Combing RGB and Depth Map Features for human activity recognition

By Yang Zhao, Zicheng Liu, Lu Yang, Hong Cheng. 2012

99 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Jun Liu
2. Amir Shahroudy
3. G. Wang
4. Zehua Sun
5. Hossein Rahmani
6. Qiuhong Ke
7. Pichao Wang
8. Jianfang Hu

Citations in the corpus

• Deep Multimodal Feature Analysis for Action Recognition in RGB+D Videos (2016, methodology): “Multimodality analysis of RGB+D action videos is studied by [39, 34, 59, 20, 22, 87, 88, 31, 63, 23, 55, 34].”
• Depth Induced Feature Representation for 4D Human Activity Recognition (2014, background): “Microsoft Kinect [30]) has drawn much attention on human activity recognition from RGBD data [30], [26], [18], [33], [35], [41], [43], [42].”
• Jointly Learning Heterogeneous Features for RGB-D Activity Recognition (2015, background): “Recent works also showed that the fusion of the RGB and depth sequences can largely improve the recognition of activities with object interactions [5], [15], [16], [20], [30], [41], [51], [55].”
• Fusion of Color and Depth Video for Human Behavior Recognition in an Assistive Environment (2013, background): “[10] addressed that issue and in their work they investigated performances of different ways of extracting interest points, since interest point based approaches can handle cluttered background and partial occlusions.”
• 3D trajectories for action recognition (2014, methodology): “To improve discriminating power similarly to [8] we compute Local Depth Pattern in the neighborhood of detected SURF points.”
• Fusing Spatiotemporal Features and Joints for 3D Action Recognition (2013, background): “Zhao et al. [31] proposed to adapt the STIPs detected on RGB data to the depth data and combine the two channels for action recognition.”
• An efficient part-based approach to action recognition from RGB-D video with BoW-pyramid representation (2013, unknown): “[27] 89.”
• Multi-modal feature fusion for action recognition in RGB-D sequences (2014, methodology): “The simplest fusion method is to concatenate heterogeneous features from different sources, like [2] which binded the skeletal and silhouette-based features, [3] who applied the same method for RGB+D “combing” or [4] in which proposed depth-based comparative coding descriptors were concatenated into”
• 3D Flow Estimation for Human Action Recognition from Colored Point Clouds (2013, background): “…2012; Thuc et al., 2012) and active matricial triangulation systems (i.e.: Kinect-style cameras) (Sung et al., 2011, 2012; Yang & Tian, 2012; Zhang & Parker, 2011; Ni et al., 2011; Popa et al., 2011; Lei et al., 2012; Bloom et al., 2012; Zhao et al., 2012; Ming et al., 2012; Xia et al., 2012).”
• Human activity recognition using multi-features and multiple kernel learning (2014, methodology): “[29] extract the feature vector of each video clip by combining the RGB-based descriptor and depth-map based descriptor.”
• RGB-Depth feature for 3D human activity recognition (2013, background): “, Microsoft Kinect [2]) has drawn much attention on human activity recognition with RGBD data [2-10].”
• Evaluating spatiotemporal interest point features for depth-based action recognition (2014, background): “[25] explored the combination of RGB channel and depth for action recognition.”
• A Review on Human Activity Recognition Using Vision-Based Method (2017, background): “Zhao et al. [77] proposed a framework of combing RGB and depth map features for HAR and presented an optimal scheme.”
• Fusing Multiple Features for Depth-Based Action Recognition (2015, methodology): “There are also works using both the RGB videos and depth maps for action recognition [Sung et al. 2012; Ni et al. 2011; Zhao et al. 2012].”
• Human Action Recognition From Various Data Modalities: A Review (2020, methodology): “[355] 2012 RGB,D None Early - - - - RGGP [356] 2013 RGB,D None Early - 85.”
• Action recognition using ensemble weighted multi-instance learning (2014, background): “…layer unit is pi(x t;W,V )= √∑m k=1 Vik( ∑n j=1Wkjx t j) 2 (1) ISA learns parameters W through finding sparse feature representations in the second layer by solving min W ∑T t=1 ∑m i=1 pi(x t;W,V ) s.t.WWT = I (2) Here, W ∈ Rk×n is the weight connecting the input units to the first layer units.”