Multi-scene crowd density estimation method based on convolutional network and multi-task learning

Abstract

The invention discloses a multi-scene crowd density estimation system and method based on convolutional network and multi-task learning. It includes a crowd density map generation module, a cross-camera multi-scene learning module, and a density map calibration module for each scene. The first part of the framework is the robust density map generation module based on the convolutional neural network. The convolutional neural network is composed of three deep fusion sub-networks, and each deep fusion sub-network has three networks with different sizes and numbers of convolution kernels. Therefore, we can effectively grasp the commonality of the density estimation problem, and perform a relatively robust density map estimation for cross-camera multi-scene surveillance video frames with large differences in data distribution in background, illumination, crowd density, etc. in practical applications; the first part of the framework The second part is the data distribution learning of each scene based on multi-task learning, which learns the different crowd distribution characteristics of each scene; the third part of the framework uses the crowd distribution characteristics of each scene learned by multi-task learning to estimate the general crowd density of the first part The results are fine-tuned for calibration. In real surveillance scenarios, this system can efficiently and accurately estimate the density of crowds in multiple scenes across cameras.

Description

technical field [0001] The invention relates to a computer vision and intelligent monitoring technology, in particular to a multi-scene crowd density estimation system based on convolutional network and multi-task learning. Background technique [0002] In recent years, the phenomenon of high-density crowd gathering in cities has become more and more frequent, and stampede incidents have occurred from time to time, which seriously threatens urban public safety. Therefore, the technology of crowd control and early warning in public places has increasingly become the focus of research in the field of intelligent monitoring and urban security. Crowd density estimation technology refers to the estimation of crowd density through computer vision technology, so as to provide early warning and evacuation of high-density crowds, and has become an important technology in crowd control. [0003] There are currently solutions based on unsupervised learning methods for people counting t...

AI Technical Summary

Problems solved by technology

This method is more accurate, but the disadvantage is that it is necessary to ensure that the data of the training set and the test set are in the same distribution

However, in actual application scenarios, due to the different backgrounds of the cameras at each location, the concentrated distribution area of ​​the crowd and the density of the crowd will be quite different. Therefore, whenever a camera needs to be deployed in a scene corresponding to it, it is often necessary to collect and mark a large number of this camera. Retraining the density map regression network on the crowd images corresponding to the scene, or using the model fine-tuning method to migrate, these two mechanisms need to pay additional data collection labeling and model training costs for the deployment scene, in the actual deployment and application of a large number of cameras The cost of data labeling and model training in the process is extremely high

Images