Name
Abstract | ||
|---|---|---|
There are limited approaches using Kinect for upper body motion recognition. Most of the available approaches are conducted when there is no joint occlusion, though some performed with joint occlusion only demonstrated recognition of a few motions at low recognition rates. This paper utilizes OptiTrack and its supporting software to obtain and transfer data into a human skeleton coordinates using Kinect v2, and selects the vector among key joint points and angles as the feature values; the AP clustering algorithm was adopted for the key frames of motions which are marked; then we conduct relative normalization of the feature values, and use the method of random forest regression to realize two functions: (1) conduct derivation based on joint offset of frames detected with Kinect v2 from those detected with OptiTrack, learn the joint offset regression function, and correct the skeleton based on the predictions on joint offset; (2) determine the motions based on predicted posture. This paper performs recognition of 8 types of upper body motions at an average accuracy of 90.86%. |
Year | DOI | Venue |
|---|---|---|
2020 | 10.1007/s11042-018-6357-y | Multimedia Tools and Applications |
Keywords | Field | DocType |
Upper body motion recognition, Upper limb motion, Kinect V2, AP clustering algorithm, Key frame, Random forest regression | Computer vision,Normalization (statistics),Pattern recognition,Motion recognition,Computer science,Human skeleton,Software,Artificial intelligence,Key frame,Random forest,Cluster analysis,Offset (computer science) | Journal |
Volume | Issue | ISSN |
79 | 7-8 | 1573-7721 |
Citations | PageRank | References |
0 | 0.34 | 14 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Bo Li | 1 | 171 | 67.08 |
| Baoxing Bai | 2 | 2 | 1.04 |
| Cheng Han | 3 | 3 | 3.43 |