  |
Email: Homepage: http://fikkert.net/research.php
Onderzoeksonderwerpen: Multimodal Interactions, Human factors, Computer Vision
|
|
| Beschrijving: |
OMSCHRIJVING
Green Dino Virtual Realities is a company that specializes in simulation and virtual reality. Their main product is a drive simulator. Using that simulator, driving schools can teach their students at low costs in a controlled and safe environment. During such a driving lesson there is no need for an instructor since the simulator boasts one virtually.
When a student is driving in the virtual world, he or she will be given instructions by the simulator’s (virtual) instructor. These instructions range from telling the direction he / she should turn to at the next crossing, to gazing in the mirrors before indicating direction and to changing gears at the correct moment. However, the desired student actions based on those instructions can not all be checked. For instance, the gaze direction of a student is unknown so the virtual instructor has no idea if the instructions are followed. Another problem is the absence of knowledge on the user’s pose in the simulator. This prevents the instructor from commenting on possible student pose errors (e.g. right hand on the transmission instead of on the wheel).
The overall goal of this project will be to improve the student’s learning experience by enhancing the information available to the drive simulator’s virtual instructor. For this, a system needs to be devised which detects the student’s gaze direction and pose, extending upon the existing drive simulator. Doing so, it must be determined if a single camera suffices, if stereo vision is needed or if an entirely different approach is best for accomplishing this task. Furthermore, it must be taken into account that the simulator has a static hardware layout on which this extension must be implemented. Also, a study must be made of costs versus accuracy, e.g. whether a cheap camera (set) may do the trick well enough. The desired goal should be implemented as a working demonstration model.
|
|
| Beschrijving: |
|
Ambient intelligence applications often use technologies from the computer vision research area to obtain information on their user(s). An existing computer vision application uses pose estimation to obtain real-time full human body poses. This paper describes how this existing application may be adapted to use stereoscopic vision in order to solve problems that persist in usage of singular camera computer vision applications such as occlusion. |
|
|
