A crowd counting method and system based on density estimation

Abstract

The invention discloses a crowd counting method and system based on density estimation, which relates to the field of image processing. The method includes: using the attention layer of the first scale-aware pyramid to extract scale-aware information features from the low-level semantic feature map to obtain the low-level multi-scale semantic feature map ; Use the second scale-aware pyramid attention layer to extract scale-aware information features from the middle-level semantic feature map, and obtain the middle-level multi-scale semantic feature map; use the third scale-aware pyramid attention layer to extract scale-aware information features from the high-level semantic feature map, and get High-level multi-scale semantic feature map; use feature fusion layer to fuse low-level multi-scale semantic feature map, middle-level multi-scale semantic feature map and high-level multi-scale semantic feature map to obtain multi-level and multi-scale fusion feature map; use crowd density map to estimate The layer estimates the crowd density on the multi-level and multi-scale fusion feature map, and obtains the estimated crowd density map. The invention can quickly and accurately obtain crowd counting results.

Description

Technical field [0001] The present invention relates to the field of image processing technology, in particular to a population counting method and system based on density estimation. Background [0002] Crowd counting is important for many real-world applications, such as video surveillance and crowd management. The researchers have proposed a large number of population counting methods, and the existing counting methods have mainly shifted from detection-based methods and regression-based methods to density estimation methods, which can provide spatial clues for relevant population analysis tasks. [0003] Given the complexity of population status in different scenarios, estimating the exact number of people has so far remained difficult and challenging, especially with the existence of population size variations. According to the main scale change problem, that is, the problem of the decline in the accuracy of population density estimation caused by the multi-scale image in th...

AI Technical Summary

Problems solved by technology

Introducing a large kernel pure convolution when using multiple pure convolutions will increase the number of parameters of the counting model, resulting in redundant calculations

Compared with multiple pure convolutions using different kernel sizes, multiple dilated convolutions with different dilation rates can reduce the number of parameters to a certain extent, but the processing of some continuous information and details is not satisfactory, thus leading to crowd counting inaccurate result

Images