Abstract: The method of obtaining regression density map based on Convolutional Neural Network (CNN) has become the mainstream method of crowd counting and positioning, however, there are still two problems in the existing methods. Firstly, density map obtained by traditional methods has adhesion and overlap problems in crowded areas, which leads to mistakes in final crowd count and location of the network. Secondly, due to weight invariance of conventional convolutional, it is difficult to achieve adaptive extraction of image features and to process images with complex background and uneven population density distribution. To solve these above problems, a method for counting and locating dense crowds was proposed based on Pixel Distance Map (PDMap) and Four-Dimensional Dynamic Convolutional Network (FDDCNet). Firstly, a new anti-european distance map was defined, which used the spatial distance relationship between pixel level points to enhance the smoothness of pixels around the center point of human head through reverse operation, hence solving the problem of adhesion and overlap in densely populated areas. Secondly, a four-dimensional dynamic convolutional algorithm was designed to adaptively change weights in the four-dimensions of convolutional, extract the prior knowledge provided by different views to deal with the challenge of counting and positioning difficulties caused by complex scenes and uneven distribution, which improved the generalization ability and robustness of the model. Finally, the threshold value was used to filter local uncertain predicted value to further improve the accuracy of counting and positioning. On NWPU-Crowd dataset, the Mean Absolute Error (MAE) and Mean Squared Error (MSE) of the proposed method are 82.4 and 431.6, respectively. MAE and MSE of the proposed method are reduced by 8.7% and 5.8%, compared with MFP-Net (Multi-scale Feature Pyramid Network). The comprehensive evaluation F1 and accuracy P of the proposed method are improved by 3.0% and 5.9% to 71.2% and 73.6%, compared with TopoCount (Topological Count). The experimental results show that the proposed method can process dense crowd images with complex background, and achieve higher count accuracy and location accuracy.