Author Affiliations
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming , Yunnan 650500, China2Yunnan Key Laboratory of Artificial Intelligence, Kunming University of Science and Technology, Kunming , Yunnan 650500, China3Yunnan Key Laboratory of Computer Technology Application, Kunming University of Science and Technology, Kunming , Yunnan 650500, Chinashow less
Fig. 1. Video acquisition system
Fig. 2. Images of quadruped walking
Fig. 3. Flow chart of automatic identification of motion feature
Fig. 4. Structure of DeeplabV3+
Fig. 5. Multiscale feature fusion module
Fig. 6. DUpsampling
Fig. 7. Improved DeeplabV3+
Fig. 8. Comparison of effects of semantic segmentation.(a) Original image; (b) PSPNet; (c) Segnet; (d) Unet; (e) DeeplabV3+; (f) improved DeeplabV3+Xception
Fig. 9. Flow chart of motion corner detection and matching
Fig. 10. Schematic diagram of corner swing angle
Fig. 11. Local maximum values
Fig. 12. Corner detection of quadrupeds. (a) Corners of rhino without overlap; (b) forelimb corners of rhino with overlap; (c) hindlimb corners of buffalo with overlap; (d) forelimb corners of alpaca with overlap
Fig. 13. Curves of corner distance variation
Fig. 14. Gait cycle of each limb
Fig. 15. Motion corner detection and matching
Fig. 16. Rhino’s corner distance variation curves
Fig. 17. Rhino’s gait cycle of each limb
Fig. 18. Buffalo’s corner distance variation curves
Fig. 19. Buffalo’s gait cycle of each limb
Fig. 20. Alpaca’s corner distance variation curves
Fig. 21. Alpaca’s gait cycle of each limb
Fig. 22. Gait sequences. (a) Gait sequence of buffalo; (b) gait sequence of rhino; (c) gait sequence of alpaca
Fig. 23. Corner swing angles. (a) Corner swing angle of rhino; (b) corner swing angle of buffalo; (c) corner swing angle of alpaca
Fig. 24. Corner detection for different scales. (a) R=1.06%; (b) R=3.25%; (c) R=6.24%
Type | DeeplabV3+ | | Improved DeeplabV3+ |
---|
Accuracy /% | Deviation value /pixel | | Accuracy /% | Deviation value /pixel |
---|
Rhino | 80.2 | 38 | | 84.6 | 32 | Buffalo | 82.4 | 36 | | 85.9 | 27 | Alpaca | 93.7 | 28 | | 100.0 | 19 |
|
Table 1. Comparison of corner detection accuracy
Type | Limb | Rhino | | Buffalo | | Alpaca |
---|
T /s | f /Hz | | T /s | f /Hz | | T /s | f /Hz |
---|
Method of this article | Forelimb 1 | 2.00 | 0.50 | | 1.37 | 0.73 | | 1.28 | 0.78 | Forelimb 2 | 2.33 | 0.43 | | 1.39 | 0.72 | | 1.28 | 0.78 | Hindlimb 1 | 2.10 | 0.48 | | 1.38 | 0.72 | | 1.28 | 0.78 | Hindlimb 2 | 1.87 | 0.53 | | 1.39 | 0.72 | | 1.28 | 0.78 | | Average value | 2.08 | 0.49 | | 1.38 | 0.72 | | 1.28 | 0.78 | Manual calculation | Forelimb 1 | 1.97 | 0.51 | | 1.38 | 0.72 | | 1.26 | 0.79 | Forelimb 2 | 2.13 | 0.47 | | 1.39 | 0.72 | | 1.27 | 0.79 | Hindlimb 1 | 2.07 | 0.48 | | 1.36 | 0.74 | | 1.30 | 0.77 | Hindlimb 2 | 1.93 | 0.52 | | 1.38 | 0.72 | | 1.21 | 0.82 | | Average value | 2.07 | 0.50 | | 1.38 | 0.73 | | 1.26 | 0.79 | Error /% | 0.50 | 2.00 | | 0.00 | 1.37 | | 1.56 | 1.27 |
|
Table 2. Values of cycle and frequency
Limb | Rhino | | Buffalo | | Alpaca |
---|
Method of this article | Manual calculation | | Method of this article | Manual calculation | | Method of this article | Manual calculation |
---|
Forelimb 1 | 1.43 | 1.35 | | 1.15 | 1.12 | | 0.55 | 0.57 | Forelimb 2 | 1.33 | 1.22 | | 1.05 | 1.10 | | 0.52 | 0.50 | Hindlimb 1 | 1.40 | 1.45 | | 1.10 | 1.15 | | 0.50 | 0.51 | Hindlimb 2 | 1.44 | 1.38 | | 1.13 | 1.17 | | 0.52 | 0.54 | Average value | 1.40 | 1.36 | | 1.11 | 1.14 | | 0.52 | 0.53 | Average error /% | 2.85 | | 2.63 | | 1.89 |
|
Table 3. Comparison of stride lengths of rhino, buffalo, and alpaca