Video Refined Agent-based Simulation
Hao Wu, Jiaohao Yu
Credits:
Data Collection, Alogirthm Design, Programming
Keywords:
Agent-based Simulation, Artificial Intelligence, Computer Vision
Agent-based model (ABM) serves as a crucial method for simulating human behavior and movement in virtual environment, relying on algorithms to abstract and model the actions and interactions of human entities into behavioral trees. However, current algorithm-based simulation, used for human movement in built environments, lacks variability and exhibit consistent biases. This research addresses this limitation by proposing a novel optimization method, Noise Plotting, to introduce variability into the simulation process, resulting in more realistic human behaviors for agents. The Noise Plotting method involves recording video data of human movement patterns in an indoor space, utilizing computer vision technology to analyze the movement patterns of agents. The traced agent patterns are then applied to a digital twin of the space, incorporating interest points for more accurate representation. To evaluate the effectiveness of Noise Plotting, movement patterns generated by the algorithm-based simulation are compared with those derived from video data. Absolute deviation is calculated as noise, providing a metric for optimizing future simulations based on real-world data. The research emphasizes the importance of validity and accuracy in simulating human behavior.
This research has been selected in the 16th International Conference on Environment-Behavior Studies, CEB- ASC, Nanjing, China, 2024
Illustration of Alglorithm
Final Result