摘要
针对传统的运动参数提取方法一直存在提取误差大、耗时长的问题,提出基于图像识别技术的中老年人下肢动作运动图像参数提取方法,使人体运动行为识别能力得到提升。首先,结合中老年人下肢运动速度特征和三维运动形状的时空梯度自相关特征,计算出边缘梯度方向空间分布与梯度之间的自相关性,将时空自相关特征与视频运动特征相结合,使特征识别具备相应的数据条件;其次,人体下肢动作的视频图像数据是典型的时间序列数据,因此,基于人体骨架局部特征,利用训练数据能够构造完备字典,完成数据编码,运用时域金字塔匹配法对编码后的向量进行下肢动作运动图像特征参数提取与识别。实验结果证明,利用基于图像识别技术对中老年人下肢动作运动图像参数实现了准确有效的提取。
The traditional motion parameter extraction method has big extraction error and long time-consumption. Therefore, an image recognition based motion parameter extraction method of lower limbs movement for elderly people is proposed to improve the recognition ability of human motion behavior. On the basis of the speed characteristics of lower limbs movement for the middle-aged and elderly people and the spatiotemporal gradient correlation characteristic of the three-dimensional motion shape, the autocorrelation between the spatial distribution and gradient in the edge gradient direction is solved. The spatiotemporal autoeorrelation characteristic and video motion feature are combined to satisfy the corresponding data condition of the feature recognition. Because the video image data of human lower limbs movement acts as the typical time series data, the training data is used to construct the complete dictionary according to the local feature of the human skeleton to realize the data encoding. The time domain pyramid matching method is adopted to extract and recognize the characteristic parameter of the lower limbs motion image for the encoded vector. The experimental results show that the proposed method based on image recognition technology can extract the image parameters of the lower limbs movement for the middle-aged and elderly people effectively.
出处
《现代电子技术》
北大核心
2018年第1期71-75,80,共6页
Modern Electronics Technique
基金
基金项目:人口老龄化背景下老年人体育健康促进研究(16TYB02)~~
关键词
图像识别
下肢动作
自相关性
运动行为识别
时域金字塔匹配法
参数提取
image recognition
lower limbs movement
autocorrelation
motion behavior recognition
time domain pyramid matching method
parameter extraction
