This sentence makes no sense in isolation, you are going to confuse people.
Yes its a hypothesis
And that's why I asked how you were able to determine the clouds angle, mis match with the planes artificial horizon was due to camera rotation and NOT as a function of elevation, the simplest explanation is, when we track the motion of the clouds, we still get what looks like pod elevation change.
If that wasn't considered as an option I am not surprised, as the angle of the clouds being a function of pod elevation change doesn't appear to of been considered in this instance either.
You might personally not see steps in those stabilized videos, but the algorithm does. The following compares jet roll, glare angle, Zaine's total rotation numbers, and pod horizon in the current model. The graph shows the change in these values relative to the first frame.
#added after getting the data in https://github.com/0x1beef/uap_nb/blob/main/src/gimbal_adjust_clouds.ipynb
from scipy.spatial.transform import Rotation
import math
def normalize(v):
return v / np.linalg.norm(v)
# rotate a vector by a certain angle around an axis.
def rotate(vec, axis, angle_degrees):
angle = math.radians(angle_degrees)
rot = Rotation.from_rotvec(angle * normalize(axis))
return rot.apply(vec)
# get the angle between vectors 'a' and 'b'.
# the sign is relative to the vector 'c' which is orthogonal to 'a' and 'b'.
def signed_angle(a, b, c):
# using https://stackoverflow.com/a/33920320
return math.degrees(math.atan2(np.dot(np.cross(a, b), c), np.dot(a, b)))
def apply_jet_roll_pitch(vec, jet_roll, jet_pitch):
vec = rotate(vec, [0, 0, 1], -jet_roll)
vec = rotate(vec, [1, 0, 0], jet_pitch)
return vec
# get the angle of the horizon in the pod's eye view without dero
def get_pod_horizon(jet_roll, jet_pitch, pod_pitch, pod_roll):
# a vector point forward along the jet's boreline:
jet_forward = apply_jet_roll_pitch([0, 0, -1], jet_roll, jet_pitch)
# a vector pointing right in the jet's wing plane:
jet_right = apply_jet_roll_pitch([1, 0, 0], jet_roll, jet_pitch)
# the pod's horizon: a vector initially pointing towards jet right, rotated by pod roll
pod_right = rotate(jet_right, -jet_forward, -pod_roll)
# a vector pointing at the target, rotated according to pod roll/pitch:
pod_forward = rotate(jet_forward, pod_right, -pod_pitch)
# pod_forward,the global 'az' viewing angle and a vector pointing up are coplanar.
# the global horizon is a vector pointing right, orthogonal to that plane.
right = np.cross(pod_forward, [0, 1, 0])
# a signed angle between the global horizon and the pod horizon:
return signed_angle(right, pod_right, pod_forward)
import matplotlib.pyplot as plt
def plot_diff(series, title):
abs(series[0:650] - series[0]).plot(label = title)
plot_diff(object_data.jet_roll, 'jet roll')
df = common.gimbal_fix_wh_to_bh(object_data, ['glare_angle'])
plot_diff(df.glare_angle, 'glare angle')
# frustum_roll_df is a dataframe containing Zaine's spreadsheet data
plot_diff(frustum_roll_df['Total rotation'], 'total rotation with "frustum roll"')
def pod_horizon_for_frame(d):
return get_pod_horizon(d.jet_roll, d.jet_pitch, d.pod_pitch, d.pod_roll_glare)
plot_diff(df.apply(pod_horizon_for_frame, axis=1), 'pod horizon')
plt.legend()
plt.ylabel('degrees')
plt.xlabel('frames')
plt.plot()You include a different set of assumptions than what the model in Sitrec is making. Those assumptions were advertised from the beginning as producing different results, and there's no consensus with regards to them being correct. So technically you're not arguing against the model in Sitrec. You're providing your own different model and testing its predictions. So it needs to be highlighted that your results might just be showing that your model is wrong while Sitrec could be right.
The intent was the opposite of a jab. I was actually trying not to be too discouraging with my remarks. I said some contributions can still be useful even if they don't always meet that high bar, and I meant in terms of transparency and reproducibility. Of course transparency and reproducibility do not, on their own, prove that something is correct, but they help others quickly review your argument, which is particularly important when people don't have a lot of time to spend on this. You asked what more you can do to show what you mean. I just answered the question.
I do appreciate your input lbf, I am acutely aware that there is a real person on the other side of the keyboard, thank you for coming back into the conversation.
I had sent the code through to @Mick West for review,
If glare, and dero explains clouds misalignment with artificial horizon, your glare angle curve (orange) should show a progressive CW rotation of about 8°, like the clouds. And not a 4° step during the 1st bank, followed by what looks like 1° rotation.
That's where I'm at with this. The old Gimbal code is too crufty to do much more with, so I'd have to re-implement it as a modern custom sitch. This is something I'll do eventually, but it's not a high priority, as I think the glare case is perfectly sound for several reasons, and this quibble over the early angle isn't likely to be consequential. All the claims being made are very unclear, and I'm not willing to waste days of my time on something so specific that nobody actually fully understands.
It's just your 1st observable, pillar of the glare theory. On which you patched the dero refinement as a quick fix, that doesn't hold water.
No, it's not. As I explained in the very first post, that observable remains. The clouds rotate with banking when the glare does not. That's it. The quibble is about the difference between the cloud motion angle (i.e. the horizon) and the artificial horizon.
Those are two different things.
Which involves dero to explain it, dero that should rotate the entire image, i.e. the glare too (slow realignment with the clouds).
What does glare have to do with it?
Oh, it seems I misread you. "If glare", there, means: "if the shape and the rotation of the apparent obect is a result of glare"?
Which is really talking about the first observable. Anyone can see that scrubbing through the 8s and 18s banks that the horizon rotates a lot more than the glare, which does not seem to rotate much, if at all. That would not happen with a real object. The 0s glare rotation on the graph seems more like an artifact of the intial jittery state of the video , and changing shape of the glare with increase in gain.
Presumably, the same something you are missing. Your explanation seems to rely on some rather specific clouds that curve upwards when stitched. If I were you, I'd wait until you make a full 3D model of what you think is going on.
Have already addressed this.
Indeed, which is partly why I have no great urgency to get back into it, we are butting up against subjective quantification.
This strikes me as odd, namely as your Gimbal a new analysis video was solely directed at demonstrating glare, I believe the quote is
When the footage is in the correct, up is up global orientation, we can see more easily the orientation changes.
What does this have to do with anything? What can you easily see here?When we test for background motion using the Formula we would expect to see a non linear progression, easiest to see via bellingcats stitch for Mosul Sphere.
Sorry, I've got no idea what you are trying to say with the Bellingcat example. It's a video shot from an MQ9 of a windblown balloon, with manual following, so the field of view moves around in various ways.
