Would like to detect abandoned objects from video using blob analysis

I would like to do project on detecting abandoned objects from video. I planned to implement it with blob algorithm. How can I use blob analysis to detect objects?

How can I set parameters like area change fraction, centroid change fraction?

Answer 1

[Brett Shoelson]

Sakthi, To detect change in an image (abandoned objects, for instance), you have to look at the differences at multiple timepoints. You have a reference (time = 0) image, and you compare that to images captured later. At t2, you can find differences. (Those differences don't necessarily reflect abandoned objects...yet.) At t3, if those objects still are present, then you start to suspect that they were abandoned.

Start by finding a color plane or colorspace in which the object(s) you are trying to detect are clearly distinguishable. Then implement some kind of segmentation routine. It might be as easy as thresholding (IM2BW, GRAYTHRESH) or as complex as some flavors of WATERSHED.

Then segment at multiple timeframes, and looking at the differences. (IMSUBTRACT)

Let's say you find some ROIs (blobs) that are present in frames 2 and 3 but were not present in frame 1. Use REGIONPROPS (or some other approaches like BWAREAOPEN) to filter out the noise--small pixels, for instance. Then use REGIONPROPS to calculate the areas, shapes (eccentricities, bounding boxes,...), positions (centroids) of the "abandoned objects."

Note, by the way, that once you have your segmented mask of "abandoned objects," you can pass that in to regionprops along with the original intensity image to determine other properties of the objects of interest.

Cheers, Brett

Answer 2

[Jiro Doke]

In Video and Image Processing Blockset, there is a shipping demo on Abandoned Object Detection.

You can also build up your algorithm using functions from Image Processing Toolbox. There are a number of functions that may be helpful, including regionprops.