TrackLineVisualization

Requirements

• python 3
  • yaml
  • opencv
  • matplotlib

Testing Environment(macOS)

• python 3.6
  • yaml 0.1.7
  • Opencv 3.4.2

Usage

Testing

1. Enter the project folder
2. Open command line，and type in following command( just as an example):

python track_line_generator.py  --tread 1.832 --wheelbase 2.871 --front_wheel_to_head_d 0.89 --head_height 0.68 --camera_yaml_path your_yaml_file_path --video_path your_video_path

Or you could type in python track_line_generator.py -h for help:

usage: track_line_generator.py [-h] -t -w -f -e -c -v 

optional arguments:
  -h, --help            show this help message and exit
  -t, --tread       car width (meters)
  -w, --wheelbase   distance between the front and rear axles of a vehicle
  -f, --front_wheel_to_head_d 
                        distance between front wheel center and car head
  -e, --head_height 
                        z position of the point in the car head
  -c, --camera_yaml_path 
                        yaml file path for camera configuration
  -v, --video_path  video path

Call from External File

from  track_line_generator.py import NewTrackLineGenerator

base_param = BaseParam(tread, wheelbase, head_height, front_wheel_to_head_d, param_yaml_path)
track_line_generator = NewTrackLineGenerator(base_param)
left_line,right_line = track_line_generator.add_track_line(steer_angle)

Visualization Example

Sample results cannot be displayed due to data privacy.

Deficiencies and Bugs

Deficiencies

1. trajectory equation

   • Curve left: $\left.x^{2}+(y+wheelbase * \cot (\phi)\right)^{2}=r_{L}^{2}$ Curve Right: $x^{2}+(y+wheelbase * \cot (\phi))^{2}=r_{R}^{2}$

   • $r_{L}^{2} = \left(wheelbase * \cot (\phi)+\frac{tread}{2}\right)^{2}+(wheelbase + front_wheel_to_head )^{2}$

     $r_{R}^{2} = \left(wheelbase * \cot (\phi)-\frac{tread}{2}\right)^{2}+(wheelbase + front_wheel_to_head )^{2}$

   

2. transform matrix is estimated without considering distortion $\mathcal{D}$

   • $X_{w}, Y_{w}, Z_{w}$ is the coordinates of track point in the real world (under vehicle coordinate system).
   • $u,v$ is the coordinates of pixel point corresponding to $X_{w}, Y_{w}, Z_{w}$.
   

   • The reason for ignore distortion $\mathcal{D}$ is that $\mathcal{D}$ cannot be represented by matrix multiplication ($\mathcal{D}$ is:

     

     $k_{1}, k_{2}, \rho_{1}, \rho_{2}, k_{3}$ are distortion parameters.

Bugs

1. The current change matrix is fixed. However, translation vector should be different with the one obtained afther calibration, and we doesn't know the plane's position when calibration.

Appendix