Automated Social Distance Recognition and Classification using Seagull Optimization Algorithm with Multilayer Perceptron

Abstract

Social distancing (SD) is a set of non-pharmaceutical disease control actions. Its main intention is to stop or decrease the transmission of a spreadable disease. The goal is to decrease the prospect of interaction among peoples who carries an infection and others who are not diseased to diminish disease spread. This can also contain actions like keeping a distance of at least 6 feet from other people, avoiding crowded locations as well as reducing physical communication. The training of SD was suited mainly at the time of COVID-19 virus, where it was suggested as a fundamental action to diminish the spread of disease. The exact rules for SD may differ reliant on the nature of the virus and references from healthcare consultants. SD recognition's main objective is to leverage machine learning (ML) and artificial intelligence (AI) based technical solutions to identify and aware persons when they are not obeying the rules and regulations of SD action. This manuscript offers the design of an Automated Social Distance Recognition and Classification using the Seagull Optimization Algorithm with Multilayer Perceptron (SDRC-SOAMLP) technique. The purpose of the SDRC-SOAMLP technique is to categorize high-risk and low-risk SD using parameter-tuned ML models. To accomplish this, the SDRC-SOAMLP technique initially undergoes two stages of preprocessing: Wiener Filter (WF) based noise removal and Dynamic Histogram Equalization (DHE) based contrast improvement. For the detection and estimation of distance between the pedestrians, the segmentation process involving running average-based adaptive background subtraction and Euclidean distance measurement is used. Followed by, the MLP model can be exploited for the detection and classification of the SD. To enhance the performance of MLP’s detection rate, the SOA can be applied for the optimal selection of the parameters related to it. The experimental validation of the SDRC-SOAMLP technique is verified on the SD dataset and the results are examined in distinct measures. The comprehensive comparison study stated the superior performance of the SDRC-SOAMLP model when compared to recent models.