[Reading] A Boundary-Fragment-Model for Object Detection

This paper present a system that uses a new shape representation, boundary-fragment-model (BFM), to perform object detection. In BFM, each edge fragment is associated with a object centroid location. Therefore several fragments should belong to the same object if their associated locations are close in the query image.

Besides the new idea, the whole system combines many other techniques, such as AdaBoost and Chamfer distance. The fragment is generated using growing and agglomerative clustering on medoids.

This is field is not what I'm familiar with, but I think many techniques they use are important. Also it is a good example about how to build a complete system.