基于密集轨迹和光流二值化图的行为识别算法

doi:10.3778/j.issn.1002-8331.2103-0504

摘要/Abstract

摘要： 针对改进的密集轨迹算法（improved dense trajectories，iDT）提取的轨迹数量较为庞大的问题，提出了一种轨迹滤除方法。密集采样兴趣点，利用光流图计算每个兴趣点下一帧的位置进而组成轨迹，对每帧光流图进行最大值归一化以及二值化，得到光流二值化图，以此反映该点的运动是否相对显著。利用光流二值化图统计轨迹上各点的有效性从而判断轨迹是否满足有效条件，并将不满足条件的轨迹滤除，得到提纯的轨迹。为了验证算法的有效性，使用了行为识别领域的常用数据集KTH和UCF sports对算法进行验证，实验结果表明，该算法能在保证准确率的同时减少轨迹数量，并且计算量较小。

关键词: 行为识别, 改进密集轨迹算法（iDT）, 光流灰度图, 光流二值化, 轨迹滤除

Abstract: To solve the problem of large trajectories extracted by improved dense trajectory（IDT） algorithm, a trajectory filtering method is proposed. Firstly, the interest points are densely sampled, and the next frame position of each interest point is calculated with the help of optical flow image to form the trajectories. Then, the maximum value normalization and binarization of the optical flow image for each frame are carried out to obtain the optical flow binarization image, which reflects whether the movement of the point is relatively significant. The validity of each point on the trajectory is counted by optical flow binarization image to judge whether the trajectory meets the effective condition, and the trajectory that does not meet the condition is filtered out and the purified trajectory is obtained. In order to verify the effectiveness of the algorithm, KTH and UCF sports, commonly used data sets in the field of behavior recognition, are used to verify the algorithm. Experimental results show that the algorithm can reduce the number of trajectories while ensuring the accuracy, and the computation is small compared with other algorithms.

Key words: action recognition, improved dense trajectory（iDT）, grayscale optical flow image, optical flow binarization, trajectory filter

周航, 刘於锡, 龚越, 寇福蔚, 许国梁. 基于密集轨迹和光流二值化图的行为识别算法[J]. 计算机工程与应用, 2022, 58(20): 174-180.

ZHOU Hang, LIU Yuxi, GONG Yue, KOU Fuwei, XU Guoliang. Action Recognition Algorithm Based on Dense Trajectories and Optical Flow Binarization Image[J]. Computer Engineering and Applications, 2022, 58(20): 174-180.

参考文献

[1] WANG H，SCHMID C.Action recognition with improved trajectories[C]//Proceedings of the IEEE International Conference on Computer Vision，2013：3551-3558.
[2] SIMONYAN K，ZISSERMAN A.Two-stream convolutional networks for action recognition in videos[J].Neural Information Processing Systems，2014，1（4）：568-576.
[3] WANG L，QIAO Y，TANG X.Action recognition with trajectory-pooled deep-convolutional descriptors[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2015：4305-4314.
[4] WANG H，KLAESER A，SCHMID C，et al.Dense trajectories and motion boundary descriptors for action recognition[J].International Journal of Computer Vision，2013，103（1）：60-79.
[5] JAIN M，JéGOU H，BOUTHEMY P.Better exploiting motion for better action recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2013：2555-2562.
[6] LAPTEV I，MARSZALEK M，SCHMID C，et al.Learning realistic human actions from movies[C]//2008 IEEE Conference on Computer Vision and Pattern Recognition，2008：1-8.
[7] ALI S，SHAH M.Human action recognition in videos using kinematic features and multiple instance learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2008，32（2）：288-303.
[8] CARMONA J M，CLIMENT J.Human action recognition by means of subtensor projections and dense trajectories[J].Pattern Recognition，2018，81（1）：443-455.
[9] LIAO Z，HU H，LIU Y.Action recognition with multiple relative descriptors of trajectories[J].Neural Processing Letters，2020，51（1）：287-302.
[10] 李岚，罗惠琼.基于时空兴趣点和密集轨迹特征加权融合的行为识别方法研究[J].微电子学与计算机，2017，34（4）：110-114.
LI Lan，LUO Huiqiong.Human action recognition with weighted feature fusion based on STIP and dense trajectory feature[J].Microelectronics & Computer，2017，34（4）：110-114
[11] 王晓芳，齐春.一种运用显著性检测的行为识别方法[J].西安交通大学学报，2018，52（2）：24-29.
WANG Xiaofang，QI Chun.An action recognition method using saliency detection[J].Journal of Xi’an Jiaotong University，2018，52（2）：24-29.
[12] DONG S，HU D，LI R，et al.Human action recognition based on foreground trajectory and motion difference descriptors[J].Applied Sciences，2019，9（10）：21-26.
[13] SENGAR S S，MUKHOPADHYAY S.Moving object area detection using normalized self adaptive optical flow[J].Optik，2016，127（16）：6258-6267.
[14] 戴斌，方宇强，孙振平，等.基于光流技术的运动目标检测和跟踪方法研究[J].科技导报，2009，27（12）：55-60.
DAI Bin，FANG Yuqiang，SUN Zhenping，et al.The detection and tracking of moving objects by using optical flow technique[J].Science & Technology Review，2009，27（12）：55-60.
[15] YI Y，ZHENG Z，LIN M.Realistic action recognition with salient foreground trajectories[J].Expert Systems with Applications，2017，75（6）：44-55.
[16] 鹿天然，于凤芹，杨慧中，等.基于显著性检测和稠密轨迹的人体行为识别[J].计算机工程与应用，2018，54（14）：163-167.
LU Tianran，YU Fengqin，YANG Huizhong.Human action recognition based on dense trajectories with saliency detection[J].Computer Engineering and Applications，2018，54（14）：163-167.
[17] 赵虎.基于改进稠密轨迹的动作识别方法研究[D].兰州：兰州交通大学，2019.
ZHAO Hu.Research on action recognition method based on improved dense trajectories[D].Lanzhou：Lanzhou Jiaotong University，2019.