Hi Steve. I think the polynomial equations for P3P are forth degree which yields 2 behind and 2 in front of the focal point. Please let me know if I am wrong, thanks!
one thingh that makes me confusing is that are we gonna to find the postion of the camera ? then why you try to fix the objects by reducing its DoF? i mean i dont see the connection.
i kinda of understand why he analyses the problem in that way, because camera , image and the objects are in a projection function, if one side is fixed then the other side will be automately determined. That means if the position of the objects are fixed then the position of the camera is alse fixed
I like this course because it's what we are trying to do: build a VR system. But I have a question, normally the points that we observe on the image are not unique. We cannot identify which is which. So how do we solve such problem assuming that all lights are identical? What other algorithms do we need?
@@roboticsinacademia3170 this is not what I meant. In the video there is an assumption that you know the ID of each point. In the real scenarios, you are just given a bunch of points that look the same so you don't know their ID. I want to know how you obtain the ID for each point. He suggests using different frequency like blinking or different color of LEDs but I saw mixed reality controller or oculus quest controller did not do that. All the points look identical.
@@anukritisingh7204 The impression that I have is that you need to identify the ID of points somehow. You can try these: - give them different LED colors. - make them blink with different frequency. - make them blink as a series of ones and zeros to report their own ID. e.g. 1001010 = 74 - given the IMU data of a specific LED and the 2D positions of all the points for a period of T seconds, try to pick the point that best explains the value of the IMU reading. - use machine learning to observe 2D points for a period of T seconds, and then it might be able to classify the ID of each point
Amazing explanation, thank you for the video!
X = V = ⚠️n
Excellent!
Thanks for this video!
I have a question: if I'm working in the IR spectrum, how can I identify the led´s label? (apart from different frequencies)
Hi Steve. I think the polynomial equations for P3P are forth degree which yields 2 behind and 2 in front of the focal point. Please let me know if I am wrong, thanks!
The other three solutions are basically symmetry exploited around the legs of triangle formed by the 3 points
Awesome explanation
one thingh that makes me confusing is that are we gonna to find the postion of the camera ? then why you try to fix the objects by reducing its DoF? i mean i dont see the connection.
i mean throught the clip you dont explain how that could determine the camera postion.
i kinda of understand why he analyses the problem in that way, because camera , image and the objects are in a projection function, if one side is fixed then the other side will be automately determined. That means if the position of the objects are fixed then the position of the camera is alse fixed
You could zoom out so you don't have to pan back and forth so much
unfortunately we will miss the stuff on the board if we zoom out
great video thanks
If you want more related videos from him:
nptel.ac.in/courses/106106138/49
I like this course because it's what we are trying to do: build a VR system.
But I have a question, normally the points that we observe on the image are not unique. We cannot identify which is which. So how do we solve such problem assuming that all lights are identical?
What other algorithms do we need?
you need feature matching algorithms like ORB, or SIFT
@@roboticsinacademia3170 this is not what I meant. In the video there is an assumption that you know the ID of each point. In the real scenarios, you are just given a bunch of points that look the same so you don't know their ID. I want to know how you obtain the ID for each point.
He suggests using different frequency like blinking or different color of LEDs but I saw mixed reality controller or oculus quest controller did not do that. All the points look identical.
hey were you able to figure out an answer for this?@@offchan
@@anukritisingh7204 The impression that I have is that you need to identify the ID of points somehow. You can try these:
- give them different LED colors.
- make them blink with different frequency.
- make them blink as a series of ones and zeros to report their own ID. e.g. 1001010 = 74
- given the IMU data of a specific LED and the 2D positions of all the points for a period of T seconds, try to pick the point that best explains the value of the IMU reading.
- use machine learning to observe 2D points for a period of T seconds, and then it might be able to classify the ID of each point