The invention discloses a grid unit characteristic-based crowd massing behavior detection method which comprises the following steps: extracting a crowd foreground from video input by adopting a mixed Gaussian model; performing median filtering processing on the crowd foreground; dividing the crowd foreground into grids taking an m*m pixel rectangular unit grid as a minimum unit; performing foreground pixel counting on each unit grid in the grids to obtain a grid unit characteristic graph; counting the unit grids of which unit grid characteristic values are 1 in the grid unit characteristic graph; performing global searching on the grid unit characteristic graph by virtue of a search box, and calculating two-dimensional probability density; calculating a two-dimensional joint distribution entropy; calculating a crowd massing behavior reference index; judging a crowd massing behavior, and if G is more than G0, determining that the crowd massing behavior occurs and giving an alarm. According to the method, the crowd massing behavior can be effectively detected, and the method can be widely applied to public places with higher requirements on public safety such as squares, stations and airports.