Scholarlink

<p>Head impacts are common in youth collision sports such as ice hockey and can contribute to brain injury. Impact biomechanics are influenced by characteristics such as impact magnitude, frequency, inter-impact interval, and cumulative exposure duration. Player velocity, particularly closing velocity, affects collision energy transfer and is therefore a key kinematic variable for characterizing head-impact events. To enable large-scale extraction of head-impact characteristics from standard youth game video, this study presents a proof-of-concept automated pipeline to estimate planar player velocity from a single panning side-view camera. A YOLOv8 player detector pre-trained on approximately 80,000 National Hockey League images was fine-tuned on a youth player-detection dataset (4200 training images with light/dark jersey labels), achieving mean average precision at an intersection-over-union threshold of 0.5 of 0.97. Players were tracked using an intersection-over-union-assisted StrongSORT configuration, reducing identity switches from 172 to 53 across 100 head-impact–centered clips (3&#xa0;s; 90 frames at 30 frames/s) and improving multiple object tracking accuracy from 89.0% to 94.5%. Using rink localization and homography (intersection-over-union = 0.96), trajectories were mapped into a top-down rink coordinate system, and planar velocity was computed from displacements over 10-frame intervals. Velocity accuracy was evaluated against a synchronized overhead drone reference across 16 skating trials, yielding a mean percentage error of 8.33% for the side-view configuration. This pipeline provides automated horizontal-plane player velocity estimates and enables future closing-velocity estimation for head-impact events when integrated with impact detection and player identification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development of an automated system to obtain the planar velocity of individual players from 2D video of youth ice hockey games

  • 期刊论文

  • 发表日期：

    2026年6月26日

查看全文
  • Parisa Dehghan,
  • Amirhossein Azadi,
  • Clara Karton,
  • Allison Clouthier,
  • T. Blaine Hoshizaki

摘要

Head impacts are common in youth collision sports such as ice hockey and can contribute to brain injury. Impact biomechanics are influenced by characteristics such as impact magnitude, frequency, inter-impact interval, and cumulative exposure duration. Player velocity, particularly closing velocity, affects collision energy transfer and is therefore a key kinematic variable for characterizing head-impact events. To enable large-scale extraction of head-impact characteristics from standard youth game video, this study presents a proof-of-concept automated pipeline to estimate planar player velocity from a single panning side-view camera. A YOLOv8 player detector pre-trained on approximately 80,000 National Hockey League images was fine-tuned on a youth player-detection dataset (4200 training images with light/dark jersey labels), achieving mean average precision at an intersection-over-union threshold of 0.5 of 0.97. Players were tracked using an intersection-over-union-assisted StrongSORT configuration, reducing identity switches from 172 to 53 across 100 head-impact–centered clips (3 s; 90 frames at 30 frames/s) and improving multiple object tracking accuracy from 89.0% to 94.5%. Using rink localization and homography (intersection-over-union = 0.96), trajectories were mapped into a top-down rink coordinate system, and planar velocity was computed from displacements over 10-frame intervals. Velocity accuracy was evaluated against a synchronized overhead drone reference across 16 skating trials, yielding a mean percentage error of 8.33% for the side-view configuration. This pipeline provides automated horizontal-plane player velocity estimates and enables future closing-velocity estimation for head-impact events when integrated with impact detection and player identification.