Multi-scale convolutional neural network-based image crowd counting method

Abstract

The invention discloses a multi-scale convolutional neural network-based image crowd counting method. The method comprises the steps of (1), generating a continuous density graph tag, and converting atagged image into a continuous estimation density graph; and (2), obtaining an accurate density graph of a predicted crowd by utilizing a convolutional neural network, setting an initial parameter for the convolutional neural network, calculating out the loss L(theta) of an input picture according to an actual density graph, and updating the parameter of the whole network in each optimization iteration, until a loss value is converged to a relatively small value. Compared with the prior art, the method has the advantages that the problem of huge scale change of the crowd in the single image is solved; based on the single convolutional neural network, features of different hierarchical networks are fused before the predicted density graph is generated, and features of different scales corresponding to different depths are extracted, so that the precision of the predicted density graph is greatly improved; and the problems of scale change, shielding and the like in the crowd image are solved.

Description

technical field [0001] The invention relates to the technical field of crowd image analysis, in particular to a crowd counting algorithm based on a multi-scale convolutional neural network. Background technique [0002] Crowd counting is an intelligent surveillance application that calculates the number of people by predicting the density map of crowd images. With the exponential growth of the world population, rapid urbanization has promoted many large-scale activities, such as sports competitions, public parades, traffic congestion and other issues leading to large-scale crowd gatherings. Therefore, in order to better manage crowds and personal safety, crowd behavior analysis algorithms are of great significance. [0003] With the continuous promotion of deep learning algorithms, the crowd counting algorithm based on convolutional neural network has greatly improved the detection accuracy compared with traditional algorithms. Algorithms based on convolutional neural netw...

AI Technical Summary

Problems solved by technology

However, because most of the crowd images are taken by surveillance cameras and high altitudes, the shooting angles vary greatly, and the size and scale of the people in the captured images vary greatly

The multi-column convolutional neural network proposed by Zhang et al. has high network complexity and large network parameters. The three-column network needs to be pre-trained and then fuse the output features of the multi-column network. It cannot grasp the multi-scale information of a single image at the same time.

Images