SPRUJB3 March 2024 AM67 , AM67A , TDA4AEN-Q1 , TDA4VEN-Q1
When a camera is viewing a scene from two different positions or when multiple cameras are viewing the scene from different positions, a transformation between the two viewing angles is needed to align the images. Under specific conditions, the class of geometric transformations known as homography, or planar-perspective transformation, will capture the geometric relationship between the images accurately. Common applications of homography transforms are to align (or stitch) multiple frames of the same scene to compute a panoramic output image. A second application is the alignment of planar surfaces in the world. Finally, perspective transforms are also useful in computing depth maps from a stereo image pair. By rectifying the two views, the search to compute disparity between the two views is simplified to a 1-D search problem. The homography is defined by a 3x3 transformation matrix, as in
The affine transform is a subset of the perspective transformation. By setting g = h = 0, hp = haff and vp = vaff.
In image alignment applications, the homography matrices are computed by locating corresponding points in the two frames and estimating the matrix parameters to transform the set of points in one frame onto the corresponding points in the second frame. In the stereo rectification application, the matrix is determined (pre-computed) at the calibration step and remains fixed.
When LDC is requested to provide a change in the frame size between the input and output, the affine/perspective parameters must be programmed to implement the coordinate scaling (see for programming details). In the following sections, the mesh table is used to program distortion correction. The table sizes are defined based on the total output frame size.