拟合稀疏信号的可穿戴设备混合现实视觉人机交互方法

张恒1; 任晓康1; 郝飞1; 高亚娟2; 王天宇3; 谢利德1

拟合稀疏信号的可穿戴设备混合现实视觉人机交互方法

引用本文：张恒1，任晓康1，郝飞1，高亚娟2，王天宇3，谢利德1.拟合稀疏信号的可穿戴设备混合现实视觉人机交互方法[J].计算技术与自动化,2025,(1):64-69

摘要点击次数: 920

全文下载次数: 227

作者	单位
张恒1，任晓康1，郝飞1，高亚娟2，王天宇3，谢利德1	（1. 承德医学院生物医学工程系,河北承德 067000；2. 北京大学第三医院放射科,北京 100191；3. 北京大学医学部医学技术研究院,北京 100191）

中文摘要:常规人机交互方法主要依托于深度自适应算法，缺少对动作特征的重构，无法对待检测样本进行动态对齐，使得动作识别错误率较高。为此，提出了拟合稀疏信号的可穿戴设备混合现实视觉人机交互方法。基于骨骼信息融合原理获取动作动态，并将动作图像从深度图像中分割出来，通过将动作图像进行超平面转换，得到动作参数峰值。结合动作边缘方程提取动作特征值，采用拟合稀疏信号算法对特征点进行空域离散化处理，利用稀疏信号阵列模型对动作特征进行重构与动态对齐，从而近似分类与识别动作类型，借助鼠标控制pynput库对角色进行动作驱动，由此实现人机交互。以可穿戴设备混合现实视觉数据集作为实验对象，将所提方法应用于人机交互的动作识别，结果表明，所提方法对于不同类别的动作识别具有更低的识别错误率。

中文关键词:拟合稀疏信号可穿戴设备混合现实视觉人机交互

Wearable Device Hybrid Reality Visual Human-machine Interaction Method for Fitting Sparse Signals

Abstract:The conventional human-computer interaction methods mainly rely on deep adaptive algorithms. Due to the lack of reconstruction of action features, they cannot dynamically align the detected samples, resulting in a high error rate in action recognition.To this end, a wearable device hybrid reality visual human-machine interaction method that fits sparse signals is proposed. Based on the principle of bone information fusion, motion dynamics are obtained, and the motion image is segmented from the depth image. Through the hyperplane transformation of the motion image, the peak value of the motion parameters is obtained. Combined with the action edge equation, the action feature values are extracted, and the feature points are processed by spatial discretization using the sparse signal fitting algorithm. The action features are reconstructed and dynamically aligned using the sparse signal array model, Thus, approximate classification and recognition of action types can be achieved, and the pynput library can be controlled by the mouse to drive the actions of characters, thereby achieving human-machine interaction. Using a wearable device mixed reality visual dataset as the experimental object, the proposed method was applied to action recognition in human-computer interaction. The results showed that the proposed method has a lower recognition error rate for different types of action recognition.

keywords:fitting sparse signals wearable devices hybrid reality vision human-machine interaction

查看全文 查看/发表评论 下载pdf阅读器