Computer Vision - CV

A collection of four CV application:

• Camera calibration
• Image Denoising
• Object recognition
• Plate recognition

The scripts are written in C++ with OpenCV 4.1.x.

Inside every model's folder there is PDF with the full description of what I do and why from a theoretical point of view.

Here a brief description for every project:

Camera calibration

Given a set of photos (all from the same camera) of a checker-board the required tasks are:

• Detects the checker-board in every image;
• Calibrates the camera;
• Compute the mean projection error;
• Show the name of the images with the best and worst re-projection error;
• Un-distort and rectify a test image acquired with the same camera as the calibration set.

Test image without correction.

Test image after correction

Image Denoising

Given a photo, the required tasks are:

• Part 1:
  • Print the histograms of the image (R,G,B);
  • Equalize the RGB channels and show the new image;
  • Change the colour space and try to equalize a single channel. Show the results.
• Part 2:
  • De-noise the result of the last picture generated from part1;
  • The filters to try are: median, Gaussian and bilateral filters.
• Part 3:
  • Try to remove the electronic cables from the "barbecue picture" using morphological filters.

Original image and its RGB histograms.

Effect of a custom morphological filter.

Object recognition

Given a set of objects photos and scenes pictures made from these objects, the required tasks are:

• for every object and scene couple, find and highlight the object in the scene;
• display the result as a figure with the object and the scene pictures side by side; the scene photo should have the found object in a coloured square.

Dataset 2, Three dimensional computer vision is detected.

Dataset 4, the big bear with a red papillon keeping only the 10% of the best matches.

Plate recognition

Given a set of photos of cars, it is required to develop an application in c++ able to:

• detect the licence plate;
• read the various single characters in the licence plate. Given the length of the task, two classes have been implemented:
• ExamClass: does the preprocessing of the input image, detects and extracts a plate;
• ExamClassPlates: detects and extracts the single characters from a plate.

The program has been tested only on the given dataset.

a), b), c), d), e), f), g) are the detected and extracted plates after preprocessing.

e), f), g) final results.

For the image recognition part, I've used the OpenCV_3_License_Plate_Recognition_Cpp.

For the papers, the license used is: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.