Respiratory rate monitoring by maximum likelihood video processing

Abstract

A novel video processing-based method for remote estimation of the respiratory rate (RR) is proposed. Relying on the fact that breathing involves quasi-periodic movements, this technique employs a generalized model of pixel-wise periodicity and applies a maximum likelihood (ML) criterion. The system first selects suitable regions of interest (ROI) mainly affected by respiratory movements. The obtained ROI are jointly analyzed for the estimation of the fundamental frequency, which is strictly related to the RR of the patient. A large motion detection algorithm is also applied, in order to exclude, from RR estimation, ROI possibly affected by unrelated large movements. The RRs estimated by the proposed system are compared with those extracted by a pneumograph and a previously proposed video processing algorithm. The results, albeit preliminary, show a good agreement with the pneumograph and a clear improvement over the previously proposed algorithm.