Player Identification in Hockey Broadcast Videos

作者：

Highlights：

• Recurrent convolutional neural network proposed for hockey player identification.

• Variable length image sequences of player bounding boxes (tracklets) are classified.

• Spatial–temporal information across video frames improves jersey number predictions.

• A secondary classifier added as a late score-level fusion method increases accuracy.

摘要

•Recurrent convolutional neural network proposed for hockey player identification.•Variable length image sequences of player bounding boxes (tracklets) are classified.•Spatial–temporal information across video frames improves jersey number predictions.•A secondary classifier added as a late score-level fusion method increases accuracy.

论文关键词：Computer vision,Recurrent models,Convolutional neural networks,Sports player identification,Jersey numbers,Broadcast hockey videos

论文评审过程：Received 20 December 2019, Revised 13 August 2020, Accepted 14 August 2020, Available online 20 August 2020, Version of Record 29 August 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.113891