High Performance, low latency 3D sensor network for live full object reconstruction

With recent advances in high speed 3D measurement sensor technologies, focus changes from merely acquiring 3D sensor data fast. An advanced application area is to fusion multiple sensor streams into a complete object representation without occlusions. Even more challenging is how to process the high speed 3D streams online, instead of the current offline processing approaches. To this end we combine our cost-effective GOBO slide-based pattern projector (GOes Before Optics) with commodity GigE vision network sensors to a multi sensor system for complete online monitoring capabilities. The targeted use-case has to deal with partial occlusions and low latency requirements for machine control. Specifically, three active NIR stereo 3D sensors are aggregated through a 10Gb-Ethernet-switch and processed by a single GPU assisted workstation. Thus a combined continuous data-stream of up to 78 million 3D points is calculated online per second out of a raw 2D data-stream of up to approximately 1250 Mb/s. The systems latency for simpler 3D analysis task, like movement tracking, is < 200 ms.