Section: New Results
Crowd motion classification
Participants : Antoine Basset, Charles Kervrann, Patrick Bouthemy.
Important research efforts have been devoted to crowd analysis for several years [58] , [65] . We are interested in this topic for two main reasons. First, views of crowded scenes are not that different of light microscopy intracellular images. Second, the addressed problem, i.e. motion understanding, is common, and we are investigating similar data-driven methodological approaches. This a way to cross-fertilize two domains.
We address the problem of classifying coherent crowd motions in videos recorded by a fixed camera. In contrast to most existing methods, which are based on trajectories or tracklets, our approach for crowd motion analysis provides a crowd motion classification on a frame-by-frame basis. Indeed, we only compute affine motion models from pairs of two consecutive video images. The classification itself relies on simple rules on the coefficients of the computed affine motion models, and therefore does not imply any prior learning stage. The overall method proceeds in three steps: we first compute a set of motion model candidates on a collection of windows of different sizes in the image, then we select the motion model at each point owing to a ML criterion, finally we determine the crowd motion class map with a hierarchical classification tree regularized by majority votes. The algorithm is almost parameter-free, and is extremely efficient in terms of memory and computation load. Experiments on computer-generated sequences [28] and real video sequences demonstrate that our method is accurate, and can successfully handle complex situations (see Fig. 12 ).
|