MOVEMENT PARAMETERS DETERMINATION OF EVACUATION FLOWS USING ARTIFICIAL NEURAL NETWORKS

Abstract

Introduction. Despite the availability of modern software simulation complexes, the evacuation participant's movement parameters for these complexes are determined by establishing dependencies between the flow density and the movement velocity of the participants. These dependencies can be detected mainly by the results of field observations and video processing of these observations, which takes a lot of time, requires a lot of effort and indicates the need to automate and optimise the video processing process.
It is necessary to note that the tools for image analysis and classification, as well as the detection and classification of moving objects in the video stream, can be successfully used to study fire evacuation problems.  The purpose of the article is to develop a conceptual model of a software system using artificial neural networks for the formation of movement parameters empirical databases based on video flow analysis.
Methods. The solution to the problem of evacuation participant’s classification by mobility groups using machine learning methods is described, in particular the method of backpropagation based on the gradient descent algorithm.
Results. The work presents a conceptual model for determining the movement parameters of evacuation flows based on the data from video surveillance cameras using an artificial neural network and the SORT algorithm. A convolutional neural network architecture is proposed for solving the problem of evacuation participants' classification by mobility groups. It is represented by 2 convolutional layers, 2 Max Pooling layers, two hidden fully connected layers and an output classification layer. A model for determining the movement velocity of evacuation participants based on the SORT algorithm is given.
Conclusion. The implementation of the proposed conceptual model allows obtaining the values of the evacuation participant's instantaneous movement velocity with their storage in a separate file, which significantly speeds up the process of empirical movement parameter database forming. The model can be very useful for further scientific studies of evacuation processes with mixed human flows.