Real Time Detection and Tracking in Multi Speakers Video Conferencing

Abstract

Currently, the videoconferencing market is growing worldwide with annual growth (CAGR) up to 10%. Several companies appreciated this technique during the Coronavirus lockdown as it allowed them to maintain continuous activity and frequent remote meetings. Despite the diversity of videoconference platforms, several are still needed especially if the conference participant moves out of the video capture window or if there is more than one person in the window. In this paper, a new videoconferencing system capable of detecting, choosing and tracking one participant is proposed. The framework suggested uses deep learning algorithms and offers the detection and tracking of a chosen person in a video stream. Person detection uses the Convolutional Neural Network model trained on a selected dataset. The tracking uses the SiamFC algorithm. In the model test phase, our system achieved an accuracy of 98%.