WarpTransform

The xFWarpTransformfunction is designed to perform the perspective and affine geometric transformations on an image. The type of transform is a compile time parameter to the function.

The function uses a streaming interface to perform the transformation. Due to this and due to the fact that geometric transformations need access to many different rows of input data to compute one output row, the function stores some rows of the input data in BRAMs. The number of rows the function stores can be configured by the user by modifying a template parameter. Based on the transformation matrix, you can decide on the number of rows to be stored. You can also choose when to start transforming the input image in terms of the number of rows of stored image.

Affine Transformation

The transformation matrix consists of size parameters, and is as shown:

Affine transformation is applied in the xFWarpTransform function following the equation:

Perspective Transformation

The transformation matrix is a 3x3 matrix as shown below:

Perspective transformation is applied in xFWarpTransform following the equation:

The destination pixel is then computed by dividing the first two dimensions of the dst1 by the third dimension

API Syntax

template void xFWarpTransform(xF::Mat & src, xF::Mat & dst, float *transformation_matrix)

Parameter Descriptions

The following table describes the template and the function parameters.

Table 1.xFWarpTransform Function Parameter Descriptions
Parameter Description
STORE_LINES Number of lines of the image that need to be buffered locally on FPGA.
START_ROW Number of the input rows to store before starting the image transformation. This must be less than or equal to STORE_LINES.
TRANFORMATION_TYPE Affine and perspective transformations are supported. Set this flag to ‘0’ for affine and ‘1’ for perspective transformation.
INTERPOLATION_TYPE Set flag to ‘1’ for bilinear interpolation and ‘0’ for nearest neighbor interpolation.
SRC_T Input pixel type. Only 8-bit, unsigned, 1 channel is supported (XF_8UC1)
ROWS Maximum height of input and output image.
COLS Maximum width of input and output image.
NPC Number of pixels to be processed per cycle; only one-pixel operation supported (XF_NPPC1).
src Input image
dst Output image
transformation_matrix Transformation matrix that is applied to the input image.

Resource Utilization

The following table summarizes the resource utilization of the Warp transform, generated usingVivado HLS 2017.1version tool for the Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to progress a grayscale HD (1080x1920) image.

Table 2.xFWarpTransform Function Resource Utilization Summary
Transformation INTERPOLATION _TYPE STORE _LINES START _ROW

Operating Frequency

(MHz)

 Utilization Estimate
LUTs FFs DSPs BRAMs
Perspective Bilinear 100 50 300 7468 9804 61 112
Perspective Nearest Neighbor 100 50 300 4514 6761 35 104
Affine Bilinear 100 50 300 6139 5606 40 124
Affine Nearest Neighbor 100 50 300 4611 4589 18 112

Performance Estimate

The following table summarizes a performance estimate of the Warp transform, as generated usingVivado HLS 2017.1tool for Xilinx Xczu9eg-ffvb1156-1-i-es1 FPGA, to process a grayscale HD (1080x1920) image.

Table 3.xFWarpTransform Function Performance Estimate Summary
Transformation INTERPOLATION _TYPE STORE _LINES START _ROW

Operating Frequency

(MHz)

Latency Estimate

Max (ms)
Perspective Bilinear 100 50 300 7.46
Perspective Nearest Neighbor 100 50 300 7.31
Affine Bilinear 100 50 300 7.31
Affine Nearest Neighbor 100 50 300 7.24