In recent years, the development of human-computer interaction (HCI) techniques is very fast and arises mainly from extensive work done at the computer vision fields
Abstract
In recent years, the development of human-computer interaction (HCI) techniques is very fast and arises mainly from extensive work done at the computer vision fields and supported by computational efficiency of hardware, allowing implementing these techniques real-time. In this work, we develop a system allowing a user to unlock his car door (or any other door) without having to carry a key or a remote control, using computer vision techniques. Naturally, this system may be used in many other HCI applications. We first present general algorithms aimed to deal with this problem. We then focus on the implementation, explaining in detail the rational and theory while demonstrating its performance using visualized examples with both computer simulations and hardware implementation.
Ways of Implementation
- Face recognition
- Only 1 function can be implemented (only open the door)
- Only people “known” for the system can use it (you can’t let your friends use the system without planning ahead)
- Hand tracking
- Many different patterns can perform various tasks
- You may tell your friends the password, allowing them to use it
We’ll use hand tracking
Block Diagram
Part I – Hand Detection
- Convert RGB image to YCbCr : Decoupling luminance (Y) from chrome (CbCr).
- Use a certain CbCr region corresponds to skin pixels
Part II – Hand Tracking
particle filter
Part III – Analyzing Trajectory
Project flow-chart
Difficulties
- Skin segmentation is a little sensitive to lighting conditions – Natural lighting or artificial lighting. Need to adjust parameters on a new scene.
- Improve performance and robustness (allow using T-shirt and overlapping face), along with…
- Minimizing runtime (mainly in tracking part)