动作识别是机器视觉领域的基础应用之一,目前动作识别算法多数基于单帧图像特征或简单综合时间维度特征和空间维度特征,一定程度上约束了特征表达能力.为了解决该问题,本文提出了一种时空特征融合方法,将时空金字塔中引入加权策略,有机地将2个维度的特征融合在一起,打破空间维度特征上的局限性.实验结果表明基于本文提出的时空加权特征融合的识别方法可有效提高动作识别精度.
Behavior recognition is a basic application of machine vision. At present, most algorithms for behavior recognition are based on the features from space domain, or other algorithms are simply merge the features on time domain and space domain. Thus these methods restrain the capability of representation. To deal with this problems, we propose a time-space domain feature fusion method. In this method, we introduce the spatiotemporal weight strategy into time-space pyramid. By doing this, we can break the limitations of feature space, integrate the features from the two dimensions together. The experiments show that our spatiotemporal weighting method on multi-feature fusion can improve the accuracy of behavior recognition.
[1] 杨敬伟,杨文柱.基于带通滤波和绿色因子的绿色识别方法[J].河北大学学报(自然科学版),2015,35(3):305-310. DOI:10.3969/j.issn.1000-1565.2015.03.013.
[2] DEAN T A, SINGH S S, JASRA A, et al. Parameter estimation for hidden markov models with intractable likelihoods[J]. Scandinavian Journal of Statistics, 2015, 41(4):970-987.DOI: 10.1111/sjos.12077.
[3] WANG H, SCHMID C. Action recognition with improved trajectories[C] // IEEE International Conference on Computer Vision, Sydney, Australia, 2013:3551-3558. DOI: 10.1109/ICCV.2013.441.
[4] MAQUEDA A I, RUANO A, DEL-BLANCO C R, et al. Novel multi-feature bag-of-words descriptor via subspace random projection for efficient human-action recognition[C] // IEEE International Conference on Advanced Video and Signal Based Surveillance, Karlsruhe, Germany, 2015:1-6. DOI: 10.1109/AVSS.2015.7301736.
[5] ROSS D A, LIM J, LIN R S, et al. Incremental learning for robust visual tracking[J]. International Journal of Computer Vision, 2008, 77(1-3):125-141. DOI:10.1007/s11263-007-0075-7.
[6] DALAL N, TRIGGS B. Histograms of oriented gradients for human detection[C] // IEEE Computer Society Conference on Computer Vision & Pattern Recognition. IEEE Computer Society, San Diego, USA, 2005:886-893. DOI: 10.1109/CVPR.2005.177.
[7] RANJAN A, BLACK M J. Optical flow estimation using a spatial pyramid network[C] // IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017:2720-2729.DOI: 10.1109/CVPR.2017.291.
[8] WANG Q, MA Q, LUO C H, et al. Hybrid histogram of oriented optical flow for abnormal behavior detection in crowd scenes[J]. International Journal of Pattern Recognition & Artificial Intelligence, 2016, 30(2):1884-1897.DOI: 10.1142/S0218001416550077.
[9] 王宇新, 郭禾, 何昌钦,等. 用于图像场景分类的空间视觉词袋模型[J]. 计算机科学, 2011, 38(8):265-268. DOI:10.3969/j.issn.1002-137X.2011.08.064.
[10] WANG H, KLASER 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.DOI: 10.1007/s11263-012-0594-8.
[11] SOOMRO K, ZAMIR A R, SHAH M. UCF101: A dataset of 101 human actions classes from videos in the wild[EB/OL].(2012-09-03)[2018-05-12].https://arxiv.org/abs/1212.0402vl.
