WIP: Update/proclivity #56
Conversation
There was a problem hiding this comment.
In general, this looks okay but I would like to improve this function by documenting it thoroughly (with comments/assumptions/requirements and variable naming, etc.) as well implementing some best-practices. We also need a unit test to prove that it works correctly. Ideally this would have a case with perfect data where the check succeeds and a few cases with bad data to prove that it fails (i.e. noisy data, bad correspondences, individual outliers). Per @drchrislewis this functionality has existed but not really been thoroughly tested
| double max_residual; //this can be omitted if residuals are compared internally | ||
| }; | ||
|
|
||
| bool checkObservationProclivity(const ProclivityParams& params); |
There was a problem hiding this comment.
Need more documentation about the purpose of this function and the assumptions made
| double max_residual; //this can be omitted if residuals are compared internally | ||
| }; | ||
|
|
||
| bool checkObservationProclivity(const ProclivityParams& params); |
There was a problem hiding this comment.
Just in general, let's use the term homography instead of proclivity to refer to this check, which is more accurate according to @drchrislewis
| Eigen::MatrixXd Alpha(8,1); //Homography matrix components | ||
| Eigen::MatrixXd A(18,8); | ||
| Eigen::MatrixXd B(1,18); //U,V coordinates of points | ||
|
|
||
| // determine rows and cols of target | ||
| int rows = params.target.rows; | ||
| int cols = params.target.cols; | ||
|
|
||
| //we want *9* points, spiraling in clockwise: UL, UR, LL, LR, UL +1, UR -1. etc. | ||
| std::vector<int> selected_point_index{0, rows-1, rows*cols -1, rows*cols - cols-1, | ||
| 1, rows-2, rows*cols -2, rows*cols - cols-2, | ||
| rows*cols/2}; |
There was a problem hiding this comment.
The row size of matrix A is related to the number of points you choose (i.e. 2*n). I would parameterize the number of points used in the calculation for clarity and in case we need to change it. I would also only use the minimum number of points to construct A so we can check use A to check more of the remaining points that weren't involved in its creation.
We should also check that the target has more points than the minimum required (4) by some constant factor (i.e. 2, so 8 points minimum). Otherwise throw an exception because this check is not valuable.
Please document that we are using corner points of the target in order to create the largest possible plane for our assumed rectangular planar target
|
|
||
| // build matrix of type Ax=b where A x is the unknown elements of the proclivity matrix "alpha" | ||
| int row = 0; | ||
| for(int i=0; i<(int) selected_point_index.size(); i++){ |
There was a problem hiding this comment.
Prefer size_t for iterators instead of casting
| bool checkObservationProclivity(const rct_image_tools::ProclivityParams& params) | ||
| { | ||
|
|
||
| Eigen::MatrixXd Alpha(8,1); //Homography matrix components |
There was a problem hiding this comment.
More documentation on the math that these objects will be used for
| row++; | ||
| } | ||
| //solve with Eigen | ||
| //beware gcc's -ffast-math, hurts accuracy of JacobiSvd |
| double EU = Ui - UV(0); | ||
| double EV = Vi - UV(1); |
There was a problem hiding this comment.
Make this an Eigen vector with a more descriptive name, i.e. homograpy_error
| // check the proclivity of every observation | ||
| bool rtn = true; | ||
| double ave_error = 0.0; | ||
| std::vector<double> errorU, errorV; |
There was a problem hiding this comment.
Make this a single vector of Eigen 2D vectors
| struct ProclivityParams | ||
| { | ||
| rct_optimizations::CameraIntrinsics intr; | ||
| rct_image_tools::ModifiedCircleGridTarget target; | ||
| rct_optimizations::Observation2D3D ob; //learn to use templates, and then do this right | ||
| double max_residual; //this can be omitted if residuals are compared internally | ||
| }; |
There was a problem hiding this comment.
Probably don't need a separate struct for inputs to the homography checking function, especially if you think we should have a default max_residual
| // construct the Proclivity matrix from Alpha | ||
| Eigen::Matrix3d P; | ||
| P << Alpha; | ||
| P(2,2) = 1.0; |
There was a problem hiding this comment.
Please document that here we are rearranging the elements of the homography matrix Alpha (which is a vector) into a 3x3 matrix, where the (2, 2) element is 1.
colin-lewis-19
force-pushed
the
update/proclivity
branch
from
563d723
to
306ba7a
Compare
…ented, and some improved methods
…ther documentation
…NaN returns or 60 deg oritentation shits
Sensor Noise Qualification Updates
…rtially implemented
colin-lewis-19
force-pushed
the
update/proclivity
branch
from
306ba7a
to
ec79db6
Compare
|
Closing; this PR was replaced by #66 |
A function for detecting if target points have been misordered. A pixel threshold can be provided, or outliers can be detected internally by comparing maximum values to interquartile range. Built atop #52