Shaping the Video Conferences of Tomorrow With AI

Abstract

Before the COVID-19 pandemic, video was already one of the main media used on the internet. During the pandemic, video conferencing services became even more important, coming to be one of the main instruments to enable most social and professional human activities. Given the social distancing policies, people are spending more time using these online services for working, learning, and also for leisure activities. Videoconferencing software became the standard communication for home-office and remote learning. Nevertheless, there are still a lot of issues to be addressed on these platforms, and many different aspects to be reexamined or investigated, such as ethical and user-experience issues, just to name a few. We argue that many of the current state-of-the-art techniques of Artificial Intelligence (AI) may help on enhancing video collabo- ration services, particularly the methods based on Deep Learning such as face and sentiment analyses, and video classification. In this paper, we present a future vision about how AI techniques may contribute to this upcoming videoconferencing-age.

Status, Challenges and Trends in Videoconferencing Platforms

“…A paper published by Mendes et al [18] presents a list of eleven situations in which AI could be used to address the needs that are commonly encountered in the usage of videoconferencing systems such as: a) verifying user identity, b) counting and possibly anonymizing participants, c) detecting user attention/engagement, or d) summarizing the meeting. Some of these use cases have already been analysed by providers of videoconferencing systems and partially or fully covered by specific features.…”

A Cluster-Based Method for Action Segmentation Using Spatio-Temporal and Positional Encoded Embeddings

“…To effectively extract information from this massive amount of data, better video understanding methods are needed. In particular, a crucial task of video understanding is the recognition and localisation in time of different actions or events that are present along the scenes [26,32] in the so called action recognition task, which has attracted much of the researchers attention nowadays. Initial efforts focused on the classification of trimmed videos with a single action [5,11,36,41].…”

A Cluster-Based Method for Action Segmentation Using Spatio-Temporal and Positional Encoded Embeddings

“…Video understanding methods aims to extract high-level semantic information from videos such as activities, actions, objects and scenes. In particular, a crucial task of video understanding is the segmentation in time of di erent actions that are present along a video (1,2). In other words, this task aims to provide answer for the following questions: (i) What are the actions and (ii) when do the actions happen?…”