2018; Autonomous Navigation System (Electronics Club)

Introduction

This project was done as a part of Kriti, the Annual Technical Competition, IIT Guwahati, 2018. The aim was to develop an autonomous miniature car from scratch that can detect lanes and traverse through it. Although the whole car couldn’t be completed by the date of the competition, the computer vision algorithm was developed and it won us the SILVER.

Details

• Since the lane is white in color with black colored borders, we first convert the incoming image to grayscale to simplify things for us.

• We provide sliders before the tracking algorithm starts running to select the following –

  • Upper and lower ends of the image to crop out the necessary areas.

  • Upper and lower thresholds for the pixel values to distinguish between the lane and the border(darker).

    

  • By selecting the appropriate grayscale value thresholds and considering no harsh changes in the lighting conditions happen, we get a noisy binary mask of the lane, which can be reduced by some morphological transformations.

    

  • Going along the y axis, we know the pixel locations of the inner sides of the lane using simple binary threshold.

  • Random sample along the y axis for several points on the inner line.

  • By performing linear regression over this (x,y) data, we get a linear fit of the points which can be considered as the inner side of the lane. For the given frame. We used sklearn’s regression module for this.

    

  • The angle between the point of intersection of the linear fits of the left and right sides(cyan line) and the center of the image(white line) gives an indication of the angle of the car w.r.t the lanes.

  • This angle is now used to control the turning of the car with turn velocity proportional to this angle.