融合改进RF算法的人体姿态识别方法在运动训练领域的应用
Application of Human Posture Recognition Method Integrated with Improved RF Algorithm in Modern Intelligent Engineering
温博1
作者信息
- 1. 西安翻译学院体育学院,西安 710105
- 折叠
摘要
对人体姿态识别及现代智能化工程设计成为人机交互领域的重要研究方向进行了研究;在实现更高效、智能的人体姿态识别中,采用了基于DBSCAN-RF算法的分类训练器,同时对RF算法加以改进,引入了 HD-SMOTE方法;该方法的技术创新和独特之处在于结合了密度聚类和随机森林的优点,能够有效地处理带有噪声的数据集,并具有较高的计算效率和可扩展性;通过实验测试,DBSCAN-RF算法的识别召回率最高达到了 98.64％,相比于传统的RF算法、K-means-RF以及Mean-shift-RF算法,其数值分别增加了 6.37％、4.28％、3.95％;同时,DBSCAN-RF算法在跌倒和正常走路的识别召回率分别达到了95.31％和96.48％;此外,DBSCAN-RF算法的测试时间均低于62 ms;经实际应用满足了现代智能化的人体姿态识别工程上的应用,为现代智能化的人体姿态识别提供了可靠的技术支持.
Abstract
Research on human pose recognition and modern intelligent engineering design has become an important research direc-tion in the field of human-computer interaction.In achieving more efficient and intelligent human pose recognition,this paper presents a classification trainer based on density-based spatial clustering of applications with random forest(DBSCAN-RF)algorithm,and im-proves the random forest(RF)algorithm by introducing the synthetic minority oversampling technique based on high-dimensional data(HD-SMOTE)method.The technological innovation and uniqueness of this method lies in the combination of density clustering and random forest advantages,which can effectively handle the datasets with noise and has high computational efficiency and scalability.Through the experimental testing,the recognition recall rate of the DBSCAN-RF algorithm reaches the highest level of 98.64％,which increases by 6.37％,4.28％,and 3.95％respectively compared with traditional RF algorithm,K-means-RF and Mean-shift-RF algorithm.Meanwhile,the recognition recall rate of the DBSCAN-RF algorithm reaches 95.31％and 96.48％for falls and normal walking,respectively.Moreover,the test time of the DBSCAN-RF algorithm is all lower than 62 ms.It meets the application of modern intelligent body posture recognition engineering,and provides a reliable technical support for modern intelligent body posture recognition.
关键词
DBSCAN-RF/分类训练器/人体姿态识别/现代智能化工程Key words
DBSCAN-RF/classification trainer/human pose recognition/modern and intelligent engineering引用本文复制引用
基金项目
2023年陕西省体育局常规课题(2023723)
2022年陕西省"十四五"教育科学计划项目(SGH22Y1789)
出版年
2024
NETL
NSTL
万方数据