OFSAI-Toolkit

An easy-to-use, open-source toolkit for self driving racing car development. Please try and improve on the tools offered and contribute to help everyone.

Features:

• Cone detection
• Cone Mapping
• Track Boundary Estimation
• Waypoint Generation

• Online Image Annotator (with Active Learning)
• Visualisations
  • Annotation Renderer
  • 2D Track Visualiser (OpenCV)
  • 2D Track Visualiser (PyGame)

TODO:

• Racing line calculation
• Car Physics
• On-the-line custom track maker
• Yolo Trains Via Data generator where each imagine is loaded differently with random alterations
• Visualise 3D track
• Driver model
• SLAM (EKF)
• Compile to Cython

Contents

1. Objects
2. Visualisations
3. Perception
4. Mapping
5. Track Boundary Estimation

1. Objects

This repository comes with, and utilises, a few objects that represent objects found in the FS-AI competition.

1.1. Point

Used to represent the position of an object in 2D.

from fsai.objects.point import Point
point = Point(x, y)

1.2. Cone

Used to represent a cone in the track. Each cone contains a point object representing the position of a cone, as well as the colour of the cone.

from fsai.objects.cone import Cone
fsai.objects.cone.CONE_COLOR_BLUE = 0  
fsai.objects.cone.CONE_COLOR_YELLOW = 1  
fsai.objects.cone.CONE_COLOR_BIG_ORANGE = 2  
fsai.objects.cone.CONE_COLOR_ORANGE = 3

The cone object can be constructed as follows:

cone = Cone()  # blue cone at 0, 0
cone = Cone(point=Point(9, 2), color=CONE_COLOR_YELLOW)  # yellow cone at 9, 2
cone = Cone(x=4, y=6, color=CONE_COLOR_ORANGE)  # orange cone at 4, 6

1.3. Line

Object containing two points. Used to represent lines such as track boundaries.

from fsai.objects.line import Line
line = Line(a=Point(0, 0), b=Point(1, 0))

1.4. Track

An object used to encapsulate cones into one handy class.

from fsai.objects.track import Track
track = Track()

Additionally this class allows you to save and load tracks in the json format:

{
    "blue_cones": [
        {"x": 1.25, "y": -14.75},
        {"x": 5, "y": -15.5}
    ],
    "yellow_cones": [
        {"x": 2.875, "y": -9.875},
        {"x": 6, "y": -10.375}
    ],
    "big_cones": [
        {"x": -2, "y": -13.625},
        {"x": -3.75, "y": -12.875}
    ],
    "orange_cones": []
}

import json
from fsai.objects.track import Track
# Loading
track = Track("examples/data/tracks/laguna_seca.json")
# -or-
track.load_track("examples/data/tracks/brands_hatch.json")
# Saving
track.save_track("output_track.json")
# -or-
with open("file", "w+") as file:
    file.write(json.dumps(track.to_json()))

2. Visualisations

2.1. Image Annotations

Annotates a given image with given annotations, returning an openCV formatted image.
<class> <x> <y> <w> <h>
(Relative to the size of the image)

Parameters:

labels_path - If provided, the labels will be loaded from a file destination
label_annotations - If provided, the annotations will be draw from the given information
image_path - If provided the image will load from the provided path
image - If provided the labels will be draw onto this opencv image
colors - If provided the annotations will use the colours provided. By default colours are procedural generated
class_names - If provided the annotations will draw class names as well
line_width - Annotations will be draw with the width provided
returns - OpenCV Image

Example Usages:

from fsai.visualisation.image_annotations import annotate
image = annotate(labels_path="labels/0.txt", image_path="images/0.jpg", colors=[(255, 0, 0), (0, 255, 0)], line_width=10)
cv2.imshow("Cones", image)
cv2.waitKey(0)

from fsai.visualisation.image_annotations import annotate
image = cv2.imread("images/1.png")
annotations = [
    [0, 0.4, 0.6, 0.25, 0.25],
    [1, 0.8, 0.8, 0.1, 0.1],
]
image = annotate(image=image, label_annotations=annotations, class_names=["dog", "plane"], line_width=2)
cv2.imshow("Cones", image)
cv2.waitKey(0)

2.2. 2D Track Renderer

This method is used to draw the given items in the scene onto an image. Cones and lines can be provided,
additionally the image can be scaled and padded for aesthetic reasons. Colours can be customised to suit
you liking by altering the method parameters.

Parameters:

track - If provided, the track data will be draw.
cones - If provided, these cones will be drawn.
blue_cones - If provided, these cones will be drawn.
yellow_cones - If provided, these cones will be drawn.
orange_cones - If provided, these cones will be drawn.
big_cones - If provided, these cones will be drawn.
blue_lines - If provided, these lines will be drawn.
yellow_lines - If provided, these lines will be drawn.
orange_lines - If provided, these lines will be drawn.
waypoints - If provided, these waypoints will be draw.
blue_line_colour - The colour in which to render blue lines.
yellow_line_colour - The colour in which to render yellow lines.
orange_line_colour - The colour in which to render orange lines.
blue_cone_colour - The colour in which to render blue cones.
yellow_cone_colour - The colour in which to render yellow cones.
orange_cone_colour - The colour in which to render orange cones.
big_cone_colour - The colour in which to render big orange cones.
background - Background greyscale color to draw the scene onto (0 - 255) (int).
scale - Scale the image of the track.
padding - Add padding around the image.
return - OpenCV Image.

Example Usages:

from fsai.objects.track import Track
from fsai.visualisation.draw_opencv import draw_track
track = Track("examples/data/tracks/laguna_seca.json")
image = draw_track(track=track)

from fsai.objects.track import Track
from fsai.visualisation.draw_opencv import draw_track
track = Track("examples/data/tracks/imola.json")
blue_lines, yellow_lines, orange_lines = track.get_boundary()
image = draw_track(
    track=track,
    blue_lines=blue_lines,
    yellow_lines=yellow_lines,
    orange_lines=orange_lines,
    background=100,
    scale=20,
    padding=50
)

3. Perception

doc coming soon…

4. Mapping

doc coming soon…

5. Track Boundary Estimation

doc coming soon…

6. Path Planning

6.1 Waypoint Generation

doc coming soon…

6.2 Path Prediction

doc coming soon…