基于极限学习机的跌倒检测分类识别研究

发布时间：2018-03-14 07:37

本文选题：极限学习机　切入点：跌倒检测　出处：《中国生物医学工程学报》2014年04期 　论文类型：期刊论文

【摘要】：由于年龄和身体条件的限制,在老年人群中跌倒是非常普遍的现象。因此,根据老年人跌倒的运动特征,远程监测他们在各个时间段的状态,以便在其摔倒或突发状况时及时采取措施显得尤为重要。针对人体运动状态进行监测,分析人体运动学特征,提出基于极限学习机的跌倒检测算法。运用三维加速度传感器采集人体的三维加速度值,建立跌倒检测特征模型。在此基础上,建立基于极限学习机的跌倒检测分类器,完成对老年人的计算机辅助跌倒检测。实验数据共540例样本,选用了不同数量的训练集和测试集,其中440例作为训练数据,其余100例为测试数据。测试结果表明,准确率为93%,敏感度为87.5%,特异性为91.7%,具有良好的分类性能。在对分类训练的运行时间方面,基于极限学习机的跌倒检测方法与传统的机器学习方法相比具有明显优势。
[Abstract]:Fall is a common phenomenon in the elderly due to age and physical constraints. Therefore, according to the movement characteristics of falls in the elderly, remote monitoring of their status at all times, In order to take measures in time in case of fall or sudden situation, it is very important to monitor and analyze the kinematics characteristics of human body. A fall detection algorithm based on ultimate learning machine is proposed. The 3D acceleration sensor is used to collect the three-dimensional acceleration value of human body, and the characteristic model of fall detection is established. On this basis, a fall detection classifier based on ultimate learning machine is established. Computer aided fall detection was completed for the elderly. A total of 540 samples were collected, of which 440 were selected as training data and the remaining 100 as test data. The accuracy is 93, the sensitivity is 87.5, the specificity is 91.7, and the classification performance is good. In terms of the running time of classification training, the fall detection method based on extreme learning machine has obvious advantages compared with the traditional machine learning method.
【作者单位】：东北大学中荷生物医学与信息工程学院;东北大学信息科学与工程学院;
【基金】：国家自然科学基金项目(61100022)
【分类号】：R310

【参考文献】