Table 1 Comparision of attitude estimation methods used in test of wind tunnel.
| Â | Author | Principles of vision | targets | methods | Test environment | accuracy |
|---|---|---|---|---|---|---|
Non-real-time | Ruyten35 | Trinocular vision | Fluorescent | Simplex iteration | Transonic | 0.005°the root-mean-square (RMS) pitch and yaw |
Chen36 | Monocular vision | Natural reflection | Generalized inverse least square | Low-sonic | Â | |
Spain37 | Monocular vision | Natural reflection | Â | Except for the supersonic | Â | |
Leifer38 | Trinocular vision | Retro-reflective | Seven-point numerical differentiation | Â | X, Y, Z-axis position accuracy is 0.0079 mm, 0.0144 mm, and 0.0186 mm rms respectively | |
Real-time | Thomas39 | Binocular vision | Retro-reflective | Least square | Supersonic | Figure 12a shows the maximum angle error is 0.3°, the average error of pitch Angle is 0.0404, and the prediction interval is 95% |
Liu40 | Trinocular depth vision | AprilTag | ORB and Bundle Adjustment Algorithm | Low-sonic | Figure 12b shows the angle and displacement are consistent with the sensor | |
Zhen41 | Binocular vision | Retro-reflective | Case Deletion Diagnostics and least square | Supersonic | Position accuracy is 0.16mm, pitch, and yaw angle accuracy is less than 0.132° and roll angle accuracy is 0.712° | |
Liu42 | Binocular vision | Retro-reflective | Centroid method with three-dimensional constraints | Supersonic | Rms errors of 0.182 mm, 0.158°, 0.212°, 0.9° for displacement, pitch, yaw and roll angles | |
Fan43 | Multi-ocular vision | Natural reflection | least square | Low-sonic | Static position accuracy is better than 0.07mm, attitude angle measurement accuracy is up to 0.05°, and dynamic measurement accuracy is 0.5° | |
Deep-learning | Huang45 | Monocular vision | Natural reflection | CNN + GAN | Low-sonic | Figure 12c shows the acceleration are consistent with the sensor |
