port over to OpenCV?

[oneshell]

The live demo is fantastic, thanks for making this available.

I am trying to understand the math/algorithm going on behind the scenes. I want to calibrate a Raspberry Pi wide FOV camera using your approach, then use these lens parameters in an OpenCV loop to dewarp a video stream.

Your live demo does everything I want in terms of finding the parameters, now I'm struggling with how to apply this to a video feed in OpenCV.

Could you point me in the right direction?

Thanks,

[jywarren]

Hi, @oneshell - sorry, saw your email but was a bit busy yesterday. Glad to answer here!

The issue is that we are doing this via WebGL, so the conversion is done via a shader. We feed the variables in here:

fisheyegl/src/fisheyegl.js

Lines 257 to 259 in 89d28f0

	gl.uniform3fv(uLensS, [lens.a, lens.b, lens.scale]);
	gl.uniform2fv(uLensF, [lens.Fx, lens.Fy]);
	gl.uniform2fv(uFov, [fov.x, fov.y]);

These are applied in the shaders:

fisheyegl/src/shaders/fragment.glfs.js

Lines 14 to 25 in 89d28f0

	float scale = uLens.w;\n\
	float F = uLens.z;\n\
	\n\
	float L = length(vec3(vPosition.xy/scale, F));\n\
	vec2 vMapping = vPosition.xy * F / L;\n\
	vMapping = vMapping * uLens.xy;\n\
	vMapping = GLCoord2TextureCoord(vMapping/scale);\n\
	vec4 texture = texture2D(uSampler, vMapping);\n\
	if(vMapping.x > 0.99 || vMapping.x < 0.01 || vMapping.y > 0.99 || vMapping.y < 0.01){\n\
	texture = vec4(0.0, 0.0, 0.0, 1.0);\n\
	} \n\
	gl_FragColor = texture;\n\

However, I believe they are determining the geometry of a GL surface, rather than doing this per pixel (in a GL paradigm rather than a pixel mapper). GL isn't my strong suit, and I am not very familiar with OpenCV, but here are some resources that may help you use the parameters in OpenCV:

• https://hackaday.io/project/12384-autofan-automated-control-of-air-flow/log/41862-correcting-for-lens-distortions
• https://stackoverflow.com/questions/26602981/correct-barrel-distortion-in-opencv-manually-without-chessboard-image
• https://answers.opencv.org/question/6332/lens-distortion-equation-changed-reference-please/

I'm not sure if they use the same model, but perhaps there is some way to convert and use the parameters in whatever model they are using?

Sorry I'm not more helpful!

[guilhermecgs]

same problem here...

I would love to have a port to open-cv