Nets for Robust Head Detection

In order to generalise about the available head-specific results for a representative number of heads, two ways seem to be promising: the one determining a minimal number of zones valid for all heads, or the other individually adapting the distances between zones for all heads. The first way is attractive, as the architecture already exists and does not need any extension. However we expect a progressive overlapping of some zones with an increasing number of heads, which may downgrade the quality of recognition compared to the 1-head case. The second way requires a new architecture which we don’t have at hand now. Both ways seem to be useful. For instance, the class of chairs does not allow for a description similar to heads. If backrest, seat and chair leg are chosen as object components, each one exists in a great variety of ways. Therefore, a description of fixed zones does not seem to work in this case.