Perspective-3-Point

p3p(points, pdn_pixels::AbstractVector{SVector{3, T}}, K)

Recover pose using P3P algorithm.

Arguments:

• points: Vector of 3D points in(x, y, z)` format.
• pdn_pixels::AbstractVector{SVector{3, Float64}}: Corresponding projections of points onto image plane, predivided by K intrinsics and normalized. E.g.: Ki = inv(K); p = Ki [x, y, 1]; p /= norm(p).
• K::SMatrix{3, 3, Float64}: Camera intrinsics.

Returns:

Vector{SMatrix{3, 4, Float64}} vector of up to 4 possible solutions. Each element is a projection matrix P = K * [R|t]. To get pure transformation matrix, multiply P by inv(K).

References:

Link: https://cmp.felk.cvut.cz/~pajdla/gvg/GVG-2016-Lecture.pdf
chapter: 7.3 Calibrated camera pose computation.
pages: 51-59

p3p_select_model(models, points, pixels; threshold = 1.0)

Select best pose from models.

Arguments:

• models: Projection matrices from which to select best one.
• points: 3D points in (x, y, z)
• pixels: Corresponding projections onto image plane in (x, y) format.
• threshold: Maximum distance in pixels between projected point and its target pixel, for the point to be considered inlier. Default value is 1.0.

Returns:

n_inliers, (projection, inliers, error).

• n_inliers: Number of inliers for the projection.
• projection: K * P projection matrix that projects points onto image plane.
• inliers: Boolean vector indicating which point is inlier.
• error: Average reprojection error for the pose.

p3p_ransac(points, pixels, pdn_pixels, K; threshold = 1.0, ransac_kwargs...)

Recover pose K*[R|t] using P3P Ransac algorithm.

Arguments:

• points: 3D points in (x, y, z)
• pixels: Corresponding projections onto image plane in (x, y) format.
• pdn_pixels: Corresponding projections onto image plane, predivided by K intrinsics and normalized. E.g.: Ki = inv(K); p = Ki [x, y, 1]; p /= norm(p).
• K: Camera intrinsics.
• threshold: Maximum distance in pixels between projected point and its target pixel, for the point to be considered inlier. Default value is 1.0.
• ransac_kwargs...: Keyword arguments passed to ransac.

Returns:

n_inliers, (projection, inliers, error).

• n_inliers: Number of inliers for the projection.
• projection: K * P projection matrix that projects points onto image plane.
• inliers: Boolean vector indicating which point is inlier.
• error: Average reprojection error for the pose.

References:

Link: https://cmp.felk.cvut.cz/~pajdla/gvg/GVG-2016-Lecture.pdf
chapter: 7.3 Calibrated camera pose computation.
pages: 51-59

p3p_ransac(points, pixels, K; threshold = 1.0, ransac_kwargs...)

Recover pose K*[R|t] using P3P Ransac algorithm.

Arguments:

• points: 3D points in (x, y, z)
• pixels: Corresponding projections onto image plane in (x, y) format. These values, WILL be predivided and normalized by the intrinsic matrix. E.g.: Ki = inv(K); p = Ki [x, y, 1]; p /= norm(p).
• K: Camera intrinsics.
• threshold::Real: Maximum distance in pixels between projected point and its target pixel, for the point to be considered inlier. Default value is 1.0.
• ransac_kwargs...: Keyword arguments passed to ransac.

Returns:

Vector{SMatrix{3, 4, Float64}} vector of up to 4 possible solutions. Each element is a projection matrix P = K * [R|t]. To get pure transformation matrix, multiply P by inv(K).

References:

Link: https://cmp.felk.cvut.cz/~pajdla/gvg/GVG-2016-Lecture.pdf
chapter: 7.3 Calibrated camera pose computation.
pages: 51-59

pre_divide_normalize(pixels, K)

Divide pixels by K intrinsic matrix and normalize them. pixels in (x, y) format.