human activity recognition in image

I am trying to implement a paper of human activity recognition, I am just started after specified the steps of the system. I found two points that I did not understand

1- calculate the normalized bounding box coordinate?!

2- If I going to use background subtraction to get silhouette, how can I get silhouette features, I found that there are many types of local features (blob, corner, region of interest) what the difference among these

Thanks a lot

Answer 1

[Walter Roberson]

regionprops() to get the bounding box. Each region will have a bounding box that is a row vector of 4 items. Divide the first and third members by the number of columns (not rows!) in the image, and divide the second and fourth members by the number of rows in the image.

[rows, cols, panes] = size(YourImage);
props = regionprops(BWimage, 'BoundingBox');
bbox = vertcat(props.BoundingBox);    %put all blob's boxes together
nbbox = [bbox(:,1)./cols, bbox(:,2)./rows, bbox(:,3)./cols, bbox(:,4)./rows];

Now nbbox will be normalized bounding box in the form [left_edge, bottom_edge, width, height] which can easily be converted to right edge and top edge instead of width and height.

[Walter Roberson]

The discussion of types of features is not a question about MATLAB. You should be researching that in other resources such as books about image processing or in research papers.

[Mammo Image]

thanks..

what is the benefit of the bounding box normalization, since I found in this paper that he did the normalization before motion detection.

Regards

[Walter Roberson]

If you had two different images of a scene taken at different resolutions, then the portion of the scene occupied by a particular object would be the same, but the number of pixels would be different. Working in portions allows you to avoid worrying about resolution.

[Mammo Image]

thanks very much Walter