基于双目视觉的帕金森症状量化识别方法研究

Research on the Identification Method of Parkinson’s Disease Based on Binocular Vision

为了实现帕金森病量化评估,同时降低评估过程中病人身体负担以及空间限制,提出了一种基于双目视觉的帕金森病识别方法。首先利用双目相机并配合光学标记点采集病人运动数据,通过目标跟踪和阈值分割提取标记点图像,其次对标记点图像进行边缘提取和最小二乘法圆心拟合得到圆心像素坐标,通过双目定位原理计算圆心三维坐标得到人体运动信号,最后在运动信号基础上进行特征参数提取并构建SVM分类模型进行帕金森病识别。通过双目视觉系统进行数据采集并构造特征参数数据集进行帕金森病识别实验,实验结果表明,对帕金森病识别的准确率和F1分数分别达到95%和95.6%,可以辅助医生对帕金森病进行诊断。

To achieve a quantitative assessment of PD and reduce the physical burden and spatial limitation of the patient during the assessment process,a binocular vision-based PD recognition method is proposed.Firstly,the patient’s motion data is collected by a binocular camera with optical marker points.The marker point images are then extracted by target tracking and threshold segmentation.Secondly,edge extraction and least-squares circle fitting are performed on the marker point images to obtain the pixel coordinates of the circle center.The human motion signal is obtained by calculating the 3D coordinates of the circle center by binocular localization principle.Finally,feature parameters are extracted based on the motion signal.Thus,a SVM classification model is constructed for Parkinson’s disease recognition.The experimental results show that the accuracy and F1 scores of Parkinson’s disease identification using the developed method can reach 95%and 95.6%,respectively,which can assist doctors in the diagnosis of Parkinson’s disease.