Crowd counting method based on multi-branch deep neural network and mixed density map

Abstract

A crowd counting method based on a multi-branch deep neural network comprises the following steps: S1, marking a crowd image, generating a density map corresponding to the crowd image, and training a crowd counting model; s2, randomly cutting an image to be identified into 9 sub-images with the size of 240 * 240 according to different resolutions; s3, performing image enhancement transformation on the sub-images of the crowd image for training to obtain enhanced sub-images; s4, sending the enhanced sub-images obtained in the step S3 into a multi-branch deep convolutional network (MCNN), and identifying head images of different sizes; and S5, stacking the results obtained in the step S4, performing 1 * 1 convolutional layer processing to obtain corresponding density map mapping, and performing integration on the density map to obtain the estimated number of people. The method can effectively cope with the conditions of over-high crowd density, serious shielding and the like, and meanwhile, the density map generation mode is adjusted according to the crowd scale, so that crowd counting errors generated under the condition that crowds are too sparse can be effectively avoided.

Description

technical field [0001] The present invention relates to the fields of artificial intelligence, machine vision, and ultra-high-definition display, in particular to a crowd counting method based on a multi-branch deep neural network and a mixed density map. Background technique [0002] Stampede incidents frequently occur in large-scale activities at home and abroad, which have caused a lot of casualties. For example, the stampede incident on the Bund in Shanghai in 2015 has reached the level of major casualty accidents stipulated by our country. Therefore, the research on crowd counting is becoming more and more popular. If the crowd density of the current scene can be accurately estimated and corresponding security measures can be arranged, the occurrence of such incidents can be effectively reduced or avoided. Crowd counting specifically refers to the use of computer vision techniques to estimate the number of people in an area. Traditional crowd counting methods are mainl...

AI Technical Summary

Problems solved by technology

[0004] The main problem in the background technology is: crowded crowds often cause events such as stampedes, and serious accidents may also cause major casualties

However, there are various difficult problems in traditional crowd counting methods. Crowd counting mainly includes detection-based methods and regression-based methods. Due to the limited camera field of view and image resolution and the existence of target occlusion, the detection effect based on the detector is often not good; while the regression-based method is faced with crowded crowds, large differences in the size of the foreground head and the background head. In this case, better results cannot be obtained

The difficulty of solving the above problems and defects is: the above problems are difficult to solve by using traditional methods. For the limitations of the scope of use and image resolution, the only solution is to use video image acquisition devices with higher resolution and wider viewing angles; For the occlusion problem and the size of people in the background and foreground, traditional solutions are difficult to deal with densely overlapping crowds. Once the crowd size increases, the existing methods will produce large errors

Images