Real-Time Substation Detection and Monitoring Security Alarm System
In real-world scenarios, conventional security alarms are commonly used to monitor and record occurrences. However, these alarms are often impractical as they lack the ability to warn security administrators in real-time when security threats are present. To address this issue, this project proposes...
Saved in:
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Conference Paper |
| Published: |
Institute of Electrical and Electronics Engineers Inc.
2024
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In real-world scenarios, conventional security alarms are commonly used to monitor and record occurrences. However, these alarms are often impractical as they lack the ability to warn security administrators in real-time when security threats are present. To address this issue, this project proposes an automated continuous security surveillance system utilizing Raspberry Pi 3B and Raspberry Pi Camera, along with deep learning tools namely Open-Source Computer Vision Library (OpenCV), You Only Look Once tiny version 4 (YOLOv4-tiny), Google Drive, Colab Notebook, and DarkNet framework. The proposed system aims to detect specific classes, such as people, cars, or fires, within a designated range and promptly send out alert notifications and sound an alarm in real-time. To achieve this, the YOLOv4-tiny machine learning model is used for object detection, and OpenCV is employed for image pre-processing tasks like scaling and normalization [21]. To overcome the computational limitations of the Raspberry Pi 3B, the YOLOv4-tiny model is trained on a custom dataset using the DarkNet framework, and Colab Notebook is utilized for training and optimizing the model, which is then saved on Google Drive for easy access. Various techniques like model quantization and network pruning are employed to improve the system's performance and efficiency. The experimental results demonstrate that the proposed system achieves high accuracy in detecting specific classes while remaining computationally efficient. Overall, this project offers a reliable and efficient solution by integrating the concepts of object recognition and detection, showcasing its potential for various practical applications. � 2023 IEEE. |
|---|