# hand_eye_calibration, surface normal by stereo vision

I'm a graduate student at mechanical engineering. I took computer vision(stereo vision) last semester but I'm still fairly new to this area. I wonder if I may ask 3 questions related to computer vision? We have a 6 DOF robot arm, and we want to orient the end effector(z axis of end effector) to the normal of a curved surface. The geometry of this curved surface is unknown.

1. I wonder if you may suggest some good tutorials/paper/implementations of the extrinsic calibration? Assuming I have a stereo camera(ZED for example), which is fixed in space. I want to get the camera's coordinate and therefore all the pixels coordinates expressed in the robot base frame. I think this might be a hand eye calibration problem, but I think there might be a better methods?

2. May I use stereo camera like ZED to get the surface normal? Assuming I want to get the surface normal of a cylinder. Assuming I have the extrinsic calibrated camera ready, which faces this cylinder. My proposed method is that I mark 3 points a1,a2,a3 in vicinity, and I use a blue marker pen to mark these 3 points. I can get the "physical coordinate" of these three points in the robot base frame, then I took the cross product between the physical coordinate of (a3-a1) and the physical coordinate of (a2-a1), then I may have a normal at the physical coordinate of a1.

3. Any suggestions/improvements on my methods? My biggest concern is that ZED camera may not be a good option here, after all 400 dollar is not cheap for an international student. More importantly, I always feel there must a better option here.

I'm all ears. I assume I may not have a good description of my problem. Thank you for your attention and help in advance.

Best,

• You sound like you are starting a discussion. Stack exchange usually does not work that way. Try something then ask a specific question when you get stuck. If you are short on cash, I would use COTS parts like existing web-cams. That said, if I were really short on cash I would try to solve as many problems in software as I could. So, instead of throwing money at 2 cameras, I would use 1 and structured light. Then use freely available software like OpenCV. Oct 17 '17 at 23:12