Abstract: A large number of shopping crowds in Beijing shopping malls are prone to crowded and trampled accidents, making it necessary to determine the location of high-density population areas. Therefore, population density maps are introduced to determine the distribution of populations in the map, and the location information of indoor populations is obtained. Firstly, the collected crowd video is divided into image frames and divided into training sets and test sets. Then, the human head label processing is performed on the training set image, and the ground truth density map is generated as the training data of the improved multi-column convolutional neural network algorithm and the model is generated, and the model is applied to the test set image to generate the population density map. Finally, use ArcGIS to georeference the population density map and the indoor plan to achieve high-density population positioning. The results show that the position coordinates of the high-density area determined by the population density map are basically consistent with the actual coordinate values, indicating that the new idea of applying the population density map to indoor localization is feasible.

Key words: mall data set, population density, MCNN, geographical registration, indoor localization