Estimating Player Positions from Padel High-Angle Videos: Accuracy Comparison of Recent Computer Vision Methods.

Area of Science:

• Sports Science
• Computer Vision
• Machine Learning

Background:

• Player position estimation is crucial for sports performance analysis.
• Padel videos typically use a standard high-angle camera view, simplifying player tracking.
• Existing computer vision methods need evaluation for padel-specific challenges.

Purpose of the Study:

• To evaluate and compare state-of-the-art computer vision methods for single-angle player position estimation in padel.
• To assess the accuracy of deep convolutional neural network (CNN)-based techniques.
• To demonstrate the feasibility of automated player tracking in padel.

Main Methods:

• Utilized a large dataset of amateur and professional padel videos.
• Applied and compared object detection, image segmentation, and pose estimation techniques.
• All methods were based on deep convolutional neural networks (CNNs).
• Evaluated performance using manually-annotated video frames, reporting accuracy and average precision.

Main Results:

• Top-down pose estimation methods achieved the highest accuracy.
• Achieved a player detection rate of 99.8%.
• Reported root-mean-square error (RMSE) below 5 cm and 12 cm for horizontal and vertical coordinates, respectively.

Conclusions:

• Deep convolutional neural network (CNN)-based pose estimation is highly suitable for player position estimation in single-angle padel videos.
• This technology enables performance and team analysis from existing video footage.
• Offers an affordable method for player positional data acquisition in amateur padel.