Estimating the number of pedestrians in surveillanceimages and videos has important applications in intelligenttransportation systems. This problem is particularly challengingwhen the scenes are densely crowded, in which the techniques oftracking a single pedestria