Stereo Depth Estimation : Python/OpenCV2 C++/DirectX11

Colab Link : https://colab.research.google.com/drive/1Ey4mfWyh3wSpEplk6TY8eJepNdbtIoJn?usp=sharing

Showcase Video Of Point Cloud Visualizer

[Note : This was a group project of 5. The techniques noted below are my personal contributions towards the project]

Features

  • Implemented Morphological transforms such as Hit or Miss transform, Erosion, Dilation, Opening, Closing, Template Matching.
  • Added several edge detection algorithms such as Sorbel edge detectors, Laplacian edge detectors, and Unsharp masking.
  • Disparity/Depth map generation by extracting relevant camera calibration data.
  • Implemented segmentation algorithms such as WaterShed segmentation and Colour space segmentation (HSV).
  • For pre-processing of images, implemented frequency domain filters such as Butterworth’s low/high pass filter, Ideal low/high pass filter
  • Created a C++/DirectX11 application to represent the depth map in form of a point cloud visualizer.

Spatial Domain Sharpening Filters

Source / Original Image N4 Laplacian Mask / Sharpened Image N8 Laplacian Mask / Sharpened Image Sorbel Mask / Sharpened Image Unsharp Masking Mask / Sharpened Image

Frequency Domain Blurring / Sharpening Filters

Ideal Low and High Pass Filter Masks Application of Ideal Low and High Pass Filters

Application of ButterWorth’s Low and High Pass Filters

Other Image Processing Techniques

Color Space Segmentation (HSV)

Homomorphic Transformations

Morphological Closing

Hit or Miss Transform (object to hit : The 2 wheeler Driver)

Template Matching

Disparity and Depth Map Generation