Visual Odometry for Autonomous Vehicle

This is a software written as a part of project for the course Perception for Autonomous Vehicles.

Contents

1. Introduction
2. Dataset and Processing
3. Dependencies
4. Pipeline

Introduction

In robotics and computer vision, visual odometry is the process of determining the position and orientation of a robot by analyzing the associated camera images. Visual Odometry is a crucial concept in Robotics Perception for estimating the trajectory of the robot (the camera on the robot to be precise). The concepts involved in Visual Odometry are quite the same for SLAM which needless to say is an integral part of Perception. In this project we are given frames of a driving sequence taken by a camera in a car, and the scripts to extract the intrinsic parameters.

Dataset and Processing

The dataset can be obtained at Link

1. The first step in the project was the preparation of the given dataset and reading all the images in the dataset. The input images are in Bayer format on which demosaicing function with GBRG alignment was used. Thus, the Bayer pattern encoded image img was converted to a color image using the opencv function: color image = cv2.cvtColor(img, cv2.COLOR BayerGR2BGR)

2. The next step in the data preparation phase was to extract the camera parameters using ReadCamer- aModel.py as follows: fx , fy , cx , cy , G camera image , LUT = ReadCameraModel ( './model')

3. The images in the given dataset were further Undistorted using the current frame and next frame using UndistortImage.py: undistorted image = UndistortImage(originalimage,LUT)

Dependencies

• opencv
• numpy
• matplotlib
• scipy

Pipeline

To estimate the 3D motion (translation and rotation) between successive frames in the sequence, the following steps were followed.

1. Find point correspondences between successive frames using a keypoint algorithm of your choice.
2. Estimate the Fundamental Matrix using the Eight-Point Algorithm within RANSAC
3. Estimation of Essential Matrix from the Fundamental matrix
4. Camera Pose computation from the Essential Matrix
5. Linear Triangulation

• For more information about the pipeline I welcome you to have a look at the Report