Starting from:

$29.99

Assignment #2 Hough transform

COMP425-6341 COMPUTER VISION 
Programming Assignment #2
Description
The objective of this assignment is to gain an understanding of the methods discussed in the class relating
to Hough transform, Harris corner detection, and feature descriptors/matching. You can use C, C++, or
Python to implement the solutions.
Part 1: Hough transform (7.5 pts)
A. Without using OpenCV’s hough transform functions, write a program that implements Hough
transform.
B. Using OpenCV’s imshow function, display the hough space
C. Using OpenCV’s line function, draw the detected lines on the original image
Part 2: Harris corner detection (7.5 pts)
A. Without using OpenCV’s goodFeaturesToTrack function, write a program that implements Harris
corner detection.
B. Using OpenCV’s imshow function, display the results of the gradients (Ix, Iy, Ixy)
C. Using OpenCV’s imshow function, display the results of the corner strength or response (R)
D. Using OpenCV’s drawKeypoints and imshow functions, display the results of the final interest points.
Part 3: Feature descriptors/matching (5 pts)
A. Without using OpenCV’s matcher classes and functions, write a program that calculates SIFT-like
descriptors and matches them using the ratio-test.
B. Using OpenCV’s drawKeypoints and imshow functions, display all the keypoints.
C. Using OpenCV’s drawMatches and imshow functions, display the matched keypoints.
C.POULLIS COMP425-6341 COMPUTER VISION WINTER 2022
Part 4: Pytorch
A. Using the Pytorch API, implement a class Sobel that creates and applies the Sobel operator. Use the
Python Jupyter notebook provided here.
 Part 5: Optional for undergraduates/ Compulsory for graduates (5 pts) :
A. Make your feature descriptor contrast invariant.
B. Implement adaptive non-maximum suppression (MOPS paper). Input parameters: N=500, and c=0.9
Part 6: Extra credit (10 pts)
A. Make your feature detector scale invariant.
B. Implement a method that outperforms the SSD ratio-test for deciding if a feature is a valid match. Your
method should outperform SSD on all provided images.
Submission (electronic submission through EAS only)
Please create a zip file containing your C/C++ or Python code(2 source files (1: hough, 2: Harris + matching)
and a readme text file (.txt). In the readme file document the features and functionality you have
implemented, and anything else you want the grader to know i.e. control keys, keyboard/mouse shortcuts,
etc if applicable.
Additional Information
● The source images for Hough transform can be downloaded here.
● The source images for feature extraction and matching can be downloaded here. 

More products