改进的归一化转动惯量对人体跌倒的识别

刘丽伟; 薛春芳; 张宏美; 余学帆

doi:10.3788/OPE.20172513.0312

您当前的位置：

首页 >

文章列表页 >

改进的归一化转动惯量对人体跌倒的识别

信息科学 | 更新时间：2020-08-13

- 改进的归一化转动惯量对人体跌倒的识别
- Recognition of human tumbles based on improved normalized inertia moment
- 光学精密工程 2017年25卷第10s期页码：312-317
- 作者机构：
  
  长春工业大学计算机科学与工程学院, 吉林长春 130012
- 作者简介：
- 基金信息：
  
  吉林省教育厅“十三五”科学技术研究项目（No.2016-35）()
- DOI：10.3788/OPE.20172513.0312
  中图分类号： TP18;TP212.9
- 收稿日期：2017-06-01，
  
  修回日期：2017-06-22，
  
  纸质出版日期：2017-11-25
- 稿件说明：
移动端阅览
刘丽伟, 薛春芳, 张宏美等. 改进的归一化转动惯量对人体跌倒的识别[J]. 光学精密工程, 2017,25(10s): 312-317

LIU Li-wei, XUE Chun-fang, ZHANG Hong-mei etc. Recognition of human tumbles based on improved normalized inertia moment[J]. Editorial Office of Optics and Precision Engineering, 2017,25(10s): 312-317
刘丽伟, 薛春芳, 张宏美等. 改进的归一化转动惯量对人体跌倒的识别[J]. 光学精密工程, 2017,25(10s): 312-317 DOI： 10.3788/OPE.20172513.0312.

LIU Li-wei, XUE Chun-fang, ZHANG Hong-mei etc. Recognition of human tumbles based on improved normalized inertia moment[J]. Editorial Office of Optics and Precision Engineering, 2017,25(10s): 312-317 DOI： 10.3788/OPE.20172513.0312.

摘要

为了实现视频中人体跌倒行为的准确快速检测，提出一种改进的归一化转动惯量（NMI）特征提取算法。首先利用加速鲁棒性特征（SURF）算法构建特征描述子的理念，并结合NMI特征值提取算法求出特征点的NMI值，然后将视频图像中NMI值相差最小的两个特征点确定为匹配点，并求出两幅图像中相匹配的总特征点数，最后将求出的特征点匹配数目与跌倒过程中特征点匹配数目的最小值相比，以此来判断视频中是否出现人体跌倒行为。实验结果表明，改进的NMI特征提取算法对跌倒检测的准确率高达96%，且与其他算法相比检测速度快，平均用时只有0.138 s。改进后的算法基本实现了行为检测的准确率高、速度快、占用内存小等要求，是一种行之有效的跌倒识别方法。

Abstract

In order to detect human tumble behavior in videos quickly and accurately

an improved Normalized Moment of Inertia (NMI) feature extraction algorithm was proposed

which used the idea of SURF to structure feature retrieve

and combine SURF with NMI to calculate the NMI values. Taking the feature points with the closet NMI values in the two images of video as the matching points

the total of matching points were calculated between the two images. Moreover

the number of the feature points was compared with the minimum of feature points in a tumble video

thus judging whether a tumble behavior has happened or not. The results show that accuracy rate of this improved NMI algorithm runs up to 96%

and the average recognition time is 0.138 s

which is faster than other algorithms. The algorithm occupies a little internal storage and detects quickly and accurately

which is a feasible and effective approach to detect human actions.

关键词

Keywords

references

裴利然,姜萍萍,颜国正. 基于支持向量机的跌倒检测算法的研究[J]. 光学精密工程,2017,25(1):182-187. PEI L R, JIANG P P, YAN G ZH. Research on fall detection system based on support vector machine[J]. Opt. Precision Eng., 2017, 25(1):182-187. (in Chinese)

张宇洋,刘满华,韩韬. 基于Mean Shift图像分割和支持向量机判决的候梯人数视觉检测系统[J]. 光学精密工程, 2013, 21(4):1079-1085. ZHANG Y Y,LIU M H,HAN T. Elevator-waiting people counting system based on Mean Shift segmentation and SVM classification[J]. Opt. Precision Eng., 2013, 21(4):1079-1085. (in Chinese)

蒲兴成, 王涛, 张毅. 基于改进Hu矩算法的Kinect手势识别[J]. 计算机工程, 2016, 42(7):165-172. PU X CH,WANG T,ZHANG Y. Kinect gesture recognition based on improved Hu moment algorithm[J]. Computer Engineering, 2016, 42(7):165-172. (in Chinese)

SULTANI W, SALEEMI I. Human action recognition across datasets by foreground-weighted histogram decomposition[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014:764-771.

HASAN M. Roy-Chowdhury A K. Incremental activity modeling and recognition in streaming videos[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014:796-803.

罗家祥,林畅赫,王加朋,等. 结合深度卷积网络与加速鲁棒性特征配准的图像精准定位[J]. 光学精密工程, 2017,25(2):469-476. LUO J X, LIN CH H,WANG J P, et al..Accurate image locating combining deep convolution network with SURF registering[J]. Opt. Precision Eng., 2017, 25(2):469-476. (in Chinese)

苗权, 谷延锋. 视频中基于场景变化分类的在线SURF特征匹配[J]. 小型微型计算机系统, 2016, 37(12):2760-2764. MIAO Q,GU Y F. On-line sorted SURF feature matching adapting to different scene changes[J]. Journal of Chinese Computer Systems, 2016, 37(12):2760-2764. (in Chinese)

屈冰广,杨晓苹. 基于基准点和NMI的手背静脉识别算法研究[D].天津:天津理工大学, 2016. QU B G, YANG X P. Hand Vein Recognition Based on Reference Point and NMI[D]. Tianjin:Tianjin university of Technology,2016. (in Chinese)

朱月秀,陈志翔. 基于n阶R-NMI特征的手势识别[J]. 漳州师范学院学报,2013, (2):28-30. ZHU Y X, CHEN ZH X. Gesture recognition based on n order R-NMI feature[J]. Journal of Zhangzhou Normal University, 2013(2):28-30. (in Chinese)

陈莹,朱明,刘剑. 高斯混合模型自适应微光图像增强[J]. 液晶与显示,2015,30(2):300-309. CHEN Y,ZHU M,LIU J. Automatic low light level image enhancement using Gaussian mixture modeling[J]. Chinese Journal of Liquid Crystals and Displays, 2015, 30(2):300-309. (in Chinese)

杨鹏生,吴晓军,张玉梅. 改进扩展卡尔曼滤波算法的目标跟踪算法[J]. 计算机工程与应用,2016,52(5):71-74. YANG P SH, WU X J, ZHANG Y M. Target tracking method based on improved Kalman filter algorithm[J]. Computer Engineering and Applications, 2016, 52(5):71-74. (in Chinese)

浏览量

617

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于分数阶微分的高光谱图像特征提取与分类

面向高光谱影像场景分类的轻量化深度全局-局部知识蒸馏网络

多分支无锚框网络密集行人检测算法

SENet生成对抗网络在图像语义描述中的应用

面向高光谱图像分类的半监督双流网络