Scholarlink

This paper presents an innovative approach to intruder detection for night-time surveillance. The primary objective of this study is to develop an integrated system for a three-stage approach comprising low-light image enhancement, human detection, and face recognition, addressing a critical gap in current surveillance technology. In the first stage of the proposed algorithm, the overall visibility in low-light images was improved. Specifically, Contrast Enhancement Algorithm Based on Gap Adjustment for Histogram Equalization (CegaHE) was utilized to evenly distribute the lower pixel range of the input image histogram. This is followed by Contrast-Limited Adaptive Histogram Equalization (CLAHE). CLAHE adjusts the over-enhanced areas of the CegaHE output, providing a balanced light correction. In the second stage, SSD MobileNet V2 was used to detect humans in the corrected input with high precision. This particular deep learning model was selected due to its lightweight and the potentially easy implementation in low-processing power single-board computers, such as Raspberry Pi. Finally, to differentiate residents from intruders, fine-tuned Siamese network with triplet loss was utilized. The proposed overall system achieves a significant improvement in intruder detection in nighttime surveillance, with a detection accuracy of 90.6%. This study highlights the importance of integrating image enhancement with advanced detection and recognition algorithms, advancing nighttime surveillance technology. Our work contributes to the growth of computer vision knowledge providing practical insights for security and monitoring systems. We aim to foster understanding and promote more effective surveillance solutions by addressing intruder detection in low-light conditions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Low-Cost Intruder Detection for Night Surveillance Using Advanced Image Enhancement and Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • M. Thanoraj,
  • S. Prannavi,
  • J. Thuvaraga,
  • M. B. Dissanayake

摘要

This paper presents an innovative approach to intruder detection for night-time surveillance. The primary objective of this study is to develop an integrated system for a three-stage approach comprising low-light image enhancement, human detection, and face recognition, addressing a critical gap in current surveillance technology. In the first stage of the proposed algorithm, the overall visibility in low-light images was improved. Specifically, Contrast Enhancement Algorithm Based on Gap Adjustment for Histogram Equalization (CegaHE) was utilized to evenly distribute the lower pixel range of the input image histogram. This is followed by Contrast-Limited Adaptive Histogram Equalization (CLAHE). CLAHE adjusts the over-enhanced areas of the CegaHE output, providing a balanced light correction. In the second stage, SSD MobileNet V2 was used to detect humans in the corrected input with high precision. This particular deep learning model was selected due to its lightweight and the potentially easy implementation in low-processing power single-board computers, such as Raspberry Pi. Finally, to differentiate residents from intruders, fine-tuned Siamese network with triplet loss was utilized. The proposed overall system achieves a significant improvement in intruder detection in nighttime surveillance, with a detection accuracy of 90.6%. This study highlights the importance of integrating image enhancement with advanced detection and recognition algorithms, advancing nighttime surveillance technology. Our work contributes to the growth of computer vision knowledge providing practical insights for security and monitoring systems. We aim to foster understanding and promote more effective surveillance solutions by addressing intruder detection in low-light conditions.