• Laser & Optoelectronics Progress
  • Vol. 55, Issue 2, 021002 (2018)
Haiyang Xu1, Jun Kong1、2、*, and Min Jiang1
Author Affiliations
  • 1 School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • 2 College of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830047, China
  • show less
    DOI: 10.3788/LOP55.021002 Cite this Article Set citation alerts
    Haiyang Xu, Jun Kong, Min Jiang. Human Action Recognition Based on Quaternion 3D Skeleton Representation[J]. Laser & Optoelectronics Progress, 2018, 55(2): 021002 Copy Citation Text show less
    Flow chart of training and testing process
    Fig. 1. Flow chart of training and testing process
    Hierarchical skeleton points and definition of skeletal vectors
    Fig. 2. Hierarchical skeleton points and definition of skeletal vectors
    (a) Representation of body part em in local coordinate system of en; (b) Representation of body part en in local coordinate system of em
    Fig. 3. (a) Representation of body part em in local coordinate system of en; (b) Representation of body part en in local coordinate system of em
    Angle cosine calculation between body parts
    Fig. 4. Angle cosine calculation between body parts
    Confusion matrix of MSR-Action3D database
    Fig. 5. Confusion matrix of MSR-Action3D database
    Confusion matrix of Florence3D-Action database
    Fig. 6. Confusion matrix of Florence3D-Action database
    MethodRecognition rate /%
    Ref. [22]91.21*
    Ref. [10]92.03*
    Ref. [10]92.20*
    Ref. [17]92.49*
    Proposed95.17*
    Ref. [16]88.20
    Ref. [23]90.00
    Ref. [17]89.48
    Proposed91.88
    Table 1. Comparison of different descriptors different descriptors on MSR-Action3D dataset
    MethodRecognition rate /%
    Quaternion withoutclassification of key frames94.12
    Quaternion withclassification of key frames95.17
    Table 2. Recognition rate based on key frames on MSR-Action3D dataset
    DatasetMethodRecognition rate /%
    UTKinect-ActiondatasetRef. [22]88.50
    Ref. [24]94.50
    Ref. [17]97.08
    Proposed97.49
    Florence3D-Action datasetRef. [20]82.00
    Ref. [25]87.00
    Ref. [17]90.88
    Proposed92.14
    Table 3. Comparison of different descriptors on database of UTKinect-Action and Florence3D-Action
    Haiyang Xu, Jun Kong, Min Jiang. Human Action Recognition Based on Quaternion 3D Skeleton Representation[J]. Laser & Optoelectronics Progress, 2018, 55(2): 021002
    Download Citation