Surveillance System for Scotopic Vision Using HDV in Deep Learning for Road-Safety

Abstract

In the Low Light image enhancement, considerable progress has been achieved, yet the emphasis has predominantly been on images captured under normal lighting conditions, overlooking the critical aspect of enhancing images in low-light environments. This oversight is particularly significant in fields like nighttime surveillance and autonomous driving, where obtaining clear images in low-light conditions is crucial. Acknowledging the limitations of traditional methods in such scenarios, researchers have turned to innovative techniques, prominently leveraging Generative Adversarial Networks (GANs) to enhance images in low-light settings. Traditional approaches often struggle to improve image quality due to reduced visibility and increased noise levels inherent in low-light conditions.By integrating GANs into the enhancement process, researchers can address these challenges by generating synthetic brightened versions of low-light images. This integration not only enriches the dataset but also enhances the model\'s ability to handle varying lighting conditions. GANs, consisting of a generator and discriminator network, collaborate to produce realistic enhancements of low-light images, thereby augmenting the training data.This approach enables the model to adapt more effectively to real-world scenarios, such as nocturnal surveillance or navigating dark environments during autonomous driving

Introduction

1. Background & Motivation

Urban rail stations face increasing challenges with crowd control, safety, and efficient surveillance. Traditional CCTV systems rely on human monitoring, which is prone to fatigue, slow response, and error. To overcome these limitations, the project integrates YOLOv5—a real-time object detection model—into existing surveillance infrastructure for automated crowd detection and behavior analysis.

The system enables proactive responses by detecting, counting, and tracking people, and generating alerts during abnormal crowd behavior or density spikes, thereby improving public safety and resource efficiency.

2. Problem Statement

Current surveillance systems are reactive and ineffective at preemptively identifying crowd risks. They lack real-time analytics to assess crowd density or movements, making them inefficient in critical situations. This project addresses the need for an intelligent, real-time monitoring solution that reduces manual oversight and enables timely interventions.

3. Literature Review

The review highlights the evolution of AI-powered surveillance, with YOLO (You Only Look Once) frameworks being central to this transformation:

• YOLOv1–v4: Improved object detection speed and accuracy, critical for real-time applications.

• YOLOv5: Offers high performance in dense and dynamic settings, with better accuracy, lightweight models, and faster inference.

• Other models (e.g., SSD, Faster R-CNN) are less effective for real-time crowd analysis due to slower speeds or difficulty detecting small/overlapping objects.

• YOLOv5 is identified as ideal for scalable, adaptable, and real-time crowd monitoring.

4. Proposed Methodology

A. System Workflow:

• Capture CCTV footage from railway stations.

• Preprocess frames and input them into YOLOv5 for crowd detection.

• Estimate density and trigger alerts based on crowd thresholds.

B. Development Phases:

1. Planning: Select pilot locations, coordinate with stakeholders, and ensure ethical and privacy compliance.

2. Design: Leverages existing infrastructure, automates alert generation, and uses Python/Flask/MySQL for system architecture.

3. Development: Set up hardware/software, train security personnel, and collect real-time video for model testing.

4. Implementation: Integrate system components—CCTV input, YOLOv5 processing, backend, and UI.

C. Architecture Layers:

• Data Collection Layer: Captures real-time video.

• Processing Layer: Prepares and analyzes data using YOLOv5.

• Decision Layer: Triggers alerts for abnormal crowd levels.

• Interface Layer: Provides a user-friendly web interface for monitoring.

D. Tools & Integration:

• Tech stack: Python, Flask, MySQL, YOLOv5.

• APIs: Include weather/location awareness, SMS/email alerts, and cloud storage for secure, responsive system operation.

5. Outcomes

• Improved Surveillance: Real-time monitoring enhances detection and response.

• Resource Optimization: Reduces dependency on continuous human oversight.

• Cost-Efficient: Uses existing infrastructure and reduces labor costs.

• Societal Benefits: Enhances public safety and supports smart city planning.

6. Results

Testing showed that the system:

• Reliably detected and tracked individuals in real-time.

• Accurately identified crowd anomalies and triggered alerts.

• Enhanced operational awareness and reduced the need for constant human monitoring.

• Was effective across public spaces like stations, offices, and events.

Conclusion

In conclusion, the integration of Generative Adversarial Networks (GANs) for image enhancement presents a promising avenue for improving visual quality across various applications. Through adversarial training, GAN models can effectively learn to generate realistic enhancements, such as increasing clarity, enhancing details, and reducing noise, in low-quality or degraded images. This approach offers significant advantages over traditional methods, as it enables the generation of visually appealing results while preserving important image characteristics. Furthermore, the versatility of GAN-based image enhancement extends to diverse domains, including photography, medical imaging, surveillance, and more. As research in GANs continues to advance, with innovations in model architectures, training techniques, and applications, the potential for further improving image enhancement capabilities remains substantial. Ultimately, the adoption of GANs in image enhancement holds promise for enhancing visual content, driving advancements in various fields, and enriching user experiences across digital platforms. Future work can focus on improving the stability and efficiency of GAN training, integrating attention mechanisms for better detail preservation, and exploring lightweight GAN architectures for realtime low-light enhancement on mobile devices. Additionally, incorporating multi-modal inputs (e.g., infrared or depth data) and leveraging unsupervised or self-supervised learning could further enhance performance and generalizability in diverse realworld scenarios.

References

[1] A. M. Reza, ‘‘Realization of the contrast limited adaptive histogram equalization (CLAHE) for realtime image enhancement,’’ J. VLSI Signal Process. Syst. Signal, Image Video Technol., vol. 38, no. 1, pp. 35–44, 2021. [2] H. Ibrahim and N. S. P. Kong, ‘‘Brightness preserving dynamic histogram equalization for image contrast enhancement,’’ IEEE Trans. Consum. Electron., vol. 53, no. 4, pp. 1752–1758, Nov. 2020. [3] Z.-U. Rahman, D. J. Jobson, and G. A. Woodell, ‘‘Multi-scale retinex for color image enhancement,’’ in Proc. IEEE Int. Conf. Image Process., vol. 3, Sep. 2020. [4] A. B. Petro, C. Sbert, and J.-M. Morel, ‘‘Multiscale retinex,’’ Image Process. Line, vol. 4, pp. 71–88, Apr. 2021. [5] C. Dong, C. C. Loy, K. He, and X. Tang, ‘‘Image superresolution using deep convolutional networks,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 38, no. 2, pp. 295–307, Apr. 2023. [6] W. Ren, S. Liu, L. Ma, Q. Xu, X. Xu, X. Cao, J. Du, and M.-H. Yang, ‘‘Low-light image enhancement via a deep hybrid network,’’ IEEE Trans. Image Process., vol. 28, no. 9, pp. 4364– 4375, Sep. 2021. [7] K. G. Lore, A. Akintayo, and S. Sarkar, ‘‘LLNet A deep autoencoder approach to natural low-light image enhancement,’’ Pattern Recognit., vol. 61, pp. 650–662, Jan. 2021. [8] C. Li, C. Guo, and C. L. Chen, ‘‘Learning to enhance low-light image via zero-reference deep curve estimation,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 8, pp. 4225–4238, Apr. 2022. [9] Z. Jiang, H. Li, L. Liu, A. Men, and H. Wang, ‘‘A switched view of retinex Deep self- regularized low-light image enhancement,’’ Neurocomputing, vol. 454, pp. 361–372, May 2021. [10] M. Ravikumar, B. J. Shivaprasad, and D. S. Guru, ‘‘Enhancement of MRI brain images using notch filter based on discrete wavelet transform,’’ Int. J. Image Graph., vol. 22, no. 1, Jan. 2022. [11] O. Ronneberger, P. Fischer, and T. Brox, ‘‘U-Net Convolutional networks for biomedical image segmentation,’’ in Proc. Int. Conf. Med. Image Comput. Comput.-Assist. Intervent. Cham, Switzerland Springer, 2021. [12] S. Parthasarathy and P. Sankaran, ‘‘An automated multi scale retinex with color restoration for image enhancement,’’ in Proc. Nat. Conf. Commun. (NCC), Feb. 2021. [13] C. Guo, C. Li, J. Guo, C. C. Loy, J. Hou, S. Kwong, and R. Cong, ‘‘Zeroreference deep curve estimation for low-light image enhancement,’’ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2020. [14] W. Wang, X. Wu, X. Yuan, and Z. Gao, ‘‘A formatted overview-based module of low-light image enhancement method,’’, 2020. [15] C. Wei, W. Wang, W. Yang, and J. Liu, ‘‘Deep retinex decomposition for low-light enhancement,’’ 2020. [16] W. Wang, X. Wu, X. Yuan, and Z. Gao, ‘‘An experiment-based review of low-light image enhancement method,’’ IEEE Access, vol. 8, , 2020 [17] C. Wei, W. Wang, W. Yang, and J. Liu, ‘‘Deep retinex decomposition for low-light enhancement,’’ 2020. [18] W. Wang, X. Wu, X. Yuan, and Z. Gao, ‘‘An experiment-based review of low-light image enhancement method,’’ IEEE Access, vol. 8, pp. 884–887, 2020 [19] W. Wang, J. Liu, ‘‘Deep lambda and gamma decomposition for low-light enhancement,’’ 2020. [20] Yuan, and Z. Gao, ‘‘An experiment-based structural review of low-light image enhancement method,’’ IEEE Access, vol. 10, , 2020 [21] J. Liu, ‘‘Deep red blue decomposition for lowlight enhancement,’’ 2023. [22] C. Li, C. Guo, and C. L. Chen, ‘‘Learning to enhance low-light image via zero-reference deep curve estimation,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 8, , Apr. 2022 [23] Arjun S. Sarkar, ‘‘LLNet An encrypted approach to natural low-light image enhancement,’’., vol. 72, , Jan. 2024. [24] C. Li, C. Guo, ‘‘Deep learning to enhance lowlight image via zero-tolerance deep curve tunnel estimation,”, Apr. 2022. [25] Z. Jiang, H. Li, L. Liu, A. Men, and H. Wang, ‘‘A type-switched feature of retinex Deep self- regularized low-light image enhancement,’’may 2024.

Copyright

Copyright © 2025 Dr. R. G. Suresh Kumar, Dr. V. Kavitha, S. Abishek . 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.