Search:

Robotics I Fall 2007

Robotics II Spring 2007

Robotics I Fall 2006

Code

# Lab2

Border detection

One of the most interesting applications of computer vision is related to the possibility to obtain information from pixel data. In computer graphics we use some primitives to describe what it should be later pictured as a virtual object. In the other hand, if what we like to analise is a picture, then we have to identify those primitives or features, that appear in the image.

One method to extract important features of an image is done by detecting high intensity contrast in a region. In this case the feature is going to possibly be a border. As we have seen in Lab1, color filtering can be as well used to determine those regions.

The Sobel Filter

An example of edge detector is the famous and well known Sobel Filter

Starting code

for( y = 0; y < height; y++ ) { for( x = 0; x < width; x ++)

``` {
Sum_X = Sobel_In[y-1][x-1] + 2 * Sobel_In[y][x-1] +
Sobel_In[y+1][x-1] -(Sobel_In[y-1][x+1] + 2 *
Sobel_In[y][x+1] + Sobel_In[y+1][x+1]);
Sum_Y = Sobel_In[y-1][x-1] + 2 * Sobel_In[y-1][x] +
Sobel_In[y-1][x+1] - (Sobel_In[y+1][x-1] + 2 *
Sobel_In[y+1][x] + Sobel_In[y+1][x+1]);
Sum = (abs(Sum_X) + abs(Sum_Y));
}
```

}

Mission

Do you remember the principles of Pose Determination explained in the introduction ? Well, once you have already design algortihms to filter colors and find edges, the next question is : How can determine the pose of an known object in the image ? Suppose that we like to find the position and orientation of the next brick in the above images.

We can visually recognize that such an object of interest has a prominent color and a very specific geometry. It contains 9 holes equally separated and the same radii. Even if we are going to see in detail some algorithms related to the 3D pose computation that requires what is known as calibration, there is quite a funny tricks one can apply to obtain such a pose without calibration: Single View Metrology is quite a new and interesting research are.

References

Another famous Edge detector is the Canny Edge Detector. In the last years the vision community have been developing diverse methods to synthetise images to relevant information. Particulary, when dealing with sequences of images - video - one is interested in found those "interest points" that continuously appear in such a sequence. On of the most famous algortihm to search distinctive features is the SIFT created by D. Lowe.

A very nice explanation of the general top-down/botton-up approaches for segmentation is found here

Video Filtering in the Blackfin : Here you will find some ideas on how to optimize a 2D filter.