Crowd density estimation method based on cascaded multilevel convolution neural network

Abstract

The invention discloses a crowd density estimation method based on a cascaded multilevel convolution neural network. The method includes the steps that (1) the multilevel convolution neural network is adopted to extract characteristics from lower layers to high layers, and lower layer characteristics and high layer characteristics are combined to form multistage characteristics, so that separability of crowd density characteristics is enhanced; (2) according to similarity of a characteristic pattern in a downsampling layer of the multilevel convolution neural network, connections of redundant neurons in the convolution neural network are eliminated, and the characteristic extraction speed is increased; (3) two multilevel convolution neural networks of different structures are trained according to the difficulty level of the separability of crowd density samples, the two multilevel convolution neural networks are in cascade connection according to sequences from simpleness to complexity to form a crowd density estimation model of the cascaded multilevel convolution neural network, and crowd density level estimation is rapidly carried out on to-be-detected images obtained from a video terminal in real time. In the aspect of detection accuracy, a better real-time effect is achieved compared with previous schemes.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, and more specifically relates to a method for accurately estimating crowd density in public areas in real time in a video intelligent monitoring system. Background technique [0002] With the development of computer vision technology and related hardware and the continuous improvement of human security awareness, intelligent monitoring technology based on public places has attracted more and more attention from the society, and it is also an important part of realizing a digital city. Especially for crowd management in public places, it can effectively improve personal safety, prevent group trampling incidents caused by excessive congestion, and realize reasonable allocation of public resources according to the regional distribution of crowd density. Therefore, crowd density estimation and related technologies for public places have also been widely used. [0003] The current cr...

AI Technical Summary

Problems solved by technology

[0018] This patent application uses the LBP feature of the obtained gradient direction map for texture description, and then uses ADABOOST to perform head detection to achieve density estimation. The detection results of this patent application are heavily dependent on the results of head detection. In complex scenes, due to occlusion, Many uncertainties such as angle of view, distance, etc. interfere, and it is impossible to accurately obtain the number of people, resulting in errors in crowd density estimation.

[0019] The above two traditional crowd density estimation methods cannot accurately describe the intrinsic characteristics of crowd images by using artificially designed texture features, resulting in the inability to accurately estimate crowd density. In addition, the real-time performance of crowd density estimation needs to be improved.

Images