Stereo Depth Estimation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Stereo Depth Estimation

Computer Vision Project - I

As a part of this whole journey of building Autonomous Vehicles, first we


need to understand the world around us. Monocular Camera are widely
used in Autonomous Cars for tasks like Lane Detection, Traffic Sign
Detection, Object Detection, etc. One disadvantage of Monocular Cameras
is that they cannot infer about the depth information of various objects in
the surrounding. This problem can be solved using more than one camera
and applying Stereo Vision Algorithms to estimate the depth information
from an image.

Brief
In this project, we are going to use images from the KITTI Dataset (Stereo Images) and
estimate the depth information from the Left and Right images. The images should be
calibrated with the calibration parameters provided with the dataset.
We will be using Traditional Computer Vision Algorithms to calculate the Disparity and
Depth from the stereo images. OpenCV will be heavily used during the course of the project.
Once we are able to get a descent depth map, then we will take that depth map to build a
3D Reconstruction of the scene.

Topics: Stereo Vision, Epipolar Geometry, Triangulation, Depth Map.


Technical Details

Language: Python3.
Tools used: OpenCV, Intel Open3D.
Input: Stereo Pair
Output: Depth Map and 3D Reconstructed Scene.
Libraries Suggested:
• OpenCV – Image Processing and Computer Vision
• Intel Open3D – Final 3D Reconstruction
• NumPy – Matrix handling
• Argparse – Command Line Argument Passing.
Platform: Independent of any platform.
Duration of this project: 15 days. (Project + Documentation)

Other Important Details

• Code should be properly commented.


• The data should be calibrated with the calibration parameters given with the dataset.
• The program should be divided into modules and then submitted.
• You may or may not use Object Oriented Style but you must define function for any
important and reusable use case.
• In case of any doubt, contact your mentor- Ayushman Kumar.

Important Links
• Go get familiarize with KITTI Dataset and Calibration -
https://www.youtube.com/watch?v=SXW0CplaTTQ
• Depth Estimation with Stereo Vision – https://www.youtube.com/watch?v=t3LOey68Xpg

All the best for your project! I hope you learn a lot through this project and also build yourself for
the later projects in the near future. BEST WISHES!

You might also like