基于BlazePose和KNN的健身计数系统设计与实现

Design and Implementation of a Fitness Counting System Based on BlazePose and KNN

目前,健身动作的识别与计数受大模型的训练以及动作种类繁多等影响,少有实时性、准确性、稳定性等各方面均表现优异的健身动作识别与计数系统。该系统利用Blaze Pose进行动作识别,以满足健身动作识别的实时性和稳定性的要求,使用KNN算法实现动作的分类与计数功能,利用Tkinter实现交互式界面,使系统具备可交互性。该系统仅需较少的自采集数据集即可实现动作识别与计数功能,测试结果表明该系统达到95.5%的计数准确率和30 FPS的实时检测速度,可广泛应用于健身场所和线上健身平台。

Currently,recognition and counting of fitness pose are limited by factors such as the training of large models and the diversity of movement types,which often results in poor real-time performance,accuracy,and stability in fitness pose recognition and counting systems.The proposed system aims to address these shortcomings by using BlazePose for action recognition,thereby meeting the requirements of real-time and stable identification of fitness poses.The K-Nearest Neighbors(KNN)algorithm is employed to facilitate movement classification and counting.Additionally,Tkinter is used to realize an interactive interface,which enhances the system's interactivity.A noteworthy aspect of this system is its ability to perform pose recognition and counting functions with a minimal self-collected dataset.Test results demonstrate that the system achieves a count accuracy rate of 95.5%and real-time detection speed of 30 Frames per Second(FPS),making it suitable for broad applications in fitness venues and online fitness platforms.