We address the problem of calibrating an embedded depth camera network designed for people tracking purposes. In our system, the nodes of the network are responsible for detecting the people moving in their view, and sending the observations to a centralized server for data fusion and tracking. We employ a plan-view approach where the depth camera views are transformed to top-view height maps where people are observed. As the server transforms the observations to a global plan-view coordinate system, accurate geometric calibration of the sensors has to be performed. Our main contribution is an auto-calibration method for such depth camera networks. In our approach, the sensor network topology and the initial 2D rigid transformations that map the observations to the global frame are determined using observations only. To distribute the errors in the initial calibration, the transformation parameters and the estimated positions of people are refined using a global optimization routine. To overcome inaccurate depth camera parameters, we re-calibrate the sensors using more flexible transformations, and experiment with similarity, affine, homography and thin-plate spline mappings. We evaluate the robustness, accuracy and precision of the approach using several real-life data sets, and compare the results to a checkerboard-based calibration method as well as to the ground truth trajectories collected with a mobile robot.

In augmented reality, marker-based tracking is the most common method for camera pose estimation. Most of the markers are black and white patterns that are visually obtrusive, but they can be hidden from the video using image inpainting methods. In this paper, we present a computationally efficient approach to achieve this. We use a high-resolution hiding texture, which is captured and generated only once. To capture continuous changes in illumination, reflections and exposure, we also compute a very low-resolution texture at each frame. The coarse and fine textures are combined to obtain a detailed hiding texture which reacts to changing conditions and runs efficiently in mobile phone environments.

In this paper, a multimodal system for augmented reality aided assembly work is designed and implemented. The multimodal interface allows for speech and gestural input. The system emulates a simplified assembly task in a factory. A 3D puzzle is used to study how to implement the augmented assembly system to a real setting in a factory. The system is used as a demonstrator and as a test-bed to evaluate different input modalities for augmented assembly setups. Preliminary system evaluation results are presented, the user experience is discussed, and some directions for future work are given.