Dense crowd counting algorithm based on cascaded high-resolution convolutional neural network

Abstract

The invention discloses a dense crowd counting algorithm based on a cascaded high-resolution convolutional neural network, and the algorithm comprises the steps: estimating the size of a single head in a dense crowd graph, and generating a supervision prediction density graph; extracting high-resolution features of the input image by using a primary high-resolution feature extraction network HRNet; predicting a density image corresponding to the primary dense crowd by using the high-resolution feature; based on the primary high-resolution feature extraction network, constructing a cascaded high-resolution feature extraction network CHRNet to extract a secondary high-resolution feature; adopting a regional loss weighting mode, and carrying out network parameter optimization by using two loss functions of MSE and counting errors; predicting a final dense crowd density map by utilizing the second-stage high-resolution characteristics; accumulating and summing the density image pixel values to obtain a final dense crowd counting result. According to the invention, the precision of dense crowd counting is improved, and the situation of inaccurate counting caused by low resolution, crowd shielding, high crowd density and the like is effectively improved.

Description

technical field [0001] The invention relates to the field of visual crowd density analysis, in particular to a dense crowd counting algorithm based on a cascaded high-resolution convolutional neural network. Background technique [0002] Crowd counting is a fundamental and important task for many applications related to visual crowd density analysis, such as security monitoring, traffic congestion control. The goal of this task is to identify each person's head in an image and get the number of all heads that appear in the image. At present, most methods are counting methods based on convolutional neural networks, including single-stage convolutional neural network counting methods: using convolutional neural networks to extract image features, performing density prediction to obtain density maps, and completing counting based on density maps; multi-stage Convolutional neural network counting method: Design a multi-stage integrated network model, introduce a multi-stage los...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a dense crowd counting algorithm based on a cascaded high-resolution convolutional neural network to solve the problem of inaccurate counting caused by low resolution, crowd occlusion, and high crowd density in dense crowd counting

Images