January 30, 2014      

The problem with robot vision

?Robots typically lack either spatial information that is resolved in real-time by humans or the necessary focus for artificial, three-dimensional seeing. In addition, too much image information is recorded and can?t be processed quickly enough to be translated into action.?

A new solution

EE TIMES (EUROPE):?The European joint research project TACO (Three-dimensional Adaptive Camera with Object Detection and Foveation) has developed a new kind of 3D-camera system that will allow robots to perform more demanding tasks.

For the fully-functional prototype, The Fraunhofer Institute for Photonic Microsystems IPMS in Dresden contributed a novel MEMS scan technology as a key hardware component, enabling ?relevant? objects in the surroundings to be detected with a higher resolution, similar to human vision, without having to increase the volume of data.

Focusing on important vision data

Robots typically lack either spatial information that is resolved in real-time by humans or the necessary focus for artificial, three-dimensional seeing. Another issue is that often too much image information is recorded and can?t be processed quickly enough to be translated into action.

Researchers at Fraunhofer IPMS have developed an extremely compact scanning technology, dubbed LinScan, for TOF (time of flight) telemeter systems that allows a three-dimensional image acquisition with a flexible scanning rate and thus scanning with an adapted resolution.

mems camera

3D camera systems equipped with LinScan could enable future generations of robots to roughly search their surroundings for objects that appear in their visual field and to only resolve the objects they are looking for at a higher accuracy.

The robot would work with a relatively small volume of data and would still be able to gain a better understanding of its surroundings so as to better interact with everyday objects and our environment.

A precondition for the realization of this so-called principle of foveation is, however, that the robot knows what it is looking for and that it is also able to identify and interpret the objects being sought in a matter of seconds. Apart from the hardware (eye) it also needs corresponding image analysis software algorithms (brain).

What’s more, the robot should also be equipped with image sensors and software for three-dimensional seeing to gain a spatial idea of its surroundings and thus be able to navigate to objects precisely.

The novel adaptive camera system developed within the scope of European joint research project TACO relies on an optical scanner with five synchronously operated LinScan mirrors from the Fraunhofer IPMS. The MEMS scanner array guarantees the necessary receiving aperture for the Time of Flight (TOF) telemeter system of effectively 5mm and was designed for an adaptive 3D-camera system with an optical scanning range of at least 40°x60°, 1 MVoxel/s measuring rate of the TOF telemeter system with a 3mm measuring uncertainty at a measuring distance of 7.5m.

The quasi-static drive of the micro-scanner allows a line-by-line image formation with a variable refresh rate ranging from under 1Hz to 100Hz, whereby the vertical measuring point density in the relevant image section can be locally increased by reducing the scanning rate.

The horizontal image acquisition by means of the gimbal-mounted 1.6 kHz resonant micro-mirror guarantees a larger receiving aperture compared to a 2D-quasi-static drive and thus a higher resolution of the TOF telemetry with a simultaneous larger optical scanning angle of up to 80 degrees.

This new MEMS-based vision system project makes it possible for current robots to perform in more sophisticated markets so that they will play a major role in the fields of cleaning, construction, maintenance, security, health care, entertainment, and personal assistance.