Automatic Synchronization of Wearable Sensors and Video-Cameras for Ground Truth Annotation -- A Practical Approach

The common practice of manual synchronization of body-worn, logging accelerometers and video cameras is impractical for integration into everyday practice for applications such as real-world behavior analysis. We significantly extend an existing technique for automatic cross-modal synchronization and evaluate its performance in a realistic experimental setting. Distinctive gestures, captured by a camera, are matched with recorded acceleration signal(s) using cross-correlation based time-delay estimation. PCA-based data pre-processing makes the procedure robust against orientation mismatches between the marking gesture and the camera plane. We evaluated five different marker gestures and report very promising results for actual use.