Stereo Disparity Estimator

Overview

Given a pair of rectified images from a stereo camera, the Stereo Disparity algorithm uses high-quality dense stereo matching to produce an output image of the same resolution as the input with left-right disparity information. This allows for inferring the depth of the scene captured by the left and right images.

Left image	Right image
Disparity map

Implementation

The stereo disparity estimator uses semi-global matching algorithm to compute the disparity. We deviate from the original algorithm by using as cost function the hamming distance of the census transforms of the stereo pair.

Usage

1. Initialization phase
   1. Include the header that defines the needed functions and structures.

      #include <vpi/algo/StereoDisparityEstimator.h>

   2. Define the stream on which the algorithm will be executed, the input stereo pair, composed of two images, and the output disparity image.

          VPIStream stream = /*...*/;
          VPIImage left    = /*...*/;
          VPIImage right   = /*...*/;

   3. Create the payload that will contain all temporary buffers needed for processing.

          VPIPayload stereo;
          vpiCreateStereoDisparityEstimator(stream, 480, 270, VPI_IMAGE_TYPE_Y16, 64, &stereo);

2. Processing phase
   1. Define the configuration parameters needed for algorithm execution.

          VPIStereoDisparityEstimatorParams params;
          params.windowSize   = 5;
          params.maxDisparity = 64;

   2. Submit the payload for execution on the stream associated with it.

          vpiSubmitStereoDisparityEstimator(stereo, left, right, disparity, &params);

   3. Optionally, wait until the processing is done.

          vpiStreamSync(stream);

3. Cleanup phase
   1. Free resources held by the payload.

          vpiPayloadDestroy(stereo);

Consult the Stereo Disparity Sample for a complete example.

For more details, consult the API reference.

Limitations and Constraints

Constraints for specific backends superceed the ones specified for all backends.

All Backends

• Left and right input images must have same type and dimensions.
• Output image dimensions must match input's.
• Output disparity images must have type VPI_IMAGE_TYPE_U16.
• Maximum disparity parameter passed to algorithm submission must be the same as defined during payload creation.
• Input image dimensions must match what is defined during payload creation.

CUDA and CPU

• Left and right accepted input image types:
  • VPI_IMAGE_TYPE_U8
  • VPI_IMAGE_TYPE_U16

PVA

• Left and right accepted input image types:
  • VPI_IMAGE_TYPE_U16
• Input and output image dimensions must be 480x270.
• windowSize must be 5.
• maxDisparity must be 64.

Performance

For further information on how how performance benchmarked, see Performance Measurement.

Jetson AGX Xavier

size	type	max disp.	CPU	CUDA	PVA
480x270	u8	64	365 ms	5.770 ms	n/a
480x270	u8	32	206 ms	6.926 ms	n/a
480x270	u16	64	357 ms	5.823 ms	14.336 ms
480x270	u16	32	206.9 ms	7.0159 ms	n/a

Jetson TX2

size	type	max disp.	CPU	CUDA	PVA
480x270	u8	64	1010 ms	23.06 ms	n/a
480x270	u8	32	483.0 ms	25.2 ms	n/a
480x270	u16	64	1002 ms	23.2 ms	n/a
480x270	u16	32	484.5 ms	25.1 ms	n/a

Jetson Nano

size	type	max disp.	CPU	CUDA	PVA
480x270	u8	64	1791 ms	54.88 ms	n/a
480x270	u8	32	863 ms	51.3 ms	n/a
480x270	u16	64	1795 ms	54.73 ms	n/a
480x270	u16	32	867 ms	51.3 ms	n/a

References

• Hirschmüller, Heiko (2005). "Accurate and efficient stereo processing by semi-global matching and mutual information".
  IEEE Conference on Computer Vision and Pattern Recognition. pp. 807–814.
• Zabih, Ramin; Woodfill, John (1994). "Non-parametric local transforms for computing visual correspondence".
  European conference on computer vision. pp. 151–158.