Maybe you remember me some days ago talking about research and object recognition. I finished it and had to present the theoretical background and practical working example. What can I say? I worked the night before my examination until seven o’clock in the morning to get it working. Without a written speech, sleep or something else helpful I talked about my program. The result was 14 of 15 possible points which is quiet good.
Now about the Imageprocessing part. I wrote a program that gets structural information from an obstacle you hold into the camera. The title is “Objectrecognition of presented obstacles” and this is done using a motion-detection for the movement. A scene background-substraction is done to get only the moved obstacle. Then there is a median-colour lookup after a lof of filters have been applied. Closing, Retinex and Blobcounter to mention only some of them.
The last final part which I like most is the artificial neuronal network that recognizes the shape of the obstacle. I used a simple feed forward net and the results were also quiet good.
You can read the full article here. I am sorry but it is not written in English so you can only read if you understand German.