The recognition of collective crowd movement is crucial for crowd management in large public places. Crowd collectiveness depends on the motion of each individual and the motion state of the local crowd overall. Considering the abovementioned factors, in this study, we propose a novel crowd collectiveness recognition convolutional neural network combining global and local features. First, we generate collectiveness measurement images using optical flow vectors. Subsequently, the channel attention mechanism is applied to obtain the global crowd features, and the dilated convolution is used to obtain the local crowd features. Finally, the effectiveness of the proposed method is demonstrated by conducting comparison experiments on the public dataset. The experimental results demonstrate that the effectiveness of our recognition method is superior in the weighted average recall, weighted average accuracy, and weighted average precision.