Universities, research centers, and labs should do research on key questions of interest to society. Most of the time, however, institutions do not take into account the actual needs of society and industry, and, thus, the development of real-life assessments that use previous research is marginalized in scientific publications. It seems that only the research itself is interesting for scientists. Many of the readers of these publications, however, are engineers, who also deal with the development of products to be used by consumers.
This is why I like this paper so much; it is a paper in a well-known scientific journal, in which the main focus is not a full description of the complete mathematical and/or statistical framework for the methods used by each module, but rather the creation of a real world application that works.
The authors’ work includes the use of robust and high confidence methods for the fusion of pictures taken from several cameras, object segmentation and tracking, background modelization and adaptive updating, and threat classification. Most of these are not today’s most popular methods (though all of them are good methods), but rather methods that were proposed several years ago, which means that they are likely to have been extensively tested.
The authors provide a lot of information about how to set up their system: specification of the camera models, the place where the camera should be installed, and so on. They present all of their examples in the context of a Minneapolis parking lot, which makes this work much more interesting.