Crowd counting is widely used in video surveillance, public security, intelligent business and many other fields. In recent years, with the continuous development of deep learning, crowd counting has become one of the hot topics in the field of computer vision. In this paper, according to the different feature extraction methods, crowd counting is divided into two categories: one is traditional method, the other is based on deep learning method, and the method based on convolutional neural network is analyzed and introduced. Further introduces the population count in the field of benchmark data sets and other representative data sets, the experimental results show that the larger changes in the crowded and scale, based on the convolution of the neural network method is superior to the traditional method, the scale change is bigger, more complex scenarios crowd more columns than a single network count more accurate, more effective; Finally, the future development direction of the algorithm is discussed.