节点文献
基于三维姿态估计的智慧体能计数算法
Intelligent Physical Fitness Counting Algorithm Based on Three-Dimensional Posture Estimation
【摘要】 针对目前体能训练中普遍采用人工监督计数所带来的效率低、误差大的问题,设计了一种Fast-3D-Pose-Counter智慧体能计数算法。该算法首先利用改进的YOLO_v3网络对单目RGB摄像头采集的视频进行目标检测,得到人体目标区域。然后使用SimplePose网络进行二维姿态估计,并将人体关键点的二维坐标输入3D Pose Baseline网络,得到三维坐标系下的人体姿态。最后基于KNN算法设计动作计数分类器对三维特征向量进行分类,实现有效动作计数。实验结果表明,本算法的推理速度达到了27.4 FPS,计数准确率达到了99.6%,具备很好的实用性。
【Abstract】 Aiming at the low efficiency and large error caused by the general use of manual supervision and counting in physical fitness training, a Fast-3D-Pose-Counter intelligent physical fitness counting algorithm is designed. The algorithm first uses the improved YOLO_v3 network to perform target detection on the video collected by the monocular RGB camera to obtain the target area of the human body. Then use the SimplePose network to estimate the two-dimensional pose, and input the twodimensional coordinates of the key points of the human body into the 3D Pose Baseline network to obtain the human pose in the three-dimensional coordinate system. Finally, based on the KNN algorithm, an action counting classifier is designed to classify the three-dimensional feature vector to achieve effective action counting. Experimental results show that the inference speed of this algorithm reaches 27 FPS, and the counting accuracy rate reaches 99.6%, which is very practical.
【Key words】 physical fitness counting; target detection; attitude estimation; KNN algorithm;
- 【文献出处】 现代计算机 ,Modern Computer , 编辑部邮箱 ,2022年07期
- 【分类号】TP391.41
- 【下载频次】55