1. Introduction

COVID-19 is a rapidly transmitted disease that can be transmitted through body droplets and can survive up to two days on a touched surface. The disease outbreak announcement also illustrated the growing concerns regarding COVID‐19's unprecedented scale of transmission and intensity. This is distinguished by its span as a global health emergency that has expanded all around the globe.

Legislative bodies of numerous countries are imposing restrictions, limits on travel, and social distancing, and are attempting to grow public knowledge of sanitation. Nevertheless, the infection tends to propagate quite rapidly. In Saudi Arabia, COVID-19 has spread through various territories of the world. The Mecca area has seen the highest incidence of COVID-19 to date. Social distancing is assumed to be an efficient measure of the spread of the pandemic virus in the present scenario. The risk of the transmission of viruses can be reduced by preventing physical interaction between individuals. The development of an automated simulation model is therefore important to implementing a social distancing scenario to prevent people from becoming infected.

A literature review revealed several studies tracking social distancing in public spaces. Harvey et al. ^[1] and Robakowska et al. ^[6] used drone technology and cameras to track movements in a crowd. They emphasized that the use of drone technology may help maintain public order and also help to detect suspicious activity in a group. Currently, the system provides a comprehensive overview of a surveillance system for patrolling and inspection. Ngu-yen et al. ^[3] provide a detailed social distance context, such as core principles, measurement systems, and prototypes, and suggest alternative socially relevant distance situations. They also describe supporting particularly efficient wire-less systems that can be broadly accepted in practice to maintain distance and promote and implement social distance in particular. They presented a survey on many emerging technologies like Bluetooth and WIFI that can work as a key aspect in a social distancing scenario.

Ahmed et al. ^[2] planned a prototype to offer a deep-learning framework for social distance monitoring. To recognize individuals in video sequences, this approach implements the YOLOv3 object detection framework. To improve the model's accuracy, the transfer learning technique is applied. They measure breaches of the social distance among persons and use a physical gap estimation. The emphasis on many technologies and using deep learning algorithms to track human movement plays a key role in maintaining social distancing. Punn et al. ^[4] and Pouw et al. ^[5] proposed a system using the YOLOv3 model to detect humans and track them using bounding boxes. They utilized information obtained from overhead commercial pedestrian monitoring sensors. They introduce the idea of the ?Corona Event? to track two people who get closer than a threshold distance. However, much of their work relies on live surveillance, and cameras to detect humans ^[7,8,9,10].

The recent pandemic of COVID-19 has prompted several scientists to find other responses. Rao et al. ^[11] suggested a framework intended for the identification of COVID-19 victims via smartphone. Yan et al. ^[12] established a forecasting framework to appraise high-risk patients at an early stage before they pass from moderate to severely sick. Several research articles on forecasting the coronavirus pandemic have been presented in recent times ^[13]. The researchers concentrated on developing a new framework focused on artificial intelligence technologies that combine machine learning algorithms and various data modalities ^[14]. The Adaptive Neuro-Fuzzy Inference System (ANFIS) is suggested by Al-Qaness and colleagues ^[15]. A regression framework was built for the estimation of rapid growth in COVID-19 dependent on the overall percentage of reported patients outdoors in China ^[16]. Many researchers produced predictions by ten common machine learning and predictive environmental area frameworks for the observation of large-scale weather variations ^[17].

Also, many mobile applications use Bluetooth technology to notify people if they have had contact with others confirmed to be infected with the coronavirus. In contrast, previous approaches only used cameras for tracking humans and for detection purposes. Nevertheless, the drawback of the existing applications is that they are not embedded with real-time detection modules to detect social distancing between people. We decided to investigate this area for the following reasons:

Following are the points why we researched this idea.

• The social distance control system in this proposal uses smartphone GPS and Bluetooth technologies that overcome the problems of tracking people's movement with cameras.

• A Bluetooth or GPS technology, therefore, is used to conduct the evaluation process in the proposed framework. With personalized data collection, the trained machine learning model is integrated with the existing applications. The applications run the machine learning algorithm and automatically identify whether or not social distance is retained in public.

• To verify if social distancing is preserved by individuals in a public zone, it also measures whether any person breaches the social distancing standard. If not, the application will send the authority a warning signal and even give the public an alert.

• The identification system uses the location information observed to classify samples. We can use a physical distance estimation through GPS and Bluetooth and set a standard to approximate social distance breaches between individuals. To determine whether the distance value violates the smallest social distance threshold, a breach threshold is defined. In particular, a monitoring algorithm would be used to identify the individuals who breach the social distancing standard. To evaluate the model's output, results are analyzed through several statistical measures. Results indicate that the proposed paradigm accurately identifies persons who approach too closely and break/violate social distancing rules. The learning method also improves the system's overall efficacy.

• The Deep Extreme Learning Machine (DELM) framework may utilize clean anomalies in the data obtained from various resources, and train existing approaches to build on the optimal framework using several training sets. In this article, a DELM-based context is proposed to achieve maximum accuracy. In the context of COVID-19, we intended to construct a DELM system that could establish the precise system for maintaining social distancing.

The system is designed to detect two states: whether individuals are not too close to each other and whether they remain with each other. The system will work based on distance tracking and the machine-learning framework. Tracking location sharing among people removes the need to track distances between people necessary to maintaining social distancing. The suggested method thus supports the community by saving time and helps mitigate the spread of the coronavirus.

It is necessary to identify COVID-19 patients early to prevent transmission of the disease to other persons. In this study, we propose using a DELM-based approach together with a location-based dataset obtained from GPS and Bluetooth for automatic detection of social distancing. This model achieved an accuracy of more than 90%. This research provides insight into how deep extreme learning machines can be implemented to manage the spread of COVID-19 in the initial phase of an outbreak. COVID‐19 has now been a challenge to the health infrastructure around the globe and many people have collapsed due to this deadly pandemic. Due to an increasing number of emergency patients, healthcare professional capacity is scarce, and a computer-controlled autonomous social distancing system can save lives by early assessment in cities and can save lives by appropriate approach.

In this study, using an AI approach to building a social distancing control framework in applications is presented. To decide if an individual's conduct violates the least social distance range, a violation threshold must be employed. Additionally, a central detection system is used to map communities. Experimental findings found that the system successfully detects individuals who become too close to others and infringe on social distance practices. The technique of transfer learning further advances the overall reliability and precision of the detection system

The suggested system identifies persons who did not sustain social distance. Therefore, this system would function effectively and automatically to assist in simple social distance inspection. The system can retrieve distance measures from the GPS position details and save them in the database of the concerned authorities.

The proposed framework effectively and efficiently trains itself to detect insufficient social distancing. We propose that the model will remarkably enhance the speed and precision of social distance tracking. It may be beneficial in a disease outbreak where the epidemic risk and the requirement for prevention steps do not suit the available resources. More appropriate, refined, and richer datasets will further mature the learning rate of the system.

2. Proposed Methodology

A proposed social distance monitoring system implemented in this work uses Bluetooth and GPS technology. The system model is seen in Fig. 1. The mobile user's dataset generated is sent to a satellite separated into sets of different users to track the distance between both users, and a machine learning-based tracking approach is used to extract the highest level of prediction of current location data. A predetermined threshold is used for calculating the centroid distance to verify whether or not the distance between any two users is lower than the standard distance. If two entities are near one another and the gap magnitude breaches the standard of the least social distance then the alarm is sent to concerned authorities and also both users. A centroid detector technique is used to observe individuals who breach the social distance threshold and trace their travel. The device presents information about social distance violations together with the detected individuals as the result.

Fig. 1. Proposed social distance monitoring system model.

Numerous hidden neurons, in addition to activation functions of several sorts, are utilized to attain the best framework for social distancing detection. The suggested architecture has three layers: the information-gathering phase, the pre-processing phase, and the application phase. Both the information-gathering layer and the pre-processing layer are included in the application layer in which one is for prediction and the other is for evaluation. For exploratory studies, actual information was acquired from GPS and Bluetooth. The data were then delivered to the data collection layer as input. Detection techniques are implemented at the data pre-processing stage to prevent any anomalies from being used in the information. In the application phase, the DELM framework is implemented to detect social distancing standards. The DELM can be organized in a variety of ways to estimate health challenges, predictions of energy consumption, transport, and traffic management, etc. ^{[18,19,20,21]}.

The concept of an ELM is defined by Huang et al. ^[22]. Since we use the standard Feed-Forward Neural Network (FFNN) method, data is flowing in only one direction across the network (predictions, as in ``from the inputs to the outputs'') through each stage in the training phase. However, we have often used the Recurrent Neural Network (RNN) approach throughout this predictive system through the training procedure, where input streams move from a specific node back through the network, and any time this flow reaches the node, the weights are changed to achieve higher precision and quality without the error rate ballooning. The constant of all the weights in the validation stage is imported in addition the actual data are predictable. The DELM context comprises of input, several hidden, in addition to one output layer.

The principal goals of this approach are:

• The DELM algorithm will be implemented to construct the most effective strategy for social distancing.

• To use an adaptive method to detect and track social distance among individuals and to identify those who are not maintaining social ties in consideration of their particular conditions.

• To incorporate and enhance Bluetooth and GPS technologies in pre-existing applications to maintain social separation criteria. In addition, a feature learning strategy is employed to improve the model's effectiveness. The newly trained layer is then deployed to the pre-trained architecture to improve the system's performance.

• To use a detecting mechanism to keep track of individuals who violate the social distance threshold.

• To evaluate the usefulness of a pre-trained system using a location dataset. The detection system's deployment is assessed with and without learning algorithms. The performance of the system is also assessed using other AI models.

• To slow the spread of the coronavirus epidemic and prevent the spread of infection.

• To evaluate the suggested approach using GPS and Bluetooth datasets for the purpose of performance estimation.

• To validate the suggested method by testing the datasets with various machine learning techniques.

Fig. 2. DELM architecture ^[18].

The primary objective of this study is to develop a smart algorithm to identify and track social distancing among people and identify those individuals who are not maintaining social distancing by improving and implementing AI in existing applications. The suggested framework relies primarily on the use of Bluetooth and GPS technology to track the distance between humans in real-time.

Let's imagine there are multiple different hidden layer feedforward neural networks, each with $n$ neurons in the hidden layer, in addition to a training data set consisting of $N$ records $(è_i , f_i)$ in which $e_i \in S_d$ and $f_i \in S_c$. The output of such a Feedforward Neural Network with Multiple Hidden Layers may look like;

Here ${{\eth }}_j$ and ${ ç}_j $ are factors that affect learning, $ß _{j}$ Output Weighted Nodes $j $and $ Î :$ $S \to S$ is the activation function.

It has been demonstrated by a perfect settlement of several hidden layers in a feedforward neural network with zero errors that with discrete intervals ${{\eth }}_j$ and ${{ ç}}_j $ there occur ${ß}_{\mathrm{j}}$ like that;

which can be denoted as

where

and

Results assessment weights can be calculated using the method below when the amount of feedback received is greater than the number of neurons in the hidden layer.

Matrix $Ú$ is inverse of ${Ú}^{{\nmid }}$. DELM is consequently a computationally economical scheme of study.

There are a few different parts to the backpropagation system, including weight setup, forward propagation, error propagation, and update to the ability to tell things apart. An activation function like ${g} ({x}) =$ sigmoid is existing in each neuron's hidden layer. This benefit projecting the sigmoid input function in addition to the DELM hidden layer;

Backpropagation error is defined in Eq. (7) as the difference between the actual and desired result squared, divided by 2. To correct for this frequent miscalculation, a rebalancing of the weights is required. Indicated by Eq. (8) are the ranges of allowed weight changes in the output layer.

Where $i= 1$, $2$, $3$, $\dots$, $10$ and $j =$ output layer.

Eq. (9) demonstrates the weights update also how the biases happen among the hidden layer and inputs.

3. Simulation & Results

4. Conclusions

A framework for social distancing has been established to increase the accuracy of safe and unsafe distance calculation predictions. Various methodological approaches were applied to evaluate the viability of this specific proposition. The suggested DELM approach is notable for its effectiveness. During validation, the suggested application had an accuracy between 95.32 and 91.2%. An additional advantage of the DELM approach is that fundamental algorithms are inexpensive and quick. We are confident that these initial results and intend to expand this work by analyzing other datasets in the future. Future research will seek to accurately characterize and measure the parameters. Future research will seek to accurately characterize and measure the parameters of the DELM approach to monitoring social distancing. Notably, the algorithm will be re-trained more often to improve its performance in a range of conditions.