APPLICATION OF RECURRENT NEURAL NETWORK ALGORITHM WITH LSTM ARCHITECTURE TO DETECT VIOLATIONS ON SOCIAL DISTANCING

• Main Author: Tri Farhan, Afif
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/63737
• Institution: Institut Teknologi Bandung

The COVID-19 pandemic which began at the end of 2019 is still plaguing the world even now. The pandemic which caused by the Novel Coronavirus-19 has had an impact on various sectors around the world. According to WHO, this virus can spread through droplets that forms when an infected person coughs or sneezes, making the virus highly contagious. There are several ways to prevent transmission of the virus, one of which is to do social distancing. Social distancing means keeping a safe distance between yourself and other people. The minimum advised distance is about 1.8 to 2 meter. Even though social distancing sounds easy, there are still a lot of people who violate the practice of social distancing, especially when there is no specific supervision for the said practice. In this final project, a social distancing violation detection system is made using a neural network model to do indoor localization based on RSSI value from a BLE device. The results of the localization form the model are then visualized on a web application. The output of this final project is to made a machine learning model using the RNN algorithm and LSTM architecture to do indoor localization. The model accepts input in the form of time-series data containing the RSSI value from the last 10 steps. The model output is the absolute location in a 2 dimensional plane that will be used by the web application. The model need to be able to be used by the web application without any constraints. The best model from this final project has a mean squared error of 0.66 when used to do indoor localization using ground truth data.