Scholarlink

<p>Current hostels are having out-dated security mechanisms which pose serious threat to student’s protection and validation. To mitigate this, the given paper introduces a three-layered hybrid framework comprising pre-processing layer, adaptive recognition-feature analysis layer and decision layer. The first layer performs noise reduction, normalization, and contextual filtering on heterogeneous data streams coming from devices like CCTV, biometric logs, and IoT sensors. The second layer switches between MTCNN (Multi-task Cascaded Convolutional Networks) and GD-HASLR approach (Gradient Direction-based Hierarchical Adaptive Sparse and Low-Rank). Under routine conditions, MTCNN is used for fast results and GD-HASLR is selectively employed in cases like partially occluded faces, dim lighting or challenging input. The last layer functions as an anomaly detector and prediction module of our framework by maintaining detailed access logs. Decision boundaries in this layer are tuned by BPO (Brood parasitism optimization) during high-risk periods to dynamically optimize minimizes false acceptance and rejection, ensuring reliable access management. Eventually, the performance is evaluated based on context-relevant metrics like Recognition Accuracy (RA %), Latency (ms), and Energy Consumption (J). The results show superior performance of the proposed framework as compared to the existing studies.</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

A hybrid framework to improve hostel security using MTCNN, GDHASLR and BPO

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Yousef Methkal Abd Algani

摘要

Current hostels are having out-dated security mechanisms which pose serious threat to student’s protection and validation. To mitigate this, the given paper introduces a three-layered hybrid framework comprising pre-processing layer, adaptive recognition-feature analysis layer and decision layer. The first layer performs noise reduction, normalization, and contextual filtering on heterogeneous data streams coming from devices like CCTV, biometric logs, and IoT sensors. The second layer switches between MTCNN (Multi-task Cascaded Convolutional Networks) and GD-HASLR approach (Gradient Direction-based Hierarchical Adaptive Sparse and Low-Rank). Under routine conditions, MTCNN is used for fast results and GD-HASLR is selectively employed in cases like partially occluded faces, dim lighting or challenging input. The last layer functions as an anomaly detector and prediction module of our framework by maintaining detailed access logs. Decision boundaries in this layer are tuned by BPO (Brood parasitism optimization) during high-risk periods to dynamically optimize minimizes false acceptance and rejection, ensuring reliable access management. Eventually, the performance is evaluated based on context-relevant metrics like Recognition Accuracy (RA %), Latency (ms), and Energy Consumption (J). The results show superior performance of the proposed framework as compared to the existing studies.