Conclusion

Computer vision has made enormous strides thanks to advances in electronics and computing, as well as methodological advances. The former have made it possible to miniaturize sensors and computers and make them more reliable, enabling robust and sophisticated image processing. Methodological advances, particularly in geometry (modeling, calibration, multiview) and in motion-based image processing, have made it possible to use vision in robotics. In addition, vision sensors, because they are easier to embed and because they provide rich information (geometry and photometry of a scene in real time) are now integrated into a wide variety of applications. This article, after giving an overview of vision systems, geometry, calibration and primitive detection, presented several convincing applications demonstrating that vision has now become a reality in robotics. But the performance of a vision...