摘要
针对滑雪人员目标检测研究中,存在的检测精度低、速度慢,不同姿态识别效果差等问题,采用YOLOv5s网络模型,改进损失函数,增加平衡因子,在自制滑雪人员数据集上对网络进行训练,利用训练好的网络进行图像特征提取,实现滑雪人员的快速检测。基于YOLOv5s的滑雪人员检测模型可以有效识别不同姿态下的滑雪人员,mAP值达到99.87%,Recall值达到97.66%,检测速度可以达到7ms/帧。实验结果表明,改进的YOLOv5s滑雪人员检测模型,检测速度快,检测精度高,鲁棒性强,有较好的可扩展性,既满足检测精度要求,又满足检测速度要求。
Aiming at the problems of low detection accuracy,slow speed and poor recognition effect of different postures in the target detection of skiers.In this paper,the YOLOv5 s network model is used to improve the loss function and increase the balance factor.The network is trained on the self-made dataset of skiers.The trained network is used for image feature extraction to realize the rapid detection of skiers.The ski personnel detection model based on YOLOv5s can effectively identify the ski personnel under different postures.The mAP value reaches 99.87 %,the Recall value reaches 97.66 %,and the detection speed can reach 7 ms/frame.The experimental results show that the improved YOLOv5s ski personnel detection model proposed in this paper has fast detection speed,high detection accuracy,strong robustness and good scalability,which meets the requirements of both detection accuracy and detection speed.
作者
彭雅坤
曹伊宁
刘晓群
Liu Xiaoqun;Peng Yakun;Cao Yining(College of Information Engineering,Hebei University of Architecture,Zhangjiakou 075000,China)
出处
《长江信息通信》
2021年第8期24-26,共3页
Changjiang Information & Communications
基金
河北省技术创新引导计划项目-科技冬奥专项资助《基于5G的VR场景下冰雪突发事故高精度定位技术研究》(20470302D)。
