Skip to Main content Skip to Navigation
Theses

Optical flow-based navigation algorithms for virtual characters

Axel López 1, 2
2 RAINBOW - Sensor-based and interactive robotics
Inria Rennes – Bretagne Atlantique , IRISA-D5 - SIGNAUX ET IMAGES NUMÉRIQUES, ROBOTIQUE
Abstract : Simulation of real humans in virtual environment is an open problem with many applications, from the entertaining industry to safety in architectural planning. Humans poses a complex visual system capable of navigation in highly dynamic environments. This thesis attempts to reproduce human behavior by imitating the human visual system. We designed an agent model capable of perceiving visual information and define a control model from it. In particular, we focus on the usage of optical flow, in other words, the apparent motion of objects. Two control loops are proposed, the first one operates with synthetic optical flow. It is an approximation available in virtual simulations that allows our agents to perceive this map with high precision. The second control loop computes optical flow from a sequence of color images using a numerical solver. This reproduces better the human perception system but the resulting flow loses accuracy. A novel characteristic of the latest control loop is the possibility of reacting to different lighting conditions. Finally, we evaluate our model by testing our agents in many scenarios with static and dynamic environments.
Complete list of metadatas

Cited literature [89 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02519441
Contributor : Abes Star :  Contact
Submitted on : Thursday, March 26, 2020 - 9:18:09 AM
Last modification on : Wednesday, September 9, 2020 - 4:12:14 AM
Long-term archiving on: : Saturday, June 27, 2020 - 12:30:43 PM

File

LOPEZ_Axel.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02519441, version 1

Citation

Axel López. Optical flow-based navigation algorithms for virtual characters. Computer Vision and Pattern Recognition [cs.CV]. Université Rennes 1, 2019. English. ⟨NNT : 2019REN1S069⟩. ⟨tel-02519441⟩

Share

Metrics

Record views

83

Files downloads

62