/PX4-OpticalFlow

This repository contains different algorithms to calculate the optical flow. It can be used as input for a position estimator

Primary LanguageC++

PX4 Optical Flow repository

This package provides different algorithms to calculate the optical flow. Currently, there is the PX4Flow algorithm (SAD-block-matching) and an OpenCV based (KLT) version.

Please feel free to improve the current algorithms or add new versions.

Usage

PX4Flow version

The constructor OpticalFlowPX4() needs the following parameters (in this order):

  • Focal length in X-Direction [pixels]
  • Focal length in Y-Direction [pixels]
  • Output rate [Hz] (Optional, default = 15Hz) If smaller than frame rate, flow gets integrated. set to -1 to use the image frame rate
  • Image width [pixels] (Optional, default = 64)
  • Image height [pixels] (Optional, default = 64)
  • Search size [pixels] (Optional, default = 6)
  • Flow feature threshold (Optional, default = 30)
  • Flow value threshold (Optional, default = 3000)

For the actual flow calculation the function calcFlow() has to be used. It is the same for both versions and needs the following arguments:

Input:

  • Image array [* uint8_t]
  • Image time stamp [us, uint32_t]

Output:

  • Delta time [us, int]
  • Angular flow in X-Direction [rad/s, float]
  • Angular flow in Y-Direction [rad/s, float]

Return:

  • Flow quality [int], 0-255, -1 if it should not be used yet (output rate)

OpenCV version

The constructor OpticalFlowOpenCV() needs the following parameters (in this order):

  • Focal length in X-Direction [pixels]
  • Focal length in Y-Direction [pixels]
  • Output rate [Hz] (Optional, default = 15Hz) If smaller than frame rate, flow gets integrated. set to -1 to use the image frame rate
  • Image width [pixels] (Optional, default = 64)
  • Image height [pixels] (Optional, default = 64)
  • Number of tracked features (Optional, default = 20)
  • Confidence multiplier for outlier rejection (Optional, default = 1.645, 90% confidence interval)

For the actual flow calculation the function calcFlow() has to be used. It is the same for both versions and needs the following arguments:

Input:

  • Image array [* uint8_t]
  • Image time stamp [us, uint32_t]

Output:

  • Delta time [us, int]
  • Angular flow in X-Direction [rad/s, float]
  • Angular flow in Y-Direction [rad/s, float]

Return:

  • Flow quality [int], 0-255, -1 if it should not be used yet (output rate)

If you have a camera calibration, you can undistort the tracked features (only for OpenCV version). To use that functionality you have to set the camera matrix setCameraMatrix() AND distortion parameters setCameraDistortion() after calling the constructor.

setCameraMatrix() needs the following arguments:

  • Focal length in X-Direction [pixels]
  • Focal length in Y-Direction [pixels]
  • Principal point in X-Direction [pixels]
  • Principal point in Y-Direction [pixels]

setCameraDistortion() needs the following arguments (see here):

  • Radial coefficient k1
  • Radial coefficient k2
  • Radial coefficient k3
  • Tangential coefficient p1
  • Tangential coefficient p2

An example can be found here.

Code style

For code formatting astyle should be used with the .astylerc options file.

astyle <file> --options=.astylerc