How to generate a point cloud?

Using the setup above, I have the hough equations of the laser-lines on both sides of the plane. I have even filtered out the pixel points on the object. I also have extrinsic parameters of my camera taken from the checkerboard patterns I put on both sides of the p开发者_开发知识库lane (so I know {R1,T1} and {R2,T2})

Now I want to generate a point-cloud of the object. I see that I need to intersect the laser plane (Elaser) with viewing ray (r) from the camera to get a single 3D point (p), any clues on how can I do this?

---

UPDATE:

This is what I think I have to do, somebody may correct me if I am wrong

STEP1: Transform all points from (left plane) coordinate system to (right plane) coordinate system to get a single {R,T} from {R1, T1} and {R2, T2}

STEP2: Find the equation of laser plane using 2 laser points on left-plane and 1 laser point on right-plane using the same coordinate system (transformed to in STEP1)

STEP3: Find the 3D point on object using equation of laser plane and 2D point on object

---

UPDATE

This is what I have currently -

---

If I am not mistaken (I've never dealt with structured light), you need i) to recover the epipolar constraint between the camera and the laser, ii) determine the epipolar line for each point from the imaged laser stripe, iii) intersect this line with the known position of the laser stripe in laser coordinates, and iv) recover the 3D information by triangulation.

For best results, you should orientate the laser stripe perpendicular to the translation vector between the camera and the laser because this way the intersection in iii) above is well conditioned and is consequently most confidently recovered in presence of noise.

---

Everything about ssegvic's post is correct but this can be easier said then done. Similar to stereo camera calibration you will need to calibrate between the laser and the camera. This is done in practice by performing measurements on an object with a known geometry relative to reference plane and then solving for the mapping between the image coordinates to the object coordinates. Once you have performed this calibration then you can solve for the depth using the resulting calibration matrix.

There are many possible calibration objects that you can use. I have used a plane that is black with thin white lines perpendicular to the plane of your laser. This will give you several points in the image that you can isolate using image processing. Once you have gathered all of the points you need by translating the position of the laser-camera apparatus. You can then translate in a certain direction(towards the camera?) while keeping track of the change in depth of the plane, since these represent the real world coordinates that you need to map to from the image coordinates. You should calibrate over the depth or operating range of the object that you plan on measuring (avoid extrapolation) in reasonable fine increments to your choosing.