Image-based crowd counting method and equipment

Abstract

The invention discloses an image-based crowd counting method and image-based crowd counting equipment. The image-based crowd counting method comprises the steps of inputting an image to be analyzed into a neural network comprising a multi-layer convolution layer and a feature fusion layer, performing scale normalization processing on output features of feature extraction sub-networks by means of the feature fusion layer, combining the output features after scale normalization processing with weighting coefficients corresponding to the output features to generate a crowd density map of the image to be analyzed, and finally integrating the crowd density map to determine the number of people in the image to be analyzed. The image-based crowd counting method and the image-based crowd countingequipment can count the total number of people based on one single image, have the advantages of high accuracy rate and good adaptability to the scene, can simultaneously satisfy dense crowd countingand sparse crowd counting, and can greatly reduce the workload of the statistician.

Description

technical field [0001] The invention relates to the technical field of image analysis, in particular to an image-based crowd counting method. The invention also relates to an image-based crowd counting device. Background technique [0002] With the rapid development of social economy, urban public construction is becoming more and more perfect, and the phenomenon of large crowds gathering in public places is becoming more and more serious, and the scale is getting bigger and bigger. Crowd counting is an important part of intelligent monitoring technology, which plays an important role in maintaining social stability and preventing crowd congestion and illegal gathering in public places. [0003] The commonly used dense people counting method first extracts features on the basis of foreground blocks, and then uses the features to perform regression to determine the number of people in dense scenes. Affected by the angle of view, people closer to the camera occupy more pixel...

AI Technical Summary

Problems solved by technology

[0004] 1. Relying on the running area extracted by the foreground block, it is impossible to determine the slow-moving crowd in the large scene, and the accuracy of the foreground block also affects the accuracy of the crowd count;

[0005] 2. It is necessary to construct different scene perspective maps in advance, which leads to poor adaptability of the algorithm to unknown scenes;

[0006] 3. The accuracy of sparse population statistics is low

[0013] In the process of implementing the present invention, the inventor found that the scheme divides the image into blocks and normalizes its size to 32*32, which causes the image to be deformed and the regression accuracy is poor; secondly, the scheme used The feature map for regression is the 100-dimensional feature output by the fully connected layer of the network. This feature loses position information and cannot accurately return the crowd density map. In addition, the network constructed by this scheme has only three layers of convolution, and the network is too shallow, so the obtained features are not good. , unable to accurately distinguish between the crowd and the background

Images