Update!
All the files have been relinked due to the recent domain merge by the university.
https://global.vcu.edu/newsroom/2020/email/
Please use the link below:
https://drive.google.com/drive/folders/1yO10ty1IbC9aInYz7wdauiFoKlnjZCsc?usp=sharing
In order to obtain the ground truth trajectory in the IMU/body’s coordinate system, the extrinsic transformation matrix ($T_M^b$) from MoCap to IMU is required. MoCap can detect and track the LED markers which are installed on the SC, shown in figure below.
The transformation matrix from MoCap to the marker’s coordinate system is described as $T_H^M$. Since the camera-IMU extrinsic transformation matrix ($T_c^b$) is known, we need the camera-maker transformation matrix ($T_c^H$) to compute $T_M^b$. $T_c^H$ can be obtained through a hand-eye calibration approach 1. To solve $T_c^H$, we recorded a data sequence by moving the SC around a checkboard (maintaining the checkboard in the SC’s field of view). Then we employed Zhang’s method 2 to compute the camera poses. Given the camera poses and the marker’s poses observed from the MoCap, we used Daniilidis’ method 1 to estimate $T_c^H$ and achieved a final pose residual about 0.6703 mm. Given $T_c^H$ , $T_M^b$ is given by $T_{M}^{b}=T_{c}^{b}\left(T_{c}^{H}\right)^{-1}\left(T_{H}^{M}\right)^{-1}$
Dataset download hand-eye - Google Drive
References:
- [1]: Daniilidis, K. “Hand-eye calibration using dual quaternions,” The International Journal of Robotics Research, 18(3), 286-298.
- [2]: Zhang, Zhengyou. “A flexible new technique for camera calibration.” IEEE Transactions on pattern analysis and machine intelligence 22.11 (2000): 1330-1334.