Mater Thesis Presentation, Albin Ohlsson
Kontakt: carl [dot] olsson [at] maths [dot] lth [dot] se
Spara händelsen till din kalender
Bundle-Adjustment with Visual- and IMU-data using B-Splines
The Structure from Motion problem is well known in computer vision. However, the scale of the trajectory cannot be determined from just image data if only a single camera is used. But in many applications, like measuring, getting the reconstruction to scale is of high importance.
This work attempts to use IMU data together with visual data to get a reconstruction with scale. This is done by approximating the camera trajectory with two types of B-splines; one regular B-spline curve for the translation and one quaternion B-spline curve for the rotation, both of order 4. These are twice differentiable, and can thus be used for comparison with IMU data.
Several steps of pre-processing are done before the visual- and inertial-data can be combined. An initial spline is constructed from discrete cameras, then the resolution in time is increased. Biases of gyroscope and accelerometer as well as gravity are estimated and the splines are fitted to the IMU data as an initialization for the final optimization where the visual and inertial data is combined.
The results shows that the different types of data can be combined, but estimating scale is difficult for short sequences.