Accurate 3D human pose estimation from single images is possible with sophisticated deep-net architectures that have been trained on very large datasets. However, this still leaves open the problem of capturing motions for which no such database exists. Manual annotation is tedious, slow, and error-prone. In this paper, we propose to replace most of the annotations by the use of multiple views, at training time only. Specifically, we train the system to predict the same pose in all views. Such a consistency constraint is necessary but not sufficient to predict accurate poses. We therefore complement it with a supervised loss aiming to predict the correct pose in a small set of labeled images, and with a regularization term that penalizes drift from initial predictions. Furthermore, we propose a method to estimate camera pose jointly with human pose, which lets us utilize multiview footage where calibration is difficult, e.g., for pan-tilt or moving handheld cameras. We demonstrate the effectiveness of our approach on established benchmarks, as well as on a new Ski dataset with rotating cameras and expert ski motion, for which annotations are truly hard to obtain.

Inertial sensor drift is usually corrected on a single-sensor unit level. When multiple sensor units are used, mutual information from different units can be exploited for drift correction. This study introduces a method for a drift-reduced estimation of three dimensional (3-D) segment orientations and joint angles for motion capture of highly dynamic movements as present in many sports. 3-D acceleration measured on two adjacent segments is mapped to the connecting joint. Drift is estimated and reduced based on the mapped accelerations' vector orientation differences in the global frame. Algorithm validity is assessed on the example of alpine ski racing. Shank, thigh, and trunk inclination as well as knee and hip flexion were compared to a multicamera-based reference system. For specific leg angles and trunk segment inclination mean accuracy and precision were below 3.9° and 6.0°, respectively. The errors were similar to errors reported in other studies for lower dynamic movements. Drift increased axis misalignment and mainly affected joint and segment angles of highly flexed joints such as the knee or hip during a ski turn.

The management of product-related data in a uniform and consistent way is a big challenge for many manufacturing enterprises, especially the large ones, like DaimlerChrysler. So-called product data management (PDM) systems are a promising way to achieve this goal. For various reasons, PDM systems often sit on top of a relational DBMS, using it (more or less) as a simple record manager. User interactions with the PDM systems are translated into a series of SQL queries. This does not cause too much harm when the DBMS and PDM system are located in the same local area network, with high bandwidth and short latency times. The picture may change dramatically, however, if the users are working in geographically distributed environments. Response times may rise by orders of magnitude, e.g. from 1-2 minutes in the local context to 30 minutes and even more in the "inter-continental" context. This paper shows how a more sophisticated utilization of the (advanced) SQL features coming along with SQL:1999 can help to cut down response times significantly.