An accurate estimation of preterm infants’ limb pose from depth images using deep neural networks with densely connected atrous spatial convolutions

作者：

Highlights：

• Monitoring preterms’ movement during and after the hospitalization is crucial.

• We present a deep-learning-pipeline to estimate preterms’ limb pose from depth frames.

• The pipeline, trained on 27000 frames, is able at generalizing on challenging frames.

• The pipeline underwent architectural variations to make it computationally efficient.

摘要

•Monitoring preterms’ movement during and after the hospitalization is crucial.•We present a deep-learning-pipeline to estimate preterms’ limb pose from depth frames.•The pipeline, trained on 27000 frames, is able at generalizing on challenging frames.•The pipeline underwent architectural variations to make it computationally efficient.

论文关键词：Pose estimation,Preterm birth,Atrous convolutions,Convolutional neural networks,Depth images,Neonatology

论文评审过程：Received 17 May 2021, Revised 17 March 2022, Accepted 28 April 2022, Available online 13 May 2022, Version of Record 9 June 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117458