Environement Perception for Self Driving Cars

An Implementation of Environment Perception for a Self Driving Car for the Course Introduction to Self-Driving Cars (Coursera)

DESCRIPTION:

Implemention of Drivable Surface Estimation, Lane estimation and 2D object and obstacle detection using the output of semantic segementation implemented using a convolutional neural network

• Input to the system is semantic segmentation output and depth map for every pixel in the images.

  The inputs are automatically imported and could be used through the following functions:

  DatasetHandler().rgb: a camera RGB image

  DatasetHandler().depth: a depth image containing the depth in meters for every pixel.

  DatasetHandler().segmentation: an image containing the output of a semantic segmentation neural network as the category per pixel.

  DatasetHandler().object_detection: a numpy array containing the output of an object detection network.

• Output :

  1. Drivable surface plane parameters
  2. Lane boundary equations for the drivable surface
  3. minimum distance to the obstacles in 3D cordinates

SOLUTION :

DRIVABLE SURFACE SHOWN IN YELLOW:

LANE MARKINGS FOR DRIVABLE SURFACE:

LOCATION AND DISTANCE TO OBSTACLE:

Dependencies:

• Jupyter Notebook
• Python 3.5 or 3.6