High Performance 3D Sensor Network for Real-Time Reconstruction
Fast, high precision 3D measurement of people and large objects by structured light based methods remains a challenge. Reasons for this are, for example, large datasets which should be processed quickly and different reflection properties of the objects’ surfaces. Furthermore, for the acquisition of people, their motions should be compensated and the structured illumination must not disturb the person.A high-precision 3D sensor network with short latency time for 3D measurements of people and other complex objects up to a size of 2 m x 1 m x 0.5 m has been developed at Fraunhofer IOF. This network consists of several independent 3D sensors operating with structured light projection in the near-infrared (NIR) range.
Application fields of such sensor systems are the apparel industry, fitness and sports, medicine, and industrial production. The main feature of the sensor network is the short latency time of less than 200 ms from image recording to the calculated 3D point cloud. This makes the system real-time capable.
The network consists of several independent 3D sensors, each composed of a projector and two NIR cameras. The projection unit was realized according to the GOBO (graphical optical blackout) principle using aperiodic fringe patterns. The cameras have a 2 MPix resolution and can record up to 110 frames per second. The system operates at 850 nm wavelength. Using appropriate filters, the visible light can be blocked. Each sensor is calibrated separately. Afterwards the positions and viewing directions of all sensors are calibrated in a common world coordinate system with high accuracy.
Each sensor generates a 3D point cloud of object points in its viewing area. Due to the simultaneous 3D measurement by all sensors, a whole body measurement can be realized in a very short time.Figure A shows an example of the sensor arrangement. The network is suitable e.g. for whole body measurements of slow moving people for adapted machine control.
Figure B shows the 3D representation of a measured mannequin.
Authors: Christoph Munkelt, Daniel Höhne, Peter Kühmstedt, Gunther Notni