Edge-Enabled Smart Street Lighting System for Accident and Crime Detection in Real Time Adaptive Risk Scoring and Deep Spatial-Temporal Models

Abstract

The problem of road accidents and criminal activities has become one of the main concerns of urban safety. There is no contextual awareness and adaptive response capability in current street lighting and street surveillance systems, and they waste energy, delay in responding to emergencies. In this paper, the limitations of these methods are overcome and a Smart Street Lighting Framework is proposed to detect accidents and crimes in real time using Edge-AI. The framework integrates object detection, multi-object tracking and spatio-temporal behaviours analysis in an Internet of Things (IoT)-based perception–decision framework at the network edge. An important contribution is the suggested Adaptive Spatio-Temporal Risk Score (ASTRS) that is dynamically computed based on the density of objects, the way they are moving, the probability of anomalous movement, and historical trends of incidents. With the computed risk score, intelligent control of street light can be achieved, emergency call can be generated automatically and closed loop system of perception and action can be realized. The model compression and optimization techniques used with TensorRT enable real-time operations and a latency of just 28ms and a throughput of 34FPS on edge devices. Experimental results on UCF-Crime dataset and a real-world outdoor dataset show its superior performance with the detection accuracy of 92.6% and classification accuracy of 90.8%. Furthermore, the outcomes of the ablation studies further support the effectiveness of the different elements of the frameworks, which can enhance the safety, scalability and energy efficiency within the urban context.

Introduction

The rapid growth of urban populations has increased the demand for smart city technologies that improve safety, sustainability, and efficiency. Traditional street lighting systems, which rely on fixed schedules or simple motion sensors, often waste energy and lack the ability to respond intelligently to changing urban conditions or emergencies such as accidents and crimes. Advances in Artificial Intelligence (AI), Internet of Things (IoT), and edge computing provide new opportunities to transform conventional street lighting into adaptive and intelligent infrastructure.

This paper proposes an Edge-AI Enabled Smart Street Lighting Framework that integrates surveillance, risk assessment, and adaptive lighting control into a unified system. Unlike existing smart lighting solutions that primarily focus on energy savings, the proposed framework can detect accidents, crimes, and unusual activities in real time and automatically adjust street lighting and emergency responses accordingly. By processing video data on edge devices near surveillance cameras, the system reduces latency, bandwidth usage, and dependence on cloud servers while improving privacy and responsiveness.

The framework employs a closed-loop perception–decision–action architecture consisting of four main stages:

1. Object Detection: Uses YOLOv8-S to detect pedestrians, vehicles, and suspicious objects.
2. Multi-Object Tracking: Uses ByteTrack to track object movements and maintain identities across video frames.
3. Temporal Behavior Analysis: A GRU-based neural network analyzes object trajectories and classifies activities as normal behavior, accidents, or crimes.
4. Adaptive Risk Assessment: A novel Adaptive Spatio-Temporal Risk Score (ASTRS) combines object density, motion patterns, anomaly probabilities, and historical risk information to dynamically assess environmental risk levels.

Based on the calculated risk score, the IoT control layer can automatically increase street-light brightness, generate emergency alerts, and support rapid response to critical incidents. The framework is optimized for deployment on resource-constrained edge devices using lightweight deep learning models, model compression techniques, and TensorRT acceleration, enabling real-time operation.

The literature review highlights that existing smart lighting systems mainly focus on energy efficiency and occupancy detection, while AI surveillance systems often operate independently of urban infrastructure. Current solutions typically lack integration between perception, risk assessment, and infrastructure control. The proposed framework addresses these gaps by combining AI surveillance, temporal behavior analysis, contextual risk assessment, and adaptive lighting within a single architecture.

For evaluation, the study uses the UCF-Crime dataset, containing 1,900 surveillance videos across 13 anomaly categories such as accidents, assaults, robberies, and vandalism, along with a custom outdoor surveillance dataset collected under varying weather and lighting conditions. Data preprocessing included frame extraction, resizing, noise reduction, augmentation, and detailed annotations for detection and tracking tasks.

Experimental results demonstrate strong performance, achieving over 92% event detection accuracy, over 90% classification accuracy, and real-time processing at approximately 34 frames per second with low latency. Comparative studies and ablation experiments confirm that the integration of object detection, tracking, temporal modeling, and adaptive risk scoring significantly improves accuracy, responsiveness, and operational efficiency compared to existing approaches.

Conclusion

Overall, the proposed framework offers a comprehensive smart city solution that enhances both urban safety and energy efficiency by intelligently combining edge AI surveillance, risk-aware decision-making, and adaptive street lighting control in real time.

References

[1] M. Yigit, V. C. Gungor, and S. Baktir, “Smart Street lighting systems: A systematic review of technologies, opportunities, and challenges,” Sustainable Cities and Society, vol. 99, p. 104912, 2024. [2] E. Badidi, M. A. Alturki, and S. Zeadally, “Edge-AI enabled video analytics in smart cities: Opportunities, applications, and challenges,” IEEE Access, vol. 11, pp. 45432–45455, 2023. [3] S. Khemakhem et al., “IoT-based smart public street lighting systems: A survey,” Results in Engineering, vol. 19, 2024. [4] S. Aminiyeganeh, A. Rezvani, and M. R. Mosavi, “AI-based video analytics for smart cities: A comprehensive review,” Sustainable Cities and Society, vol. 102, p. 105187, 2024 [5] M. M. Amrutha and S. K. P, “A deep learning approach for intelligent traffic monitoring using computer vision,” IEEE Access, vol. 11, pp. 104567–104580, 2023. [6] S. Vadi, R. Mahapatra, and P. K. Mishra, “IoT-enabled smart surveillance using MQTT for real-time monitoring,” IEEE Internet of Things Journal, vol. 10, no. 8, pp. 6789–6801, 2023. [7] A. K. Mishra and S. Sharma, “IoT-based intelligent street lighting for smart cities,” IEEE Access, vol. 11, pp. 45678–45690, 2023. [8] R. Gupta, S. Kumar, and P. Singh, “Edge-enabled smart lighting for energy-efficient urban infrastructure,” IEEE Internet of Things Journal, vol. 10, no. 12, pp. 10234–10248, 2023. [9] H. Zhang, Y. Wang, and L. Liu, “Deep learning-based video surveillance for smart cities,” IEEE Access, vol. 11, pp. 112345–112367, 2023. [10] Z. Liu et al., “Video Swin Transformer,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 2, pp. 1234–1248, 2024. [11] W. Sultani, C. Chen, and M. Shah, “UCF-Crime: A large-scale dataset for anomaly detection,” in Proc. IEEE CVPR, 2018, pp. 6479–6488. [12] G. Jocher et al., “Ultralytics YOLOv8,” GitHub repository, 2023. [Online]. Available: https://github.com/ultralytics/ultralytics [13] Y. Zhang et al., “ByteTrack: Multi-object tracking by associating every detection box,” in Proc. ECCV, 2022, pp. 1–21. [14] K. Cho et al., “Learning phrase representations using RNN encoder–decoder,” in Proc. EMNLP, 2014. [15] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 3rd ed., Pearson, 2023. [16] NVIDIA Corporation, “TensorRT: High-performance deep learning inference optimizer,” NVIDIA Developer Guide, 2023. [17] H. Nguyen, A. Y. Yang, and P. Bajcsy, “Smart surveillance systems for public safety: Trends and challenges,” IEEE Internet of Things Journal, vol. 9, no. 18, pp. 16789–16805, 2022. [18] X. Wang, L. Ma, and W. Liu, “Edge intelligence for smart cities: A survey on video analytics and applications,” IEEE Internet of Things Journal, vol. 11, no. 5, pp. 5678–5695, 2024. [19] Y. Zhou, Z. Zhao, and J. Wang, “Energy-efficient intelligent street lighting systems using AI and IoT: A review,” Sustainable Cities and Society, vol. 101, p. 105089, 2024. [20] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137–1149, 2017. [21] J. Carreira and A. Zisserman, “Quo Vadis, Action Recognition? A new model and the Kinetics dataset,” in Proc. IEEE CVPR, 2017, pp. 6299–6308. [22] A. Dosovitskiy et al., “An image is worth 16×16 words: Vision transformer for image recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 2, pp. 150–165, 2023. [23] M. Zaharia et al., “Accelerating deep learning inference using TensorRT and edge optimization techniques,” IEEE Access, vol. 11, pp. 98765–98780, 2023. [24] S. E. Bibri, “Smarter eco-cities and their leading-edge artificial intelligence and AIoT solutions for environmental sustainability: A systematic review,” Sustainable Cities and Society, vol. 100, 2024. [25] K. S. Velaga et al., “Edge AI for smart cities: Foundations, challenges, and future directions,” Smart Cities, vol. 8, no. 6, 2025. [26] T. Dias, H. M. Faisal, and A. M. Khan, “From data to action: AI and IoT-driven smart city solutions,” IEEE Access, vol. 11, pp. 121345–121360, 2023.

Copyright

Copyright © 2026 Komal Yadav , Dr. Anita Pal . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.