Bibliometric Analysis of Publications on Action Recognition, Convolutional Neural Network, Video Surveillance During 2011-2021

Action recognition based on convolutional neural networks (AR-CNN) has been developingrapidly in recent years. It is of great significance to conduct a deep analysis to understand the recentdevelopment of AR-CNN. However, a limited number of studies examining the research status of this fieldcould be found. Therefore, this study aims to quantitatively assess the publications related to the SciValtopic "Action Recognition; Convolutional Neural Network; Video Surveillance (T.561)" in computer visionresearch. This study focused on six aspects: literature distribution characteristics analysis, the developmenttrend, citation analysis, collaborative analysis, keyword analysis, and thematic evolution, using VOSviewerand Bibliometrix. The relevant publications were retrieved from Scopus in the period 2012–2021. A total of6633 publications were identified by 9088 different authors; 62% were conference papers, and 35% wereresearch articles. China and the USA contributed 39.7% and 17.9% of the total publications, respectively. Theauthors’ productivity demonstrated variability in alignment with Price’s Law, yet exhibited consistency whenevaluated under the framework of Lotka’s Law. Ling Shao was the most productive author, with 48 papers(0.7%). Chinese Academy of Sciences was the most productive affiliation, with 259 papers (3.9%). The firstBradford site consisted of Computer Science Lecture Notes with 617 publications. A moderately significantcorrelation was revealed between the country’s publications and GDP per capita. The overall results show thatthe number of AR-CNN-related documents has increased significantly in recent years, with rapid growth from2016. Although publications on AR-CNN were published mainly in European journals, China led the scientific production.