Final Report of FDUROP: 'The development of an inaccurate graph isomorphism algorithm and its application in prototype machine'


This program(FDUROP) aims at giving undergraduate students with scientific interest and aptitude an opportunity to launch an individual project under the supervision and guidance of capable mentors. My project is 'The development of an inaccurate graph isomorphism algorithm and its application in prototype machine'. This project is profound and sophisticated, which requires basic knowledge in pattern recognition, machine vision, artificial intelligence, electronic design, mechanic and auto-control, but with the help of my mentor and funded by FDUROP, I have finished the project and will have an achievement exhibition at the end of the year. Some part of my achievement has been applied to a new visual examination system and exhibited in Shanghai Industrial Exhibition Show.

This research lasts more than one year, from Oct, 2013 to Nov, 2014. During that time, I have done much work in the related area.

Finding out the development status of automatic industry.

Learning basic knowledge of computer vision and image processing. 

Learning to use OPENCV (Windows, linux and DSP edition) 

Using DSP and raspberry embed system board.

Learning to control a robot and made a robot arm from scratch.


Detailed introduction to this large project is demonstrated in series of articles.

Articles link:

1. A matlab code for detection of tiles on production line

2. Tile detection algorithm using opencv

3. Tile detection algorithm on linux platform

4. Noodles package detection using opencv

5. EMCV Guide —— opencv for DSP

6. Real test on square or circle object determination


The following is a brief summary:

The system has three versions, each marking a step of success.


The first version is to find the contour from a tile production line, telling location and rotation of each tile.

blob.png


testing picture contour capture process in Debian

The 1.5 version is to move the system to embed system. Linux system on raspberry(an embedded computer) and DSP 642

blob.png

raspberry (connected to a camera)

blob.png

DSP with camera and monitor


The second version is to make the system stronger to complicated environment, including partially overlap, and dirty marks.

blob.png

simulating a noodles production line with comlicated environment

the vision system is still able to capture noodles package accurately(marked in colorful rectangles)


The third version not only needs to find contour, but also tells its shape(square or circle). And the robot arm is added to perform the consequent action(simulating the situation when different productions are sorted by robot arm in a production line).

determine square and circle and put them to different baskets using robot arm

blob.png

The robot arm I assembled


Although, unfortunately, I met with a enermy of my tutor in the final presentation, who criticized casually and unscientifically, as a result I did not rank first, I still felt so satisfied, for the knowledge I learned, for the skill s I mastered, for the opportunity I met many people who helped me during this research. So I never regret to spend a whole year, including summer and winter vocation to do this reserch with countless time and energy with my whole heart.


Final report 成果汇报书.docx

Final representation ppt:机器视觉图匹配算法.pdf

Video of demonstration version 3: http://youtu.be/UnEwTq3et98

http://www.tudou.com/programs/view/ixN_rFdjFcY


youtube video:


Last Article Next article

Comment 评论



Share 分享

New Users 最新加入

  • hokurikustr

  • refrain

New comments 最新评论

test123: aasdas Details Apr 13 16:39
admin: Thanks! Details Apr 09 11:46
admin: Google map api Details Apr 09 11:46
lqj12: cooooooooool Details Apr 08 21:34
Yunhan Huang: 这个功能是如何实现的? Details Apr 08 13:23