Black Widow Optimization Algorithm with Deep Learning Driven Accurate Social Distance Detection

Abstract

Social distance (SD) detection is a critical part of public health surveillance and depends on computer vision tools to evaluate and safeguard adherence to secure personal distance. Leveraging methods like object detection and deep learning (DL) models can precisely detect people in crowded settings and size the physical gap among them. By examining real-time video streams, these systems deliver direct responses to SD compliance, providing an active tool for public healthcare experts to monitor and diminish the transmission of the virus. This advanced use of computer vision improves safety procedures as well as demonstrates the incorporation of technology in the protection of community well-being. In this manuscript, we propose a Black Widow Optimization Algorithm with Deep Learning Driven Accurate Social Distance Detection (BWODL-SDD) technique. The BWODL-SDD technique exploits a hyperparameter tuned DL model for the identification and classification of the SD. For enhancing the input image quality, the noise gets removed by the median filtering (MF) approach and adaptive histogram equalization (AHE) based contrast enhancement is made. Besides, UNet model can be employed for segmentation process which aims to properly detect the presence of pedestrians and the hyperparameter tuning of the UNet model takes place with Adam Optimizer. Followed by, the distance among the detected pedestrians can be computed by the use of Manhattan distance based measurement. At last, the BWO algorithm with deep recurrent neural networks (DRNNs) can be applied for the recognition and classification of SD. The design of the BWO algorithm helps in the optimal hyperparameter assortment of the DRNN model. The experimentation study of the BWODL-SDD model is carried out on a benchmark SD dataset. The simulation outcomes highlighted the supremacy of the BWODL-SDD technique in the SD detection procedure.