Approach
The proposed tracking approach consists of a hybrid method based on computer vision and inertial technologies. The environment in the area in which the system is to be used is first modelled, in order to provide a set of images and feature points that may be used as 'beacons' by the tracking process. Depending on the application, this modelling process may make use of existing CAD models, or may build a model from images of the scene (WP5). Suitable feature points and image textures are extracted from the model and enhanced to improve the ruggedness of the tracking (WP4). Initial calibration is performed on the camera and IMU (Inertial Measurement Unit), to calibrate parameters such as the relative position and orientation of the IMU with respect to the camera, and lens distortion and focal length (WP3).
- (Fig 1) MATRIS system architecture
During tracking, images from the camera are captured live and processed by the same signal enhancement process as used on the reference scene (WP4), in order to provide a set of enhanced textures and features. The data from the IMU (WP3) is also processed to compensate for effects such as drift. At the start of the tracking process (WP6), the image and sensor data is used by the initial view registration process to calculate the initial camera pose; this may take several seconds as a search through the entire model space may be necessary. After initialisation, the predictive tracking module takes over, tracking the movement of the camera from frame-to-frame. If the predictive algorithm should fail, the initialisation process can be called again, to re-establish tracking. An overview of the data flow and the principal modules is given in the figure above.