions. Environmental factor analysis showed that the office environment achieved the highest detection accuracy owing to controlled lighting conditions and predictable occupancy patterns, while the residential environment presented greater challenges owing to variable lighting and diverse occupancy behaviors. The laboratory environment demonstrated intermediate performance with good overall accuracy, despite 24/7 operational requirements and varying activity levels. Energy Efficiency Analysis Energy consumption measurements demonstrated substantial improvements over conventional lighting systems across all deployment environments, validating the energy efficiency benefits of intelligent lighting control. The analysis compared the baseline energy consumption of the conventional fluorescent and fixed LED systems with that of the proposed smart lighting system over identical operational periods. Table 2: Energy Consumption Comparison Environment Baseline (kWh) Smart System (kWh) Savings (kWh) Percentage Office 234.7 132.8 101.9 43.4% Residential 89.3 54.2 35.1 39.3% Laboratory 312.5 175.6 136.9 43.8% Average 212.2 120.9 91.3 42.3% Table 2 shows that the office environment achieved 43.4% energy savings, reducing monthly consumption from 234.7 kWh to 132.8 kWh while maintaining equivalent or superior illumination quality. The residential environment demonstrated 39.3% energy savings, whereas the laboratory environment achieved the highest percentage savings of 43.8%. Figure 3: Energy Consumption Analysis: (a) Monthly comparison by environment, (b) Weekly pattern in office environment Figure 3 reveals that the greatest savings occurred during weekends and non-business hours, when occupancy-based control could significantly reduce unnecessary illumination. Peak energy efficiency was observed during periods of variable occupancy, when the intelligent system could dynamically adapt to actual usage patterns, whereas conventional systems operated at fixed output levels. Figure 4: Energy Savings Breakdown and Monthly Trends Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 111 The detailed energy savings analysis in Figure 4 shows that occupancybased switching contributes 67.3% of the total savings, adaptive brightness control contributes 21.7%, circadian rhythm optimization provides 8.2%, and predictive maintenance accounts for 2.8% of the energy savings. Anomaly Detection Performance The ensemble anomaly detection system demonstrated robust performance across different types of system anomalies and operational conditions, validating the effectiveness of the multi-algorithm approach for comprehensive system monitoring. Table 3: Anomaly Detection Performance by Category Anomaly Type Count Precision Recall F1-Score MTTD (min) Hardware Failures 47 0.923 0.851 0.885 3.2 Software Anomalies 23 0.870 0.913 0.891 1.8 Security Events 18 0.889 0.944 0.916 4.1 Environmental 12 0.917 0.833 0.873 5.7 Overall 100 0.892 0.870 0.881 3.4 As presented in Table 3, hardware failure detection achieved 92.3% precision and 85.1% recall with a mean time to detection of 3.2 minutes, whereas software anomaly detection achieved 87.0% precision and 91.3% recall with the fastest detection time of 1.8 minutes. The overall system performance achieved 89.2% precision and 87.0% recall across all the anomaly categories. Figure 5: Anomaly Detection Performance: (a) Performance by category, (b) Ensemble weight optimization Figure 5 illustrates the performance across different anomaly types, and shows the ensemble weight optimization results. The optimal ensemble configuration assigns the highest weight (0.45) to the LSTM autoencoder owing to its superior performance in detecting temporal anomalies, which represent the majority of system failures in smartlighting applications. User Satisfaction and System Reliability User satisfaction was evaluated through daily surveys and behavioral analysis across all deployment environments, providing a comprehensive assessment of system effectiveness from the occupant perspective. Table 4: User Satisfaction Metrics (1–5 Scale) Metric Office Residential Laboratory Visual Comfort 4.2 ± 0.3 4.4 ± 0.2 4.1 ± 0.4 Energy Awareness 4.6 ± 0.2 4.3 ± 0.3 4.5 ± 0.2 System Responsiveness 4.3 ± 0.3 4.5 ± 0.2 4.2 ± 0.3 Lighting Quality 4.1 ± 0.4 4.4 ± 0.3 4.0 ± 0.4 Overall Satisfaction 4.3 ± 0.2 4.4 ± 0.2 4.2 ± 0.3 Table 4 shows high user satisfaction across all categories, with energy awareness achieving the highest scores (4.5 ± 0.2 average), reflecting strong user appreciation for environmental benefits and energy efficiency improvements. Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 112 Figure 6: User Satisfaction Analysis: (a) Satisfaction by category and environment, (b) System uptime comparison over 90-day period Figure 6 demonstrates consistently high satisfaction scores across environments and shows an improved system uptime with predictive maintenance capabilities. Behavioral analysis revealed a 73.2% reduction in manual lighting adjustments, 18.7% increase in workspace utilization under optimal lighting conditions, 24.1% improvement in reported sleep quality (residential environment), and 15.3% reduction in eye strain complaints. Comparative Analysis and Novel Contributions This research introduces several novel techniques that contribute to the enhanced performance observed in experimental validation. Building on existing commercial solutions and research foundations, the proposed system incorporates three key innovations that distinguish it from previous approaches. The multi-modal sensor fusion algorithm represents a significant methodological advancement by implementing adaptive weighting mechanisms that respond to the environmental conditions and sensor reliability. Unlike traditional fixed-weight approaches, this dynamic fusion technique optimizes detection performance across various operational scenarios. The ensemble anomaly detection framework introduces a novel combination of the Isolation Forest, LSTM Autoencoder, and One-Class SVM algorithms specifically calibrated for smart lighting applications. This approach addresses the gap in existing commercial systems that typically lack comprehensive anomaly detection capabilities for predictive maintenance. The adaptive learning system for user-preference integration represents an innovative approach for personalizing lighting control while maintaining energy-efficiency objectives. This technique builds on the basic programmable features found in commercial systems by implementing continuous learning and adaptation capabilities. A comparison with existing commercial and research solutions revealed the significant advantages of the proposed approach across multiple performance dimensions. Figure 7: Performance Comparison with Existing Solutions: (a) Energy savings comparison, (b) Energy savings vs detection accuracy scatter plot Figure 7 illustrates that the proposed system achieves superior performance in terms of both energy efficiency (42.3% average savings) and detection accuracy (94.7%) compared to existing commercial solutions, such as Philips Hue (23% savings, 78% accuracy) and LIFX Professional (28% savings, 82% accuracy). Comprehensive anomaly detection capabilities represent a significant advancement over existing systems that provide limited or no anomaly-detection functionality. Vangalapat T Euro. J. Adv. Engg. Tech., 2019, 6(8):105-117 113 DISCUSSION AND IMPLICATIONS The experimental results demonstrate significant improvements across all the evaluated metrics through the application of novel techniques in multi-modal sensing, adaptive control, and predictive maintenance. Rather than displacing existing commercial solutions, these innovations provide complementary capabilities that can enhance the current lighting system architectures. The 42.3% average energy savings achieved in this study represent a significant advancement over existing commercial solutions, which typically achieve 20– 30% energy reduction through basic occupancy sensing and manual control optimization. This improvement is attributed to the novel combination of multi-modal sensor fusion, AI-driven adaptive control, and predictive maintenance capabilities, which build upon the solid foundations established by existing commercial systems. The multi-modal sensor fusion approach proved particularly effective, achieving an average detection accuracy of 94.7% while reducing false posi-tives by 66.2% compared to single-sensor systems. This improvement demonstrates the value of combining complementary sensor technologies, where PIR sensors provide low-power continuous monitoring and computer vision systems offer high-accuracy validation and detailed occupancy tracking. These advances build upon the proven reliability of individual sensor approaches, while addressing their respective limitations through intelligent fusion techniques. The ensemble anomaly detection system demonstrated robust performance with 89.2% precision in identifying system anomalies, while maintaining acceptably low false alarm rates. The weighted combination of the three distinct algorithms provides comprehensive coverage of different failure modes while adapting to the specific characteristics of smart lighting system operations. Scalability and Economic Analysis Scalability analysis indicates that the proposed system architecture supports practical deployment in larger building installations through distributed processing and hierarchical control structures. Edge computing capabilities reduce bandwidth requirements (2.3 Mbps per 100 fixtures), while enabling autonomous operation during network disruptions. Hardware costs an average of $47 per fixture, including sensors and processing, with payback periods of 18–24 months based on energy savings alone. The economic analysis demonstrates a positive return on investment when considering the total cost of ownership, including maintenance savings ($2,340 in prevented emergency repairs over 90 days) and productivity improvements from enhanced user comfort and lighting quality. Limitations and Challenges There are several limitations that warrant consideration for practical deployment. Privacy concerns related to computer vision components require careful implementation of data-handling policies and user consent procedures, particularly in residential and sensitive commercial environments. Environmental dependencies affect the system performance under varying conditions, necessitating robust calibration procedures and adaptive algorithm parameters. These research contributions have broader implications for building automation and sustainable technology development. The demonstrated energysaving potential, when applied across the building sector representing 40% of the total energy consumption, could contribute significantly to global energy conservation efforts. CONCLUSION This study presents a comprehensive Smart Lighting System framework that demonstrates substantial improvements in energy efficiency, user comfort, and system reliability through intelligent human detection and AIbased anomaly detection. Experimental validation across three distinct building environments over 90-day deployment periods provides robust evidence of system effectiveness under realistic operational conditions. The key technical contributions include a multi-modal sensor fusion algorithm achieving 94.7% human detection accuracy, an ensemble anomaly detection system with 89.2% precision across diverse failure modes, adaptive lighting control that reduces energy consumption by 42.3% on average, and a real-time processing architecture suitable for practical deployment in diverse building environments. The practical achievements demonstrated comprehensive field validation in realistic operational environments, user satisfaction improvements across multiple comfort dimensions (4.3/5.0 average rating), proven reliability and maintenance cost reductions through predictive capabilities, and significant energy savings that build upon existing commercial solutions through novel technical approaches. This broader impact contributes to sustainable building technologies by demonstrating practical approaches to significant energy reduction while improving occupant comfort and safety. System architecture and algorithms provide a foundation for next-generation building automation systems capable of autonomous adaptation to user needs while maintaining optimal efficiency. The integration of artificial intelligence with IoT sensor networks represents a paradigm shift from reactive to predictive building management systems. The demonstrated energy savings of over 40% indicate substantial potential for reducing the building sector energy consumption and contributing to global sustainability goals. Experimental evidence validates the effectiveness of AI-driven approaches for building automation and provides a practical roadmap for implementing intelligent systems in diverse building environments. As sensor technologies