Scholarlink

<p>This research presents LAXMI, a secure and lightweight hybrid video analytics pipeline for motion detection and real-time object classification. The end objective is to design a system that can perform computationally efficient scene analysis in the edge but securely transmit full video data to the cloud for archival or additional processing. The pipeline aims to minimize processing delay, minimize bandwidth usage, and guard surveillance content with a combination of optimized machine learning and cryptographic methods. LAXMI integrates a temporal gradient-based motion detection (TGMD) algorithm at the edge, which precisely identifies motion segments within continuous video streams. Upon detection of motion, the system conducts YOLOv8 object detection for classification and annotation of related entities such as people, cars, animals, and other moving objects. One of the key novelties is employing a hybrid tokenization strategy where the video is encrypted with AES-GCM, and the encryption key is securely transported using RSA public-key cryptography. This ensures data confidentiality as well as secure edge-to-cloud communications. The edge server then generates a JWT-based token for authenticating the cloud upload. Unlike conventional IoT frameworks, LAXMI explicitly targets scalable surveillance workloads at city or enterprise level, where real-time analytics over distributed networks, parallel processing across heterogeneous edge nodes, and secure high-speed archival to the cloud require coordination similar to high-performance computing (HPC)/supercomputing environments. The design leverages parallel inference and distributed edge-cloud scheduling to meet ultra-low-latency and high-throughput demands. An 8-week duration performance test reveals that the edge-based motion detection subsystem achieved a 92.7% accuracy and F1-score of 96.2%, while the YOLOv8 classifier achieved 94.8% accuracy and F1-score of 94.4%. Further, the edge pipeline outperformed the cloud baseline across processing latency, energy expense, and bandwidth consumption constantly, thereby setting the effectiveness of the proposed LAXMI architecture in actual smart surveillance applications.</p>

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

LAXMI: a novel AI-based system for smart, energy-efficient, and secure video footage processing

  • 期刊论文

  • 发表日期：

    2025年10月19日

查看全文
  • Tathagata Bhattacharya,
  • Srikant Ponaganti,
  • Adithya Peddi Vardhan,
  • Yaswanth Sai Venkatesan

摘要

This research presents LAXMI, a secure and lightweight hybrid video analytics pipeline for motion detection and real-time object classification. The end objective is to design a system that can perform computationally efficient scene analysis in the edge but securely transmit full video data to the cloud for archival or additional processing. The pipeline aims to minimize processing delay, minimize bandwidth usage, and guard surveillance content with a combination of optimized machine learning and cryptographic methods. LAXMI integrates a temporal gradient-based motion detection (TGMD) algorithm at the edge, which precisely identifies motion segments within continuous video streams. Upon detection of motion, the system conducts YOLOv8 object detection for classification and annotation of related entities such as people, cars, animals, and other moving objects. One of the key novelties is employing a hybrid tokenization strategy where the video is encrypted with AES-GCM, and the encryption key is securely transported using RSA public-key cryptography. This ensures data confidentiality as well as secure edge-to-cloud communications. The edge server then generates a JWT-based token for authenticating the cloud upload. Unlike conventional IoT frameworks, LAXMI explicitly targets scalable surveillance workloads at city or enterprise level, where real-time analytics over distributed networks, parallel processing across heterogeneous edge nodes, and secure high-speed archival to the cloud require coordination similar to high-performance computing (HPC)/supercomputing environments. The design leverages parallel inference and distributed edge-cloud scheduling to meet ultra-low-latency and high-throughput demands. An 8-week duration performance test reveals that the edge-based motion detection subsystem achieved a 92.7% accuracy and F1-score of 96.2%, while the YOLOv8 classifier achieved 94.8% accuracy and F1-score of 94.4%. Further, the edge pipeline outperformed the cloud baseline across processing latency, energy expense, and bandwidth consumption constantly, thereby setting the effectiveness of the proposed LAXMI architecture in actual smart surveillance applications.