I did research in the field of Computer Vision at the University of Central Florida under Dr. Mubarak Shah and his student Haroon Idrees.
For my research project, we worked on developing a new algorithm for tracking in crowds. In this new method, we build on top of a preexisting tracker, and therefore, our method can be used with any tracker. However, we assume that a confidence value is returned by the tracker, as this is a critical step for our contribution. The first step is to select the prominent people in the crowd, who are called Queen Bees. These people will be tracked first. The second step is to update all members in the crowd in a hierarchical order, starting with the Queen Bees and the expanding outward toward to the neighbors. Each person is updated using information based on their neigh- bors who have already been updated in the current frame. The final step is to adjust for the error in the current frame. If the returned confidence value is low for a person, that person will be updated a second time using more information. With these three steps, we are able to vastly improve the results of tracking in dense crowd situations.