Scholarlink

<p>This study aimed to assess the concurrent validity of DeepLabCut to estimate a player’s position and velocity in a pre-delimited course in an indoor environment. Ten young male basketball players (age: 16.5 ± 0.5 years; height: 181.7 ± 2.4&#xa0;cm; body mass: 75.7 ± 3.6 Kg) were submitted to static and dynamic tasks. The proposed tracking method with a single camera was compared to an optoelectronic system. Validity was analyzed by means of absolute and relative errors, Bland-Altman plot analyses, intraclass correlation coefficients, root mean square errors, and statistical parametric mapping for velocity time series. Overall, intraclass correlation coefficients values showed good to excellent (0.78–0.94) reliability between systems. Errors were higher as the proposed speed of the task increased. Statistical parametric mapping indicated significant differences (<i>p</i> &lt; 0.05) in velocity curves of both systems during moments of 90-degree change of direction. However, no significant differences were found regarding linear displacements or on change-of-direction instants at lower velocities. Collectively, our results showed that DeepLabCut can be considered a valid and cost-effective method for determining the distance traveled and the instantaneous velocity during on-court activities, making high-quality tracking accessible to organizations with limited resources.</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

Automatic tracking of indoor sports players using a video-based deep learning approach: a concurrent validity study

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Bruno Giovanini,
  • Felipe Arruda Moura

摘要

This study aimed to assess the concurrent validity of DeepLabCut to estimate a player’s position and velocity in a pre-delimited course in an indoor environment. Ten young male basketball players (age: 16.5 ± 0.5 years; height: 181.7 ± 2.4 cm; body mass: 75.7 ± 3.6 Kg) were submitted to static and dynamic tasks. The proposed tracking method with a single camera was compared to an optoelectronic system. Validity was analyzed by means of absolute and relative errors, Bland-Altman plot analyses, intraclass correlation coefficients, root mean square errors, and statistical parametric mapping for velocity time series. Overall, intraclass correlation coefficients values showed good to excellent (0.78–0.94) reliability between systems. Errors were higher as the proposed speed of the task increased. Statistical parametric mapping indicated significant differences (p < 0.05) in velocity curves of both systems during moments of 90-degree change of direction. However, no significant differences were found regarding linear displacements or on change-of-direction instants at lower velocities. Collectively, our results showed that DeepLabCut can be considered a valid and cost-effective method for determining the distance traveled and the instantaneous velocity during on-court activities, making high-quality tracking accessible to organizations with limited resources.