Wearing detection and action recognition of protective gear based on pose estimation

Summary: Guard wearing detection and motion recognition are an important part of the smart safety monitoring system.Aiming at the problems of low accuracy and low real-time performance of traditional parallel recognition methods, a parallel recognition scheme based on pose estimation is proposed.Using a data interaction method based on distance and Hungarian algorithm, the human torso and protective gear are matched, and the protective gear wearing detection and motion recognition are quickly and parallelly realized.Parallel recognition scheme utilizes depthwise separable convolutions to OpenposeThe parameters of the model are compressed so thatOpenposeThe model is lightweight, which improves the real-time performance of the model's attitude estimation, and proposes a quadrilateral detection method combined with attitude information, which solves the problem of misjudgment without protective gear (hand-held helmet).In the real-time detection experiment, the false positive rate of the hand-held helmet of the parallel recognition scheme decreased29.3%, the overall accuracy of the system reaches 93.5