Keyframes selection from multiscene videos for stress detection

Junrui Tian, Zexi Lin, Yi Dai, Yang Ding, Jinlei Liu, Lei Cao, Ling Feng

Abstract: Highlights•Propose keyframes selection for stress detection in multiscene videos.•Develop a fine-grained two-stage framework to optimize selection process.•Design peer-attended collaboration mechanism, balance efficiency and accuracy by DRL.•Establish video quality metrics after keyframes selection.•Create confidence evaluation methods for stress detection decisions.•Build a multiscene video dataset with stress annotations.