Data-driven crowd movement simulation method based on generative adversarial network

Abstract

The invention provides a data-driven crowd movement simulation method based on a generative adversarial network. The method relates to the fields of crowd simulation and deep learning. The method is used on the basis of a pedestrian trajectory data set extracted from pedestrian movement video data. Virtual pedestrians which do not exist in a data set are generated in a simulation scene, and complete path planning closer to real pedestrian response is carried out on the generated virtual pedestrians according to given conditions such as initial positions and destinations and other factors in the whole scene. According to the method, a generative adversarial network (GAN) based on a long short-term memory (LSTM) is applied to train a simulation model. Compared with a traditional crowd simulation method based on rules, the simulation effect of the movement trail of the virtual pedestrian simulated through the method is more realistic, and the movement trail of the virtual pedestrian is closer to the movement situation of a real pedestrian. According to the invention, the trajectory planning task for the virtual pedestrian is completed, and the authenticity of the crowd movement simulation effect is effectively improved.

Description

technical field [0001] The invention relates to the fields of crowd simulation, deep learning and the like, and is especially oriented to the task of path planning for any pedestrian in a simulation scene. Background technique [0002] Crowd simulation has always had very important applications in evacuation drills, computer games, movies and other fields and has attracted researchers' research. In the past few years, researchers have proposed a large number of crowd motion simulation methods, but with the advancement of related technical fields and the maturity of new fields such as autonomous driving and express delivery robots, people have higher demands for crowd simulation. Among them, the scale, authenticity, and flexible interaction of crowd simulation are key research points. Traditional crowd simulation methods can be mainly divided into rule-based methods and data-driven methods, both of which can simulate crowd movement in the macro or micro field, but these two ...

AI Technical Summary

Problems solved by technology

[0003] Existing research on pedestrian trajectory prediction models based on generative confrontation network training is to predict the next trajectory of pedestrians by observing existing pedestrian trajectory data. It is difficult to generate virtual pedestrians for scenarios without separating from the data set, and the application scenarios are relatively single. (It is mostly used to predict the future movement of pedestrians in automatic driving)

However, rule-based simulation methods such as social force models can flexibly simulate various pedestrian trajectories, but the trajectories generated based on artificially formulated action rules are far from real human trajectories. Tracks are extremely unrealistic

Images