基于改进YOLOv5的目标智能检测方法在罚球姿势中的应用

Application of target detection method based on improved YOLOv5 in the prediction of different free throw positions

摘要:

针对智能体育视域下的罚球智能命中预测问题,研究提出了基于YOLOv5网络改进损失函数和引进注意力机制的目标检测网络完成对比赛视频中罚球球员的准确识别与有效提取。然后设计了一种基于轨迹优化的识别方法进行罚球球员的姿态检测,并利用支持向量机(Support vector machine,SVM)分类器进行篮球罚球命中预测。实验结果显示研究提出的目标检测算法在训练15轮时就达到了85%的平均精度,而命中预测算法的进球预测准确性也增加了5%,优于实验中的其他算法。实验结果表明改进损失函数并引入注意力机制的视频罚球队员检测跟踪方法能够有效检测罚球队员所在区域,并准确将罚球队员的特写视频提取出来。篮球教练可以根据经验,结合姿态估计和罚球命中预测数据,有效评价动作表现并给出指导建议,既能实现运动姿态矫正还能帮助评估运动员水平,进而建立智能化运动员数据档案。

Abstract:

To address the problem of intelligent hit prediction of free throws in the context of smart sports, the study proposes a target detection network based on the improved loss function of YOLOv5 network and the introduction of an attention mechanism to complete the accurate identification and effective extraction of free throw players in game videos. Then a recognition method based on trajectory optimization is designed for the pose detection of free throw players, and a support vector machine (SVM) classifier is used for basketball free throw hit prediction. The experimental results show that the target detection algorithm proposed in the study achieved an average accuracy of 85% after 15 rounds of training, while the accuracy of the hit prediction algorithm also increased by 5%, which is superior to other algorithms in the experiment. The experimental results show that the improved Loss function and the introduction of attention mechanism can effectively detect the area where the free throws are located, and accurately extract the close-up video of the free throws. Basketball coaches can combine posture estimation and free throw hit prediction data based on experience to effectively evaluate movement performance and provide guidance suggestions. This can not only achieve posture correction but also help evaluate athlete level, thereby establishing intelligent athlete data archives.