摘要
针对手势识别普遍存在对指尖运动轨迹、加速度变化等细粒度特征提取不足导致复杂微手势的识别精度受限这一挑战,提出融合多维度特征分析的细粒度识别框架(FGRF)。通过部署双指惯性测量单元(IMU)传感器捕获原始信号,从时频域联合分析构建高区分度特征集,结合主成分分析(PCA)与随机森林(RF)分类器实现动作特征的深度挖掘与高效识别。实验基于包含35名受试者的多种数据集验证,本文方法相较主流机器学习与神经网络模型展现出显著优势,识别准确率提升至97.37%。
To address the challenge that gesture recognition systems often suffer from limited recognition accuracy for complex micro-gestures due to insufficient extraction of fine-grained features such as fingertip motion trajectories and acceleration variations,a fine-grained recognition framework(FGRF)that integrates multi-dimensional feature analysis is proposed.By deploying dual-finger inertial measurement unit(IMU)sensors to capture raw signals,the framework constructs a highly discriminative feature set through joint time-frequency domain analysis.Principal component analysis(PCA)and a random forest(RF)classifier are employed to enable deep mining and efficient recognition of motion features.Experimental evaluations conducted on diverse datasets collected from 35 subjects demonstrate that the proposed method significantly outperforms mainstream machine learning and neural network models,achieving a recognition accuracy increased to 97.37%.
作者
燕远征
陈荟慧
陈冠成
关柏良
YAN Yuanzheng;CHEN Huihui;CHEN Guancheng;GUAN Boliang(School of Computer Science and Artificial Intelligence,Foshan University,Foshan 528225,China)
出处
《传感器与微系统》
北大核心
2025年第12期135-140,共6页
Transducer and Microsystem Technologies
基金
国家自然科学基金资金项目(61972092)
广东省教育厅新一代信息技术重点领域专项项目(2022ZDZX1026)。
