Wadaboa/cv-con-rod-inspection

Visual inspection of motorcycle connecting rods for Computer Vision class at UNIBO

Connecting rods inspection

Description

This university project is an implementation of a visual system which is able to analyze motorcycle connecting rods, by applying the following steps:

1. A grayscale image is loaded and binarized using Otsu's method
2. The image is bitwise inverted to get a white foreground and a black background
3. Morphological operations are carried out in order to detach touching rods
4. Blobs are extracted using connected components labeling
5. Blobs are filtered using a threshold on the blob's area, to remove iron powder and distractors
6. Circles are calculated using a custom contour finding method and the Haralick's circularity measure
7. Blob's moments are calculated using a custom method
8. Blob's orientation and its bounding box are determined
9. Blob's shape features, like length and width, are calculated
10. The number of holes inside each rod is computed, based on the Euler number
11. Finally, results are printed out

In the whole execution, images of the happening processing are shown for debug purposes.

Dependencies

This software is written in Python 3.7.6, using the following third-party libraries:

• plac 1.1.3, to parse from CLI "the easy way"
• scipy 1.4.1, to efficiently perform scientific computations
• opencv-python 4.2.0.32, to exploit some computer vision algorithms

Installation & execution

To install the software, just clone this repository locally.
To execute it, cd into the downloaded folder (cv-con-rod-inspection) and run

python inspection.py -i "<image_path>"

Todo

The last thing to refine is how to detach touching rods, without altering their main structure.