Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Adding example to rectify lens distortion in depth images #5

Open
interactivetech opened this issue Jan 3, 2019 · 5 comments
Open

Comments

@interactivetech
Copy link

Hi there!

Thank you very much for creating this open source repo, I plan to use it for future ios machine learning projects! In the WWDC 2017 Talk, Apple discusses that the depth output is geometrically distorted to align with images produced by the camera. They mention to get precise true depth measurements, you need to correct for lens distortion. the WWDC Talk says that a reference implementation to correct for lens distortion is commented in the AVCameraCalibrationData.h file.

It would be great if you can add an example view controller that a user taps to see a recitfied depth image and enable developers to work with precise, true depth measurements. I attached the reference implementation from AVCameraCalibrationData.h for ease of reference and if anyone can add this it would be amazing!

The following reference implementation illustrates how to use the lensDistortionLookupTable, 
    inverseLensDistortionLookupTable, and lensDistortionCenter properties to find points in the 
    lens-distorted or undistorted (rectilinear, corrected) space. If you have a distorted image (such as a photo taken by a camera) and want to find a particular point in a corresponding undistorted image, you would call the sample method below using the inverseLensDistortionLookupTable. If you have an undistorted (aka distortion-corrected) image and want to find a point in the distorted image's space, you would call the sample method below using the lensDistortionLookupTable.
 
    To apply distortion correction to an image, you'd begin with an empty destination buffer and iterate through it 
    row by row, calling the sample implementation below for each point in the output image, passing the 
    lensDistortionLookupTable to find the corresponding value in the distorted image, and write it to your 
    output buffer. Please note that the "point", "opticalCenter", and "imageSize" parameters below must be
    in the same coordinate system, i.e. both at full resolution, or both scaled to a different resolution but
    with the same aspect ratio.
 
    The reference function below returns floating-point x and y values. If you wish to match the results with 
    actual pixels in a bitmap, you should either round to the nearest integer value or interpolate from surrounding
    integer positions (i.e. bilinear interpolation from the 4 surrounding pixels).
 
- (CGPoint)lensDistortionPointForPoint:(CGPoint)point
                           lookupTable:(NSData *)lookupTable
               distortionOpticalCenter:(CGPoint)opticalCenter
                             imageSize:(CGSize)imageSize
{
    // The lookup table holds the relative radial magnification for n linearly spaced radii.
    // The first position corresponds to radius = 0
    // The last position corresponds to the largest radius found in the image.
 
    // Determine the maximum radius.
    float delta_ocx_max = MAX( opticalCenter.x, imageSize.width  - opticalCenter.x );
    float delta_ocy_max = MAX( opticalCenter.y, imageSize.height - opticalCenter.y );
    float r_max = sqrtf( delta_ocx_max * delta_ocx_max + delta_ocy_max * delta_ocy_max );
 
    // Determine the vector from the optical center to the given point.
    float v_point_x = point.x - opticalCenter.x;
    float v_point_y = point.y - opticalCenter.y;
 
    // Determine the radius of the given point.
    float r_point = sqrtf( v_point_x * v_point_x + v_point_y * v_point_y );
 
    // Look up the relative radial magnification to apply in the provided lookup table
    float magnification;
    const float *lookupTableValues = lookupTable.bytes;
    NSUInteger lookupTableCount = lookupTable.length / sizeof(float);
 
    if ( r_point < r_max ) {
        // Linear interpolation
        float val   = r_point * ( lookupTableCount - 1 ) / r_max;
        int   idx   = (int)val;
        float frac  = val - idx;
 
        float mag_1 = lookupTableValues[idx];
        float mag_2 = lookupTableValues[idx + 1];
 
        magnification = ( 1.0f - frac ) * mag_1 + frac * mag_2;
    }
    else {
        magnification = lookupTableValues[lookupTableCount - 1];
    }
 
    // Apply radial magnification
    float new_v_point_x = v_point_x + magnification * v_point_x;
    float new_v_point_y = v_point_y + magnification * v_point_y;
 
    // Construct output
    return CGPointMake( opticalCenter.x + new_v_point_x, opticalCenter.y + new_v_point_y );
}
@CanCanZeng
Copy link

Hi, @interactivetech , do you know how to get the lensDistortionLookupTable? I searched for a long time but did not find any example. I also want to undistort the image to see the true depth image

@chaiyujin
Copy link

@CanCanZeng
Copy link

@chaiyujin Thank you for your advice. I find this API but I do not know how to use it. Actually I do not know much about swift or object-c, so I tried to find an example but without any luck.

@CanCanZeng
Copy link

I'm trying to build an application on top of ARKit, but I suspect that the image returned by ARKit is distorted, so I want to undistort the image and see what will happen.

@chaiyujin
Copy link

@CanCanZeng The lensDistortionPointForPoint is straightforward, if you simply follow the comments. To undistort image, you have give inverseDistortionLookupTable to lookupTable. Returned 2d-point is the undistorted position for input point in unit of pixel. Remember to keep point, opticalCenter and imageSize in the same coordinate system.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants