Blob detection algorithm

CaraPost uses a four-step process to detect and locate blobs on each camera image. The Vicon CaraPost blob detection algorithm is based upon machine learning techniques, producing a classifier that has learned what a blob looks like, based on many examples. The classifier gives as output a probability (between 0 and 1) that a given image pixel is at the center of a blob. Vicon CaraPost has two classifiers: one to detect white blobs, and the other to detect black blobs.

Having used the classifier to detect possible blobs, CaraPost then uses image processing techniques to extract the boundary of each blob, and finally fits an ellipse to each extracted blob.

Using the blob-detection algorithm simply involves defining the range of blob sizes which should be considered, and setting a threshold on the probability above which a blob is considered to be present. Usually a probability threshold of around 0.5 will work well for a wide range of skin-types and lighting scenarios.

Tip:          Although the blob-detection algorithm copes well with very low lighting conditions, it will fail to detect blobs in parts of an image which are fully saturated (i.e. pixel values for a region are all set to 255).