An embedded stereovision system: Aspects of measurement precision

Computational performance of embedded systems heavily increased during last decade and tasks, which used to have a real time solution only in PC based vision systems are being ported to stand alone devices. One of the typical demanding vision tasks is 3D vision. This work describes design of a simple embedded stereo vision sensor - smart camera that can be used for spatial measurement. The system consists of a single core DSP processor and can work in several different setups because of its modular conception. The camera performs all computations itself and it is able to transmit the measurement results to a master unit using various interfaces including USB 2.0 High Speed and ethernet. Basic aspects influencing the overall accuracy of 3D position measurement are discussed in the contribution. This work is primarily focused on a simple task of object position evaluation in 3D. The problem is significantly simplified by utilizing an easily detectable object marker. Having realized the basic measurement we have shown that the overall position measurement precision nonlinearity is better than 0.5% from the full range (1000 mm). The aspects influencing the measurement accuracy and repeatability of measurement are also discussed.