Author Affiliations
School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou, Inner Mongolia 014010, Chinashow less
Fig. 1. Pixel gray value of 3×3 window
Fig. 2. Neighborhood of circular symmetry. (a) (1,8); (b) (1.5,12); (c) (2,16)
Fig. 3. Processing results of the two modes. (a) Gray image; (b) original LBP operator; (c) improved LBP operator
Fig. 4. Neighborhood of pixel point
Fig. 5. Adjacent pixel
Fig. 6. Edges extracted by different thresholds. (a) Depth image; (b) automatic threshold; (c) threshold range is [0.32,0.8]; (d) threshold range is [0.08,0.2]
Fig. 7. Specific steps of the ScSPM model
Fig. 8. Flow chart of our algorithm
Fig. 9. Recognition results of the CAD-60 dataset
Fig. 10. Recognition results of the MSR Action Pairs dataset
Fig. 11. Recognition results of the SBU Kinect interaction dataset
Reference | Recognitionmethod | Recognitionrate /% |
---|
Ref. [16] | Full tow-layer MEMM | 61.7 | Ref. [17] | HMM | 82.3 | Ref. [18] | CFt+RBF-SVM | 94.0 | Ours | S+ScSPM | 99.0 |
|
Table 1. Recognition rates of different algorithms (CAD-60 dataset)
Reference | Recognitionmethod | Recognitionrate /% |
---|
Ref. [19] | skeleton+LOPskeleton+LOP+pyramid | 61.782.2 | Ref. [20] | DMM | 66.1 | Ref. [21] | HON4D | 93.3 | Ours | S+ScSPM | 93.1 |
|
Table 2. Recognition rates of different algorithms (MSR Action Pairs dataset)
Reference | Recognitionmethod | Recognitionrate /% |
---|
Ref. [22] | joint features+CFDM | 89.4 | Ref. [23] | SVM+LCNN | 92.8 | Ref. [24] | BOW+HOG | 92.5 | Ref. [25] | motion feature+shape feature(depth) | 98.4 | Ours | S+ScSPM | 95.4 |
|
Table 3. Recognition rates of different algorithms (SBU Kinect interaction dataset)