人物交互检测研究进展综述

龚勋; 张志莹; 刘璐; 马冰; 吴昆伦

doi:10.3969/j.issn.0258-2724.20210339

人物交互检测研究进展综述

doi: 10.3969/j.issn.0258-2724.20210339

龚勋^{1, 2,},
张志莹²,
刘璐²,
马冰²,
吴昆伦¹

1.
西南交通大学计算机与人工智能学院，四川成都 611756
2.
西南交通大学唐山研究生院，河北唐山063000

基金项目: 国家自然科学基金（61876158）；中央高校基本科研业务费专项资金（2682021ZTPY030）

详细信息

作者简介:
龚勋（1980—），男，教授，博士生导师，研究方向为计算机视觉与模式识别，E-mail： xgong@swjtu.edu.cn

中图分类号: TP391
计量
- 文章访问数: 1127
- HTML全文浏览量: 716
- PDF下载量: 233
- 被引次数: 0
出版历程
- 收稿日期: 2021-04-28
- 修回日期: 2021-09-14
- 刊出日期: 2021-10-27

A Survey of Human-Object Interaction Detection

GONG Xun^{1, 2
,},
ZHANG Zhiying²,
LIU Lu²,
MA Bing²,
WU Kunlun¹

1.
School of Computing and Artificial Intelligence, Southwest Jiaotong University, Chengdu 611756, China
2.
Graduate School of Tangshan, Southwest Jiaotong University, Tangshan 063000, China

摘要

摘要:
作为目标检测、行为识别、视觉关系检测的交叉学科，人物交互（human-object interaction，HOI）检测旨在识别特定应用场景下人与物体的相互关系. 本文对基于图像的人物交互检测研究成果进行了系统总结及论述. 首先，从交互关系建模的原理出发，把人物交互检测方法分为基于全局实例和基于局部实例两类，并对代表性方法进行了详细阐述和分析；进而，根据所采用视觉特征的差异将基于全局实例的方法进行细分，包括融合空间位置信息、融合外观信息与融合人体姿态信息；然后，讨论了零样本学习、弱监督学习以及Transformer模型在人物交互检测中的应用；最后，从交互类别、视觉干扰以及运动视角三方面出发，总结了人物交互检测面临的挑战，并指出领域泛化、实时检测和端到端网络是未来发展的趋势.
- 人物交互 /
- 视觉关系 /
- 目标检测 /
- 动作识别
Abstract:
As an interdisciplinary subject of object detection, action recognition and visual relationship detection, human-object interaction (HOI) detection aims to identify the interaction between humans and objects in specific application scenarios. Here, recent work in the field of image-based HOI detection is systematically summarized. Firstly, based on the theory of interaction modeling, HOI detection methods can be divided into two categories: global instance based and local instance based, and the representative methods are elaborated and analyzed in detail. Further, according to the differences in visual features, the methods based on the global instance are further subdivided into fusion of spatial information, fusion of appearance information and fusion of body posture information. Finally, the applications of zero-shot learning, weakly supervised learning and Transformer model in HOI detection are discussed. From three aspects of HOI, visual distraction and motion perspective, the challenges faced by HOI detection are listed, and it is pointed out that domain generalization, real-time detection and end-to-end network are the future development trends.
- human-object interaction (HOI) /
- visual relationship /
- object detection /
- action recognition

HTML全文

图 1 人物交互关系检测流程

Figure 1. Flowchart of HOI detection

下载: 全尺寸图片幻灯片

图 2 人物交互的相对空间关系

Figure 2. Relative spatial relationship in HOI

下载: 全尺寸图片幻灯片

图 3 基于人-物区域位置信息的HO-RCNN网络

Figure 3. HO-RCNN networks based on human-object regional information

下载: 全尺寸图片幻灯片

图 4 空间引发模型框架

Figure 4. Framework of spatial priming model

下载: 全尺寸图片幻灯片

图 5 PPDM与同类方法在HICO-DET上的推理时间、平均准确度以及速度

Figure 5. Inference time, m_AP, speed between PPDM and similar methods on HICO-DET dataset

下载: 全尺寸图片幻灯片

图 6 融合视觉语义姿态特征的VSP-GMN网络

Figure 6. VSP-GMN network integrating visual, semantic and pose features

下载: 全尺寸图片幻灯片

图 7 整体人体姿态与局部特征的对比

Figure 7. Comparison of overall human posture and local features

下载: 全尺寸图片幻灯片

图 8 基于实例行为的局部标签

Figure 8. Local annotations based on instance behavior

下载: 全尺寸图片幻灯片

图 9 零样本目标检测流程

Figure 9. Flowchart of zero-shot object detection

下载: 全尺寸图片幻灯片

表 1 人物交互图像数据集

Table 1. HOI image datasets

数据集	发布机构/发布者	图片/张	动作/类	注释类型	示例
HAKE （2020）^[13]	上海交通大学	118 000	156	检测框结合语义描述
HOI-A （2020）^[14]	北京航空航天大学	17 606	10	检测框结合语义描述
HICO-DET （2018）^[12]	密西根大学安娜堡分校	48 000	117	检测框结合语义描述
HCVRD （2018）^[15]	澳大利亚阿德莱德大学	788 160	9 852	检测框结合语义描述
V-COCO （2015）^[11]	加州大学伯克利分校	10 346	80	检测框
HICO （2015）^[9]	密西根大学安娜堡分校	47 774	600	语义描述
MPII （2014）^[10]	普朗克信息学研究所	40 522	410	2D 姿态
TUHOI （2014）^[8]	意大利特伦托波沃大学	10 805	2 974	语义描述

下载: 导出CSV

表 2 基于视觉特征方法在HICO-DET数据集的m_AP结果对比

Table 2. Result comparison of m_AP with visual feature based methods on HICO-DET data set

来源	年份	方法	特征	m_AP（Default）/%			m_AP（Know Object）/%
来源	年份	方法	特征	Full	Rare	None-Rare	Full	Rare	None-Rare
文献[46]	2021 年	GGNet	A	29.17	22.13	30.84	33.50	26.67	34.89
文献[14]	2020 年	PPDM	A	21.73	13.78	24.10	24.58	16.65	26.84
文献[17]	2020 年	VS-GATs + PMN	A + S + P	21.21	17.60	22.29
文献[45]	2021 年	多级成对特征网络	A + S + P	20.05	16.66	21.07	24.01	21.09	24.89
文献[34]	2019 年	PMFNet	A + S + P	17.46	15.65	18.00	20.34	17.47	21.20
文献[31]	2019 年	TIN	A + S + P	17.22	13.51	18.32	19.38	15.38	20.57
文献[20]	2019 年	多分支网络	A + S	16.24	11.16	17.75	17.73	12.78	19.21
文献[19]	2018 年	iCAN	A + S	14.84	10.45	16.15	16.26	11.33	17.73
文献[12]	2018 年	HO-RCNN	A + S	7.81	5.37	8.54	10.41	8.94	10.85
注：A、S、P 分别表示外观、空间和骨架；Full、Rare、None-Rare 分别表示完整类、罕见类、非罕见类.

下载: 导出CSV

表 3 基于视觉特征方法在V-COCO数据集结果对比

Table 3. Results comparison of visual feature based methods on V-COCO data set

来源	年份	方法	特征	A_P/%
文献[46]	2021 年	GGNet-Hourglass	A	54.70
文献[45]	2021 年	多级成对特征网络	A + S + P	52.80
文献[17]	2020 年	VS-GATs	A + S + P	49.80
文献[34]	2019 年	PMFNet	A + S + P	52.00
文献[31]	2019 年	TIN	A + S + P	48.70
文献[41]	2019 年	Body-part	A + P	47.53
文献[20]	2019 年	多分支网络	A + S	47.30
文献[19]	2018 年	iCAN	A + S	45.30
文献[24]	2017 年	InteractNet	A	40.00

下载: 导出CSV

表 4 其他新技术总结

Table 4. Summary of other new technologies

来源	年份	数据集	方法	主要工作概述	m_AP/%
文献[57]	2021 年	V-COCO	QPIC	利用注意力机制有效地聚合特征以检测各种 HOI 类	58.80
文献[56]	2021 年	HICO-DET	HOTR	首次提出基于Transformer编码器-解码器结构的预测框架	25.10
文献[54]	2020 年	HICO	GCNCL	以弱监督的方式训练未见类	16.02
文献[51]	2019 年	COCO-a UnRel HICO-DET	三联体模型	提出了一个类比迁移模型，可计算从未见过的视觉短语嵌入信息	7.30 17.50 20.90
文献[50]	2019 年	VT60	ZSL + S2S	语义到空间体系结构融合了零样本学习共同捕获信息并查询	50.47
文献[49]	2018 年	ILSVRC-2017 Unseen (all) Seen	端到端的深度架构	将零样本任务扩展到目标检测领域，联合建模了视觉和语义领域信息融合的端到端深度网络	16.40 26.10
文献[48]	2018 年	HICO-DET	多任务训练网络	通过零样本学习方法将 HOI 识别扩展到长类别，实现对未见的动词-对象对的零样本目标检测	6.46

下载: 导出CSV

参考文献(57)

[1]	JOHNSON J, KRISHNA R, STARK M, et al. Image retrieval using scene graphs[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE Computer Society, 2015: 3668-3678.
[2]	LI Y K, OUYANG W L, ZHOU B L, et al. Scene graph generation from objects, phrases and region captions[DB/OL]. (2017-06-31)[2021-02-02]. https://arxiv.org/abs/1707.09700.
[3]	XU D F, ZHU Y K, CHOY C B, et al. Scene graph generation by iterative message passing[EB/OL]. (2017-01-10)[2021-02-02]. https://arxiv.org/abs/1701.02426.
[4]	BERGSTROM T, SHI H. Human-object interaction detection: a quick survey and examination of methods[DB/OL]. (2020-09-27)[2021-02-02]. https://arxiv.org/abs/2009.12950.
[5]	GUPTA A, KEMBHAVI A, DAVIS L S. Observing human-object interactions: using spatial and functional compatibility for recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(10): 1775-1789. doi: 10.1109/TPAMI.2009.83
[6]	ALESSANDRO P, CORDELIA S, VITTORIO F. Weakly supervised learning of interactions between humans and objects[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(3): 601-614. doi: 10.1109/TPAMI.2011.158
[7]	LI L J, LI F F. What, where and who? Classifying events by scene and object recognition[C]//Proceedings of IEEE International Conference on Computer Vision. [S.l.]: IEEE, 2007: 1-8.
[8]	LE D T, UIJLINGS J, BERNARDI R. TUHOI: trento universal human object interaction dataset[C]// Proceedings of the Third Workshop on Vision and Language. Brighton: Brighton University, 2014: 17-24.
[9]	CHAO Y W, WANG Z, HE Y, et al. HICO: a benchmark for recognizing human-object interactions in images[C]//IEEE International Conference on Computer Vision. [S.l.]: IEEE, 2015: 1-9.
[10]	ANDRILUKA M, PISHCHULIN L, GEHLER P, et al. 2d human pose estimation: New benchmark and state of the art analysis[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2014: 3686-3693.
[11]	GUPTA S, MALIK J. Visual semantic role labeling[DB/OL]. (2015-03-17)[2021-02-02]. https://arxiv.org/abs/1505.04474.pdf.
[12]	CHAO Y W, LIU Y, LIU X, et al. Learning to detect human-object interactions[C]//2018 IEEE Winter Conference on Applications of Computer Vision. [S.l.]: IEEE, 2018: 381-389.
[13]	LI Y L, XU L, LIU X, et al. Pastanet: Toward human activity knowledge engine[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2020: 379-388.
[14]	LIAO Y, LIU S, WANG F, et al. PPDM: Parallel point detection and matching for real-time human-object interaction detection[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2020: 479-487.
[15]	ZHUANG B, WU Q, SHEN C, et al. Hcvrd: a benchmark for large-scale human-centered visual relationship detection[C/OL]//Proceedings of the AAAI Conference on Artificial Intelligence, 2018. [2021-02-22]. https://ojs.aaai.org/index.php/AAAI/article/view/12260.
[16]	XU B J, LI J N, YONGKANG W, et al. Interact as You intend:intention-driven human-object interaction detection[J]. IEEE Transactions on Multimedia, 2019, 22(6): 1423-1432.
[17]	ULUTAN O, IFTEKHAR A S M, MANJUNATH B S. Vsgnet: spatial attention network for detecting human object interactions using graph convolutions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2020: 13617-13626.
[18]	GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. [S.l.]: IEEE, 2015: 1440-1448.
[19]	GAO C, ZOU Y, HUANG J B. iCAN: instance-centric attention network for human-object interaction detection[DB/OL]. (2018-08-30)[2021-02-22]. https://arxiv.org/abs/1808.10437.
[20]	WANG T, ANWER R M, KHAN M H, et al. Deep contextual attention for human-object interaction detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. [S.l.]: IEEE, 2019: 5694-5702.
[21]	PENG C, ZHANG X, YU G, et al. Large kernel matters－improve semantic segmentation by global con- volutional network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2017: 4353-4361.
[22]	GIRDHAR R, RAMANAN D. Attentional pooling for action recognition[DB/OL]. (2017-11-04)[2021-02-15]. https://doi.org/10.48550/arXiv.1711.01467.
[23]	BANSAL A, RAMBHATLA S S, SHRIVASTAVA A, et al. Spatial priming for detecting human-object interactions[DB/OL]. (2020-04-09)[2021-02-15]. https://arxiv.org/abs/2004.04851.
[24]	GKIOXARI G, GIRSHICK R, DOLLÁR P, et al. Detecting and recognizing human-object interactions[DB/OL]. (2017-04-24)[2021-02-22]. https://arxiv.org/abs/1704.07333
[25]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[26]	GUPTA T, SCHWING A, HOIEM D. No-frills human-object interaction detection: factorization, layout encodings, and training techniques[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. [S.l.]: IEEE, 2019: 9677-9685.
[27]	YU F, WANG D, SHELHAMER E, et al. Deep layer aggregation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2018: 2403-2412.
[28]	ZHOU X Y, WANG D Q, KRÄHENBÜHL P. Objects as points[DB/OL]. (2019-04-16)[2021-02-15]. http://arxiv.org/abs/1904.07850.
[29]	LAW H, DENG J. Cornernet: detecting objects as paired keypoints[C]//Proceedings of the European Conference on Computer Vision. [S.l.]: Springer, 2018: 734-750.
[30]	NEWELL A, YANG K, DENG J. Stacked hourglass networks for human pose estimation[C]//European Conference on Computer Vision. [S.l.]: Springer, 2016: 483-499.
[31]	LI Y L, ZHOU S, HUANG X, et al. Transferable interactiveness knowledge for human-object interaction detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2019: 3585-3594.
[32]	LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]//European Conference on Computer Vision. Cham: Springer, 2014: 740-755
[33]	LI J, WANG C, ZHU H, et al. Crowdpose: efficient crowded scenes pose estimation and a new benchmark[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2019: 10863-10872.
[34]	WAN B, ZHOU D, LIU Y, et al. Pose-aware multi-level feature network for human object interaction detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. [S.l.]: IEEE, 2019: 9469-9478.
[35]	CHEN Y, WANG Z, PENG Y, et al. Cascaded pyramid network for multi-person pose estimation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2018: 7103-7112.
[36]	LIANG Z J, LIU J F, GUAN Y S, et al. Pose-based modular network for human-object interaction detection[DB/OL]. (2020-08-05)[2021-02-22]. https://arxiv.org/abs/2008.02042
[37]	LIANG Z J, LIU J F, GUAN Y S, et al. Visual-semantic graph attention networks for human-object interaction detection[DB/OL]. (2020-01-07)[2021-02-22]. https://arxiv.org/abs/2001.02302
[38]	FANG H S, CAO J, TAI Y W, et al. Pairwise body-part attention for recognizing human-object interactions[C]//Proceedings of the European Conference on Computer Vision. [S.l.]: Springer, 2018: 51-67.
[39]	FANG H S, XIE S, TAI Y W, et al. Rmpe: regional multi-person pose estimation[C]//Proceedings of the IEEE International Conference on Computer Vision. [S.l.]: IEEE, 2017: 2334-2343.
[40]	MALLYA A, LAZEBNIK. Learning models for actions and person-object interactions with transfer to question answering[C]//Proceedings of the European Conference on Computer Vision. [S.l.]: Springer, 2016: 414-428.
[41]	ZHOU P, CHI M. Relation parsing neural network for human-object interaction detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. [S.l.]: IEEE, 2019: 843-851.
[42]	GIRSHICK R, RADOSAVOVIC I, GKIOXARI G, et al.Detectron[CP/OL]. (2020-09-22)[2021-02-11]. https://github.com/facebookresearch/detectron.
[43]	HE K, GKIOXARI G, DOLLÁR P, et al. Mask R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. [S.l.]: IEEE, 2017: 2961-2969.
[44]	QI S, WANG W, JIA B, et al. Learning human-object interactions by graph parsing neural networks[C]//Proceedings of the European Conference on Computer Vision. [S.l.]: Springer, 2018: 401-417.
[45]	LIU H C, MU T J, HUANG X L. Detecting human-object interaction with multi-level pairwise feature network[J]. Computational Visual Media, 2021, 7(2): 229-239. doi: 10.1007/s41095-020-0188-2
[46]	ZHONG X, QU X, DING C, et al. Glance and gaze: inferring action-aware points for one-stage human-object interaction detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2021: 13234-13243.
[47]	LAMPERT C H, NICKISCH H, HARMELING S. Learning to detect unseen object classes by between-class attribute transfer[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR). [S.l.]: IEEE, 2009: 951-958.
[48]	SHEN L, YEUNG S, HOFFMAN J, et al. Scaling human-object interaction recognition through zero-shot learning[C]//2018 IEEE Winter Conference on Applications of Computer Vision. [S.l.]: IEEE, 2018: 1568-1576.
[49]	EUM S, KWON H. Semantics to space (S2S): embedding semantics into spatial space for zero-shot verb-object query inferencing[DB/OL]. (2019-06-13)[2022-02-22]. https://arxiv.org/abs/1906.05894
[50]	RAHMAN S, KHAN S, PORIKLI F. Zero-shot object detection: learning to simultaneously recognize and localize novel concepts[DB/OL]. (2018-03-16)[2021-02-22]. https://arxiv.org/abs/1803.06049
[51]	PEYRE J, LAPTEV I, SCHMID C, et al. Detecting unseen visual relations using analogies[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. [S.l.]: IEEE, 2019: 1981-1990.
[52]	ALESSANDRO P, SCHMID C, FERRARI V. Weakly supervised learning of interactions between humans and objects[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 34(3): 601-614.
[53]	PEYRE J, LAPTEV I, SCHMID C, et al. Weakly-supervised learning of visual relations[DB/OL]. (2017-07-29)[2021-02-22]. https://arxiv.org/abs/1707.09472.
[54]	SARULLO A, MU T T. Zero-shot human-object interaction recognition via affordance graphs[DB/OL]. (2020-09-02)[2021-02-22]. https://arxiv.org/abs/2009. 01039.
[55]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[DB/OL]. (2017-06-12)[2022-02-26]. https://doi.org/10.48550/arXiv.1706.03762
[56]	KIM B, LEE J, KANG J, et al. HOTR: end-to-end human-object interaction detection with transfor- mers[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2021: 74-83.
[57]	TAMURA M, OHASHI H, YOSHINAGA T. QPIC: query-based pairwise human-object interaction detection with image-wide contextual information[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE, 2021: 10410-10419.