Why blur the face of the person (I'm guessing it's the author's face) at 3:39, when at 0:38 there's 4 shots showing the (your) face? Really impressive research, nevertheless!
Interesting, could directly integrating the regressed velocities (orange line) also work? (Edit: just read it in the paper, "Direct integration of the predicted velocities would produce positions but performs worse.")
Hi, What would be the most appropriate method for calculating the position of an object using linear acceleration data from a BNO055 sensor, given the potential presence of noise and errors in the data? Additionally, what techniques or methods can be employed to mitigate these issues and improve the accuracy of the position calculation?
What do you think might happen if you used multiple IMUs arranged in such a way that no IMU had parallel/co-planar planes to the others? Would the extra ability to isolate noise by calculating the virtual IMU help clean up the signal even more?
How would you remove noise in your example? Do you have a paper that details this technique that you could recommend? I'm using IMUs for pedestrian tracking and haven't come across two (or more) being used in this way. Interested to hear what you have to say!
various white noise, random walk noise and bias. once integrate them together, you amplify the noise. See github.com/ethz-asl/kalibr/wiki/IMU-Noise-Model for detail