Controlling the Camera of 3D World by Using Real Time Face Tracking
Tomi Lamminsaari · 2012
Gestures have become very common elements of modern user interfaces as the touch screens have gained popularity in consumer gadgets. At the same time, the importance of physical buttons has diminished. For example the latest gaming consoles have controllers that can detect how players move and hold the controllers and this motion is significant part of the gameplay. Another technology that is coming more and more common is 3D graphics. Many displays provide true 3D view already and these will increase the demand for new ways to detect gestures. This master’s thesis studies different methods to track user’s head position with web camera. Located head position works as an input that changes the viewing angle of the current 3D scene. The target was to implement a class library that can be used to introduce head tracking features to existing 3D applications. The primary tool for this study was an open source software library called OpenCV. It provides effective object detection algorithms and image color space filtering functionalities that were used in face detection. This thesis studies both the object detection and the skin color based face tracking methods. Performance tests were executed for both methods. Based on those results a hybrid solution was created for camera controller’s actual implementation. Implemented camera controller can provide approximately 15 head positions per second with low-end camera and low-end computer equipment but this requires good and stable lighting conditions. In average lighting conditions the performance of camera controller drops by half. In uneven or fluctuating lighting conditions the camera controller might fail to track user’s head. Under controlled conditions, the camera controller provides quite accurate positioning for user’s head.