Stabilization of 3D Position Measurement

Abstract

Within the particular context of ToF imaging we investigate a real-time cost-efficient filtering method for the stabilization of 3D data. The current limitation in frame rate, resolution and intrinsic depth measurement accuracy of range finding imaging systems, together with the reflective and motion properties of the objects in the scene, may lead to noisy or inaccurate depth map reconstruction. Still, in many applications such as gesture recognition or skeleton modeling and rendering, reliable and stable point location data has to be extracted from the depth map. To overcome the depth map limitations in the context of human-computer interaction, we propose a simple, fast and efficient stabilization method to filter the raw 3D data measurements or their derivatives. This filter maintains the reliability of the original measurements of an identified 3D point when smoothing the continuous change in its 3D position, avoids jerky movements without introducing noticeable latency nor impacting rapid motion.