peek

KO | EN

gitlite — search

#python #react #typescript #golang #go #android #machine-learning #rust #docker #nodejs #cli #kubernetes #security #open-source

(★ 18)

Quality Assurance System using OpenCV

peek Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Peek  
Peek is a proof-of-concept project that investigates how manufacturing quality assurance can be inspected using computer vision.

![Control and query image feature matching](https://raw.githubusercontent.com/domantasjurkus/peek/master/img/splash.jpg)

### Requirements  
* Python 2.7.13  
* OpenCV 2.4.13.2 (built-in SIFT and SURF feature detection algorithms)

### Demo  
`python main.py` - run example of alignment and difference  
`python src/align.py` - alignment example  
`python src/difference.py` - difference example

`python test.py` - run script on a number of examples under good and less good lighting conditions (ideally, the score would be ~100 for good images and ~90 for damages ones)

![Damage detection](https://raw.githubusercontent.com/domantasjurkus/peek/master/img/splash2.png)

### Approach  
For an arbitrary product image, a quality confidence score is computed by the following steps:  
1. Control and query images are loaded (control image has to be cropped to the very edges of the product)  
2. The query image is scaled, rotated and aligned to the control image  
3. Images are blurred and smoothed to minimise misalignment  
3. Images are subtracted and thresholded to produce a binary "damage map"  
4. A final confidence score is output based on damage map

#### Assumptions  
- Images should ideally be captured under similar lighting conditions  
- The camera should be fixed 90 degrees vertically above the products for inspection  
- The product is not rotated from the query image more than 45 degrees

### Nice-to-haves  
- Automatically crop control image  
- Change feature detection algorithm from SIFT (patented) to BRISK or some other free alternative  
- Update warping to rotate the query image any number of degrees  
- Detect main geometric shape of the product to remove background  
- Automatically compute blur and threshold values (color histogram comparison?)

peek

peek Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

peek

peek Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README