| Autor: |
Alnajjar, Ibrahim Ali, Almazaydeh, Laiali, Odeh, Ali Abu, Salameh, Anas A., Alqarni, Khalid, Ban Atta, Anas Ahmad |
| Předmět: |
|
| Zdroj: |
International Journal of Intelligent Engineering & Systems; 2024, Vol. 17 Issue 4, p649-665, 17p |
| Abstrakt: |
Anomaly detection from a video surveillance camera is a time-critical application that combines the capabilities of computer vision-based object detection algorithms to monitor and analyze various anomalous activities in Industry 4.0 scenarios. An intelligent video surveillance system for automated monitoring and analysis of video streams without human supervision is paramount in industrial scenarios. Nevertheless, the detection of anomalous objects is often hindered by the scarcity of data and privacy restrictions inherent in the centralized storage and processing of surveillance videos. To overcome these shortcomings, Federated Learning (FL) has emerged as a promising solution for the privacy-preserved processing of decentralized data. Despite significant advancements in computer vision, accurately identifying surveillance anomalies through object detection within resource-constrained edge networks remains a formidable challenge for FL-assisted anomaly detection. This difficulty arises from the limited computational capabilities and constrained resources inherent to edge devices, which impedes the performance and accuracy of anomaly detection algorithms relying on the object recognition method. Thus, this work proposes a hierarchical FL-assisted surveillance anomaly detection by integrating the YOLOv8n and Flownet models for motionaware, accurate anomalous object detection. To design time-efficient anomaly detection for time-critical applications, the proposed approach applies the hierarchical FL that comprises multiple edge aggregators instead of cloud aggregators. The primary objective of adopting the hierarchical FL is to alleviate the communication costs associated with object detection tasks. By distributing the aggregation process across multiple edge nodes, the proposed approach enhances the efficiency of anomaly detection while minimizing latency, thereby ensuring timely responses. Finally, the FL-assisted detection system accurately identifies anomalous human activities in the manufacturing industry through the hierarchical aggregation associated with the local model of YOLOv8n and Flownet-based object detection in the edge network. Thus, the experimental results prove its anomaly detection ability in the surveillance vides by yielding 88.95% accuracy and 0.85 as the average Anomaly score while testing on the Avenue dataset. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
