New-Tech Europe | April 2018

Stereo Applications Need a Dedicated Lens Choosing

FRAMOS

Stereo Vision Systems Stereo vision systems create a 3D reconstruction of the industrial environment from a pair of two- dimensional images from 2 different perspectives; usually, these perspectives are two separate sensors and lenses, or ‘stereo camera’. The selection criteria for these lenses requires additional consideration for stability and the system environment. Essentially, a stereo camera is a setup of two similar sensors and lenses separated by a distance x, referred to as the "baseline". The field of view of the two lenses overlaps at a distance y in front of the stereo camera system, depending on the focal length of both lenses. Two images are acquired from both lenses; each with a slightly different perspective. After an image of a known scene

Stereo cameras perceive the world in a manner similar to humans; this similarity allows machines using stereo technology to perform and interact like human partners. Robots equipped with a stereo camera can grasp objects autonomously, without the need for additional sensors or lights. Autonomous cars can have the ability to detect a human crossing the street and accurately determine their distance from the moving car, and their walking speed. The selection of the correct lens for a stereo application is much more important than for a setup with a single camera. The position and perspective of each camera with respect to the other determines the system’s accuracy. Environmental parameters require consideration, in addition to temperature, shock, and vibration.

has been acquired (for example, an image of a test chart for calibrating the system), the images of the two sensors can be used to calculate the depth information in the scene. Creating a Stable Setup In most cases, a high precision calibration can be provided solely in- factory; it is necessary to implement high resolving test charts to perform this level of quality calibration. After the calibration is completed, the algorithm calculating the depth information assumes that the distance between the two cameras and their orientation to each other is constant. If one of these parameters varies, it is likely that the calculation of the depth map will be inaccurate, or potentially lead to a more serious error. Therefore, the setup and alignment of the two “cameras”,

34 l New-Tech Magazine Europe