Bilkent University
Department of Computer Engineering


Augmenting Virtual Agents in Real Crowd Videos


Yalım Doğan
MS Student
Computer Engineering Department
Bilkent University

We propose a framework for augmenting virtual agents in real crowd videos using Unity game engine. We utilize pedestrian detection and tracking algorithms to locate the pedestrians in video frames as accurately as possible to project them into our simulated environment. The navigable area of the simulated environment is to be generated from the video record using segmentation and mesh generation algorithms. We represent the real pedestrians in the video as simple three-dimensional (3D) models in our simulation environment. We feed the simple 3D models representing real agents and the augmented virtual agents into a collision avoidance algorithm (RVO) for local path planning. The virtual agents augmented into the real video feed move plausibly without colliding with static and dynamic obstacles, including other virtual agents and the real pedestrians. Each virtual agent will have personality, based on the OCEAN Personality model, which will determine its behavior and reaction to other agents in the environment.


DATE: 23 October, 2017, Monday @ 15:40