Deep Learning Algorithm-Based Target Detection and Fine Localization of Technical Features in Basketball.

Area of Science:

• Computer Vision
• Sports Analytics
• Machine Learning

Background:

• Accurate player detection and action recognition are crucial for sports analytics.
• Existing methods struggle with complex backgrounds and precise localization of technical features in basketball.

Purpose of the Study:

• To develop an advanced player segmentation and action recognition system for basketball.
• To improve the accuracy and efficiency of detecting and localizing key basketball technical features.

Main Methods:

• A super-pixel-based Fully Convolutional Network-Convolutional Neural Network (FCN-CNN) algorithm was developed for player segmentation using Single Shot Detector (SSD).
• High-resolution Convolutional Neural Networks (CNNs) were employed for image preprocessing and feature extraction.
• The system was trained to identify specific basketball actions: rebounds, shots, and passes.

Main Results:

• The proposed FCN-CNN algorithm effectively filters complex backgrounds, aiding subsequent pose estimation and fine localization.
• The system achieved a high accuracy rate of up to 95.6% in identifying typical basketball sports actions from video streams.
• Comparative analysis demonstrated the superiority of the proposed target detection system over three classical classification algorithms.

Conclusions:

• The developed system provides effective target detection and fine localization of basketball sports technical features.
• The integration of SSD, FCN-CNN, and high-resolution CNNs offers a robust solution for basketball action analysis.
• This research contributes to advancements in automated sports performance analysis through computer vision.