论文集--借鉴别人的特征

Online RGB-D Gesture Recognition with Extreme Learning Machines

归一化两关节点间的距离

image.png
image.png

Sequence of the Most Informative Joints (SMIJ): A New Representation for Human Skeletal Action Recognition

image.png
image.png
image.png

Posebits for Monocular Human Pose Estimation

image.png

《Skeleton Based Action Recognition Using Translation-Scale Invariant Image Mapping And
Multi-Scale Deep CNN》
身体分为五部分，骨骼点转化为图片。
NTU - RGBD CS 85.02% CV 92.3%
UTD - MHAD 96.27% ( captured by one Microsoft Kinect camera and one wearable inertial sensor )
MSRC - 12 99.41 %
G3D DATASET 93.9%

《2 streams 3D cnn for human skeleton based action recognition》
CS 66.85% CV 72.58%

《Skeleton-Based Action Recognition Using Spatio-Temporal LSTM Network with Trust Gates》

image.png

NTU -RGBD

SKELETON-BASED ACTION RECOGNITION USING LSTM AND CNN
相对距离 or 关节点N 到 关节点 j、k之间线段JK的距离
《Mining Key Skeleton Poses with Latent SVM for Action Recognition 》
关键帧的选取 和 成对的相对距离

《skeleton based human action recognition with profile hidden markov models》

相对距离和 角度

《Co-occurrence Feature Learning for Skel..》
直接输入

《HIF3D: Handwriting-Inspired Features for 3D skeleton-based action recognition》

image.png

《A New Representation of Skeleton Sequences for 3D Action Recognition》

image.png
image.png

绿色点为参考点，算相对距离。