Low-complexity dense crowd analysis method based on deep learning

Abstract

The invention discloses a low-complexity dense crowd analysis method based on deep learning. The method comprises the steps of (1) sampling to obtain a picture data set; (2) labeling the picture dataset; (3) selecting a Gaussian convolution kernel, performing Gaussian convolution operation on each marking point of each picture to obtain a corresponding crowd distribution thermodynamic diagram, and counting the real number of pedestrians in each picture; (4) randomly dividing the obtained picture data set into a training set and a test set according to a proportion of 8:2, respectively takingthe training set and the test set as inputs of a training model and a test model, and carrying out offline training to obtain a low-complexity deep learning model for analyzing dense crowd distribution conditions; and (5) decoding the new crowd monitoring video stream in real time, inputting the decoded new crowd monitoring video stream into the training model, representing the crowd distributionsituation by a thermodynamic diagram output by the model, and integrating the thermodynamic diagram to realize real-time analysis of dense crowds. According to the invention, a more excellent crowd monitoring analysis effect can be obtained.

Description

technical field [0001] The invention relates to various fields such as image processing, video coding, and deep learning, and in particular, a low-complexity model scheme for analyzing dense crowds based on deep learning. Background technique [0002] In recent years, with the continuous growth of the population, the safety of the crowd has received more and more attention from all walks of life. Many public places use video surveillance technology to help security personnel manage, but the frequent stampede incidents reflect that it is a big burden to judge the safety of the crowd only by manpower. In addition, the crowd has the characteristics of strong mobility, uneven distribution, large occlusion area, and unfixed moving direction, which makes it very difficult to analyze crowd surveillance video only by manpower. [0003] With the rapid development of big data and computer computing power, Deep Learning (DL) has achieved satisfactory results in many fields. In partic...

AI Technical Summary

Problems solved by technology

Many public places use video surveillance technology to help security personnel manage, but the frequent stampede incidents reflect that it is a big burden to rely on manpower to judge the safety of the crowd

In addition, the crowd has the characteristics of strong mobility, uneven distribution, large occlusion area, and unfixed moving direction, which makes it very difficult to analyze crowd surveillance video only by manpower

Images