GstCUDA - Example - opencvfilter

From RidgeRun Developer Wiki


Previous: Example - opencvwarp Index Next: Performance Profiling





Introduction

The opencvfilter element exposes a series of image filters from the popular computer vision framework OpenCV. The following figure shows some of the available filters applied to the same image.

Figure 1. Example of opencv available filters.

Element properties

  • size

The size of the filter. Depending on the filter, it could either be the kernel size or the aperture size. Must be odd.
Type: Integer
Range: 1 - 2147483647
Flags: readable, writable
Default: 3

  • filter

Type of the filter to create. Available options:

box: Normalized box filter
box-max: Maximum filter
box-min: Minimum filter
column-sum: Vertical 1D box filter
deriv: General deriv filter
gaussian: Gaussian filter
laplacian: Laplacian filter
linear: General linear filter
morphology: Morphological filter
row-sum: Horizontal 1D box filter
scharr: Scharr filter
separable: Separable linear filter
sobel: Sobel filter

Flags: readable, writable
Default: "gaussian"

  • dx

Derivative order in respect of X (for deriv, sobel and scharr).
Type: Integer
Range: 0 - 2
Flags: readable, writable
Default: 0

  • dy

Derivative order in respect of Y (for deriv, sobel and scharr).
Type: Integer
Range: 0 - 2
Flags: readable, writable
Default: 0

  • scale

Optional scale factor for the computed values (for deriv, laplacian, sobel and scharr).
Type: Integer
Range: 1 - 2147483647
Flags: readable, writable
Default: 1

  • iterations

Number of times for the morphological operations to be applied (for morphology).
Type: Integer
Range: 1 - 2147483647
Flags: readable, writable
Default: 1

  • morph-op

Type of morphological operation (for morphology). Available options:

erode: Normalized box filter
dilate: Maximum filter
open: Minimum filter
close: Vertical 1D box filter

Flags: readable, writable
Default: "erode"

  • kernel

-2D matrix of NxM to use as kernel (for linear and morphology filters).
-2D matrix of 1xM to use as a row kernel (for separable filter).
Type: GstValueArray of GValues of type GstValueArray
Flags: readable, writable
Default: <<0.0,0.0,0.0>,<0.0,1.0,0.0>,<0.0,0.0,0.0>>

  • kernel2

-2D matrix of 1xM to use as a column kernel (for separable filter).
Type: GstValueArray of GValues of type GstValueArray
Flags: readable, writable
Default: <<0.0,0.0,0.0>,<0.0,1.0,0.0>,<0.0,0.0,0.0>>

Examples

The following examples pipe will help you apply different types of filters to the input video stream from your camera. These pipes were tested on both, the Nvidia Jetson TX1 and TX2.

Border Detection

gst-launch-1.0 v4l2src ! nvvidconv ! opencvfilter filter=sobel dx=0 dy=1 size=1 ! queue ! nvvidconv ! autovideosink

Linear Filter with Custom Kernel

For a Gaussian filter given by the following 2D kernel:

You may use the following pipeline:

gst-launch-1.0 v4l2src ! nvvidconv ! opencvfilter filter=linear kernel="<<0.0625,0.125,0.0625>,<0.125,0.25,0.125>,<0.0625,0.125,0.0625>>" ! queue ! nvvidconv ! autovideosink

Separable Filter with Custom Row and Column Kernels

For a Gaussian filter given by the row kernel:

and the column kernel:

You may use the following pipeline:

gst-launch-1.0 v4l2src ! nvvidconv ! opencvfilter filter=separable kernel="<<0.25,0.5,0.25>>" kernel2="<<0.111,0.388,0.388,0.111>>" ! queue ! nvvidconv ! autovideosink

Performance Measurements

Previous: Example - opencvwarp Index Next: Performance Profiling