Depth Pooling Based Large-Scale 3-D Action Recognition With Convolutional Neural Networks

• Published in: IEEE transactions on multimedia Vol. 20; no. 5; pp. 1051 - 1061
• Main Authors: Wang, Pichao, Li, Wanqing, Gao, Zhimin, Tang, Chang, Ogunbona, Philip O.
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.05.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: action recognition; Artificial neural networks; convol-utional neural networks; Datasets; depth; Dynamics; Feature extraction; Gesture recognition; Image recognition; Image segmentation; Large-scale; Motion segmentation; Moving object recognition; Neural networks; Object recognition; Representations; Three-dimensional displays
• ISSN: 1520-9210; 1941-0077
• DOI: 10.1109/TMM.2018.2818329

This paper proposes three simple, compact yet effective representations of depth sequences, referred to respectively as dynamic depth images (DDI), dynamic depth normal images (DDNI), and dynamic depth motion normal images (DDMNI), for both isolated and continuous action recognition. These dynamic images are constructed from a segmented sequence of depth maps using hierarchical bidirectional rank pooling to effectively capture the spatial-temporal information. Specifically, DDI exploits the dynamics of postures over time, and DDNI and DDMNI exploit the 3-D structural information captured by depth maps. Upon the proposed representations, a convolutional neural network (ConvNet)-based method is developed for action recognition. The image-based representations enable us to fine-tune the existing ConvNet models trained on image data without training a large number of parameters from scratch. The proposed method achieved the state-of-art results on three large datasets, namely, the large-scale continuous gesture recognition dataset (means the Jaccard index 0.4109), the large-scale isolated gesture recognition dataset (<inline-formula> <tex-math notation="LaTeX">\text{59.21}\%</tex-math></inline-formula>), and the NTU RGB+D dataset (<inline-formula> <tex-math notation="LaTeX">\text{87.08}\%</tex-math></inline-formula> cross-subject and <inline-formula> <tex-math notation="LaTeX">\text{84.22}\%</tex-math></inline-formula> cross-view) even though only the depth modality was used.