A Systematic Review of Facial Detection and Expression Recognition in Groups of People

Resumo

Expression recognition from facial inputs based on scene-based group of individuals is a very important approach, presenting potential applications for business, security, education, and healthcare areas. Recently, many approaches have been proposed to address this problem, having as a general solution the formulation of the problem as an extension from the single-face detection approach, while other approaches are specifically designed taking into account the multi-face scenario. This study presents a systematic literature review on the state-of-the-art group-level expression detection and recognition technique, based on facial images. The paper has, as its main goal, the identification of the most used strategies published over the past few years to interpret and recognize facial emotion expressions in groups of people. For this purpose, a total of 319 papers were collected from multiple well-established scientific databases (ACM Digital Library, IEEE Xplore, and Scopus) and, after applying the methodology for systematic literature and its inclusion and exclusion criterion, a total of 14 papers were analyzed from the literature, totaling 16 distinct methods. The obtained analysis demonstrates an extensive application for Convolutional Neural Networks (CNNs) compared to other categories of methods. Although predominantly used, the overall scores presented by CNNs were not the best suited across the evaluated methods. This literature review suggests that besides the good results achieved, there is still an open problem and a significant range for improvements, especially for the CNN counterpart.