累积方向-数量级光流梯度直方图的人体动作识别

PDF下载HTML

Title:

Human action recognition based on accumulated orientation-magnitude histograms of optical flow

作者:

郭利¹, 曹江涛¹, 李平¹, 姬晓飞²

1. 辽宁石油化工大学 信息与控制工程学院, 辽宁 抚顺 113001;
2. 沈阳航空航天大学 自动化学院, 辽宁 沈阳 110136

Author(s):

GUO Li¹, CAO Jiangtao¹, LI Ping¹, JI Xiaofei²

1. School of Information and Control Engineering, Liaoning Shihua University, Fushun 113001, China;
2. School of Automation, Shenyang Aerospace University, Shenyang 110136, China

关键词:

人体动作识别; Horn-Schunck光流; 方向-数量级直方图; 梯度直方图

Keywords:

human action recognition; Horn-Schunck optical flow; orientation-magnitude histograms; gradient histograms

分类号:

TP391.41

DOI:

10.3969/j.issn.1673-4785.201305001

摘要:

为了提高光流信息在人体动作识别系统中应用的效果和效率, 提出一种累计方向-数量级光流梯度直方图的人体动作特征表示方法。该方法首先利用Horn-Schunck充流算法计算图像光流, 然后将光流矢量按照不同的方向-数量级进行直方图统计, 得到单帧图像的方向-数量级的光流梯度直方图, 最后将单帧图像的直方图特征在时间维上进行累积来表示整个视频动作的特征。利用该特征在KTH动作视频库上进行动作识别测试, 4个场景的混合测试得到了87.5%的平均正确识别率, 验证了算法的有效性。

Abstract:

In order to improve the recognition rate and efficiency of optical flow in the human action recognition system, a novel method for human action representation based on the accumulated orientation-magnitude gradient histograms of the optical flow is proposed in this paper. First the image optical flow is computed, and then every flow vector is counted according to the orientation-magnitude to obtain orientation-magnitude histograms of single frame image. Finally information of the video sequence can be represented by accumulating orientation-magnitude histograms in time dimension. The proposed feature is evaluated on a standard database of human actions: KTH. The experiment conducted on the four scenes demonstrates that this algorithm is effective and achieves a correct recognition rate of 87.5% with the KTH dataset.