Improved attention and node binding-based icu hand hygiene compliance monitoring method

Abstract

The application relates to an ICU hand hygiene compliance monitoring method based on improved attention and node binding, and relates to the technical fields of medical informatization, artificial intelligence and edge computing; in view of the problems of high cross-lens tracking computing power, easy invasion of privacy and target loss caused by serious double-hand shielding when hands are washed and rubbed, the first camera in a ward and the second camera of a corridor washbasin are configured as a binding monitoring node pair, macro statistical period aggregation calculation compliance rate is adopted, and individual tracking is avoided; the first camera is vertically overhead, and multi-target horizontal shielding is eliminated; a local-global sparse attention model is adopted to reduce edge computing amount, left and right hand query vectors are forced to separate in a hidden space through feature orthogonal regularization and space-time comparison loss, and double-hand independent trajectories are maintained when overlapping; while realizing physical isolation of privacy, the application greatly reduces edge end computing power consumption, and significantly improves the accuracy and system robustness of action recognition in a complex environment.