A Target Density Estimation Method with Attention Mechanism

Abstract

The invention discloses a method for estimating target density with an attention mechanism, which relates to the field of image processing and includes: generating a corresponding preset target coordinate table; based on the preset target coordinate table, generating the real value density map; generate a global attention area map based on the true value density map of the preset target corresponding to each training image; input the training image and the true value density map of the preset target in the training image into the parallel density prediction network and Attention area network, output the preset target density prediction map and attention area map; merge the attention area map and the preset target density prediction map to obtain the final preset target density feature map; expand the final preset target density feature map into the final The preset target prediction map is used in the present invention to solve the problem that it is difficult to estimate the number of high-density targets in an area based on the detection method in the existing method, and to provide an accurate basis for the application based on the number of targets.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an object density estimation method with an attention mechanism. Background technique [0002] At present, target monitoring applications are mainly based on target recognition technology. These detection methods usually circle the target with a detection frame, and count the number of targets by counting the number of detection frames. These methods are used in environments with high target density and many occlusions. The recognition effect is poor, some targets cannot be circled by the detection frame, and many targets will be missed during statistics. The common convolutional neural network is a single-column network. When the target object has a large perspective zoom in the image, it is difficult for the single-column neural network to learn all the features. At the same time, existing neural networks usually require the size of the input image, so that these networks need ...

AI Technical Summary

Problems solved by technology

[0003] The present invention provides a target density estimation method with an attention mechanism, which is used to solve the problem that the existing method is difficult to estimate the number of high-density targets in an area based on detection, and provides an accurate basis for the application based on the number of targets

Images