Top view image target counting method and system and storage medium

Abstract

The invention provides a target counting method for a top view image, which is characterized by comprising the following steps of: S1, acquiring the top view image and generating an image data set; and S2, processing the image data set by using a geometric adaptive Gaussian convolution method. And S3, constructing a regression counting neural network model. And S4, inputting the real density graph set as a training set into the regression counting neural network model for supervised training to obtain a trained regression counting neural network model. And S5, processing the to-be-counted image by using a geometric adaptive Gaussian convolution method. And S6, inputting the real density map of the to-be-counted image into the trained regression counting neural network model. And S7, integrating the prediction density map of the to-be-counted images to obtain the target number of the to-be-counted images. The invention provides a target counting system for a top view image. The invention provides a target counting storage medium for a top view image.

Description

technical field [0001] The invention relates to a target counting method, system and storage medium of an overhead image. Background technique [0002] With the acceleration of urbanization, the number of road vehicles has increased, and the urban road system has become more complex, making the relationship between vehicles and roads more difficult to handle. Accurate vehicle counting and understanding of road vehicle information are of great significance for the construction of modern road systems and the redistribution of road traffic resources. However, the commonly used vehicle counting based on intersection cameras has the problem that the target is easily blocked and the direction is single. The bird's-eye view image obtained by using optical remote sensing technology and UAV aerial photography technology contains a wider spatial range, and can directly monitor a large range of roads to obtain the overall information of the road. In addition, it has the characteristics...

AI Technical Summary

Problems solved by technology

However, the current two methods have the following defects in the counting of vehicles facing overhead images: on the one hand, the general method lacks the adaptability to the target scale change, and the effect is not good in the counting of multi-scale target images; On the other hand, the general method has poor counting stability for target dense areas

Images