基于掩码自监督学习的点云动作识别方法

doi:10.3778/j.issn.1673-9418.2404045

摘要/Abstract

摘要： 点云动作识别方法可以提供精准的三维动作监测与识别服务，在智能交互、智能安防和医疗健康等领域具有广阔应用前景。现有方法通常利用大量标注的点云数据训练模型，但点云视频包含大量的三维坐标，精准标注点云非常昂贵，同时点云视频高度冗余，点云信息在视频中分布不均，这些问题都增大了标注的难度。为解决上述问题并获得更好的点云动作识别性能，提出一种无需人工标注即可捕获点云视频时空结构的掩码自监督动作识别方法MSTD-Transformer。将点云视频划分为点管并根据重要性进行自适应视频级掩码，通过点云重构和运动预测双流自监督学习点云视频的外观和运动特征。为了更好地捕获运动信息，MSTD-Transformer从点云关键点的位移中提取动态注意力并嵌入Transformer，使用双分支结构进行差异化学习，分别捕获运动信息和全局结构。在标准数据集MSRAction-3D上的实验结果表明，提出的方法对24帧点云视频动作识别准确率为96.17%，较现有最好方法提高2.09个百分点，证实了掩码策略和动态注意力的有效性。

关键词: 动作识别, 点云, 自监督学习, 掩码, 注意力机制

Abstract: Point cloud action recognition methods can provide precise 3D motion monitoring and recognition services, with broad application prospects in fields such as intelligent interaction, intelligent security, and medical health. Existing methods typically use a large amount of annotated point cloud data to train models, but point cloud videos contain a large number of 3D coordinates, precise annotation of point clouds is very expensive, and point cloud videos are highly redundant with uneven distribution of point cloud information in the video, all of which increase the difficulty of annotation. To address the aforementioned issue and achieve superior performance in point cloud action recognition, a novel masked self-supervised action recognition method called MSTD-Transformer is proposed, which can capture the spatiotemporal structure of point cloud videos without the need for manual annotation. Specifically, the point cloud video is divided into point tubes and adaptive video-level masks are generated based on importance, learning the appearance and motion features of point cloud videos through self-supervised learning of point cloud reconstruction and motion prediction dual-stream. To better capture motion information, MSTD-Transformer extracts dynamic attention from the displacement of point cloud keypoints and embeds it into a Transformer, using a dual-branch structure for differential learning to capture motion information and global structure separately. Experimental results on the standard dataset MSRAction-3D show that the proposed method achieves an accuracy of 96.17% for 24-frame point cloud video action recognition, which is 2.09 percentage points higher than the best existing method, confirming the effectiveness of the masking strategy and dynamic attention.

Key words: action recognition, point cloud, self-supervised learning, mask, attention mechanism

何允栋, 李平, 平晨昊. 基于掩码自监督学习的点云动作识别方法[J]. 计算机科学与探索, 2024, 18(12): 3235-3246.

HE Yundong, LI Ping, PING Chenhao. Point Cloud Action Recognition Method Based on Masked Self-Supervised Learning[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(12): 3235-3246.

参考文献

[1] LIU X Y, YAN M Y, BOHG J. MeteorNet: deep learning on dynamic 3D point cloud sequences[C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2019: 9246-9255.
[2] FAN H H, YU X, DING Y H, et al. Point spatio-temporal convolution on point cloud sequences[C]//Proceedings of the 2021 International Conference on Learning Representations, Vienna, May 4-8, 2021.
[3] WANG Y C, XIAO Y, XIONG F, et al. 3DV: 3D dynamic voxel for action recognition in depth video[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 511-520.
[4] FAN H H, YANG Y, KANKANHALLI M. Point spatio-temporal transformer networks for point cloud video modeling[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(2): 2181-2192.
[5] 田钰琪, 刘康, 张远辉. 基于毫米波雷达点云的人体动作识别[J]. 中国计量大学学报, 2023, 34(1): 66-73.
TIAN Y Q, LIU K, ZHANG Y H. Human activity recognition based on millimeter wave radar point cloud[J]. Journal of China University of Metrology, 2023, 34(1): 66-73.
[6] ZHONG J X, ZHOU K, HU Q Y, et al. No pain, big gain: classify dynamic point cloud sequences with static models by fitting feature-level space-time surfaces[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 8510-8520.
[7] 赵登阁, 智敏. 用于人体动作识别的多尺度时空图卷积算法[J]. 计算机科学与探索, 2023, 17(3): 719-732.
ZHAO D G, ZHI M. Spatial multiple-temporal graph convolutional neural network for human action recognition[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(3): 719-732.
[8] FAN H H, YANG Y, KANKANHALLI M. Point 4D transformer networks for spatio-temporal modeling in point cloud videos[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 14204-14213.
[9] HE K M, CHEN X L, XIE S N, et al. Masked autoencoders are scalable vision learners[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 16000-16009.
[10] FEICHTENHOFER C, FAN H Q, LI Y H, et al. Masked autoencoders as spatiotemporal learners[C]//Advances in Neural Information Processing Systems 35, New Orleans, Nov 28-Dec 9, 2022: 35946-35958.
[11] TONG Z, SONG Y B, WANG J, et al. VideoMAE: masked autoencoders are data-efficient learners for self-supervised video pre-training[C]//Advances in Neural Information Processing Systems 35, New Orleans, Nov 28-Dec 9, 2022: 10078-10093.
[12] YU X M, TANG L L, RAO Y M, et al. Point-BERT: pre-training 3D point cloud transformers with masked point modeling[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 19313-19322.
[13] PANG Y T, WANG W X, TAY F E H, et al. Masked autoencoders for point cloud self-supervised learning[C]//Proceedings of the 17th European Conference on Computer Vision. Cham: Springer, 2022: 604-621.
[14] HWANG S, YOON J, LEE Y, et al. EVEREST: efficient masked video autoencoder by removing redundant spatiotemporal tokens[EB/OL]. [2024-02-23]. https://arxiv.org/abs/ 2211.10636.
[15] SHEN Z Q, SHENG X X, FAN H H, et al. Masked spatio-temporal structure prediction for self-supervised learning on point cloud videos[C]//Proceedings of the 2023 IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE, 2023: 16580-16589.
[16] SHEN Z, SHENG X, WANG L, et al. PointCMP: contrastive mask prediction for self-supervised learning on point cloud videos[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 1212-1222.
[17] 邱云飞, 王宜帆. 双分支结构的多层级三维点云补全[J]. 计算机工程与应用, 2024, 60(9): 272-282.
QIU Y F, WANG Y F. Multi-level 3D point cloud completion with dual-branch structure[J]. Computer Engineering and Applications, 2024, 60(9): 272-282.
[18] 李海旺, 周恒可, 赵兴, 等. 机载LiDAR点云数据的建筑屋顶面提取算法[J]. 计算机工程与应用, 2024, 60(11): 233-241.
LI H W, ZHOU H K, ZHAO X, et al. Algorithm for extracting building roof surfaces from airborne LiDAR point cloud data[J]. Computer Engineering and Applications, 2024, 60(11): 233-241.
[19] LI P, CAO J, YUAN L, et al. Truncated attention-aware proposal networks with multi-scale dilation for temporal action detection[J]. Pattern Recognition, 2023, 142: 109684.
[20] LI P, CAO J, YE X. Prototype contrastive learning for point-supervised temporal action detection[J]. Expert Systems with Applications, 2023, 213: 118965.
[21] DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. [2024-02-23]. https://arxiv.org/abs/2010.11929.
[22] XIE Z D, ZHANG Z, CAO Y, et al. SimMIM: a simple framework for masked image modeling[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 9653-9663.
[23] BAO H B, DONG L, PIAO S H, et al. BEiT: BERT pre-training of image transformers[C]//Proceedings of the 10th International Conference on Learning Representations, Apr 25-29, 2022.
[24] CHEN M, RADFORD A, CHILD R, et al. Generative pretraining from pixels[C]//Proceedings of the 37th International Conference on Machine Learning, Jul 13-18, 2020: 1691-1703.
[25] QI C R, SU H, MO K, et al. PointNet: deep learning on point sets for 3D classification and segmentation[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2017: 652-660.
[26] QI C R, YI L, SU H, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space[C]//Advances in Neural Information Processing Systems 30, Long Beach, Dec 4-9, 2017: 5099-5108.
[27] MATURANA D, SCHERER S. VoxNet: a 3D convolutional neural network for real-time object recognition[C]//Proceedings of the 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems. Piscataway: IEEE, 2015: 922-928.
[28] ZHANG C, WAN H C, SHEN X Y, et al. PVT: point-voxel transformer for point cloud learning[J]. International Journal of Intelligent Systems, 2022, 37(12): 11985-12008.
[29] WANG Y, SUN Y B, LIU Z W, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 1-12.
[30] SHEN Y R, FENG C, YANG Y Q, et al. Mining point cloud local structures by kernel correlation and graph pooling[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2018: 4548-4557.
[31] BEN-SHABAT Y, SHROUT O, GOULD S. 3DinAction: understanding human actions in 3D point clouds[EB/OL].[2024-02-23]. https://arxiv.org/abs/2303.06346.
[32] WANG H Y, YANG L, RONG X J, et al. Self-supervised 4D spatio-temporal feature learning via order prediction of sequential point cloud clips[C]//Proceedings of the 2021 IEEE/CVF Winter Conference on Applications of Computer Vision. Piscataway: IEEE, 2021: 3762-3771.
[33] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances in Neural Information Processing Systems 30, Long Beach, Dec 4-9, 2017: 5998-6008.
[34] QIAN R, DING S R, LIU X, et al. Static and dynamic concepts for self-supervised video representation learning[C]//Proceedings of the 17th European Conference on Computer Vision. Cham: Springer, 2022: 145-164.
[35] SIMONYAN K, ZISSERMAN A. Two-stream convolutional networks for action recognition in videos[C]//Advances in Neural Information Processing Systems 27, Montreal, Dec 8-13, 2014: 568-576.
[36] LI W, ZHANG Z, LIU Z. Action recognition based on a bag of 3D points[C]//Proceedings of the 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2010: 9-14.
[37] FAN H H, YU X, YANG Y, et al. Deep hierarchical representation of point cloud videos via spatio-temporal decomposition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44(12): 9918-9930.
[38] WEN H, LIU Y Z, HUANG J W, et al. Point primitive transformer for long-term 4D point cloud video understanding[C]//Proceedings of the 17th European Conference on Computer Vision. Cham: Springer, 2022: 19-35.
[39] CORTES C, MOHRI M, ROSTAMIZADEH A. Algorithms for learning kernels based on centered alignment[J]. The Journal of Machine Learning Research, 2012, 13(1): 795-828.
[40] LI X, HUANG Q, WANG Z, et al. SequentialPointNet: a strong frame-level parallel point cloud sequence network for 3D action recognition[EB/OL]. [2024-02-23]. https://arxiv.org/abs/2111.08492.