摘要
随着竞技运动在全球范围内的蓬勃发展,对运动动作的识别精确度要求也逐渐提高。研究针对竞技运动动作识别领域,结合改进You Only Look Once version 7(YOLOv7)和注意力机制对健美操动作识别模型进行设计。过程中将两种跨阶段局部网络结构相结合,同时优化损失函数,得到改进的YOLOv7。再结合SimAM和时空注意力机制,完成识别模型的搭建。实验结果显示,改进后模型的健美操上肢动作识别准确率达到了90%,相比改进前提升了28.4%。模型的身体姿态变化识别率从64.5%升高到90%,召回率从60%升高到91.3%。结果表明,研究设计的竞技运动动作识别模型能够更好地理解和模拟竞技运动中动作的时空特性,对竞技运动相关研究以及实际应用场景中的动作识别需求具有推动作用。
With the flourishing development of competitive sports worldwide,the demand for accuracy in recognizing sports movements is gradually increasing.The research focuses on the field of competitive sports action recognition,and designs an aerobics action recognition model by combining improved You Only Look Once version 7(YOLOv7)and attention mechanism.During the process,two types of cross stage local network structures were combined,and the loss function was optimized to obtain an improved YOLOv7.Combined with SimAM and spatiotemporal attention mechanism,complete the construction of the recognition model.The experimental results showed that the improved model achieved an accuracy of 90%in recognizing upper limb movements in aerobics,an increase of 28.4%compared to before the improvement.The recognition rate of body posture changes in the model has increased from 64.5%to 90%,and the recall rate has increased from 60%to 91.3%.The results indicate that the competitive sports action recognition model designed in this study can better understand and simulate the spatiotemporal characteristics of movements in competitive sports,and has a driving effect on the demand for action recognition in sports related research and practical application scenarios.
作者
田荣
冯欣
TIAN Rong;FENG Xin(Xi’an Mingde Institute of Technology,Xi’an 710124,China)
出处
《自动化与仪器仪表》
2025年第7期169-173,共5页
Automation & Instrumentation
基金
2023年陕西省体育局常规课题《关于‘双减’政策背景下陕西省体育培训产业可持续发展研究》(2023769)。
