| Abstract |
An important issue in smart cities is intelligent mobility. In this line, the ATLASCAR project started in 2009 as a challenge to transpose to a real car many perception and navigation techniques developed for small scale vehicles in robotic competitions. Among many others, that challenge brought the need to use multiple sensors from different modalities for a richer and more robust perception of the road and its agents. The first issues concerned the proper registration and calibration of one LiDAR sensor and one visual camera, but rapidly evolved to multiple LiDARs and cameras of variate natures. In this paper, we present several calibration techniques that were proposed, developed, applied and tested along the several years of the project providing interesting practical solutions and the means and tools to merge and combine sensorial data from multiple sources. Based on the multiple detection of specific targets, like cones, spheres and traditional checkerboards or charuco boards, the techniques range from point cloud and data matching, using several techniques, including deep learning trained classifiers, up to holistic optimization techniques. The developments have also been applied in other contexts such as 3D scene reconstruction or in industrial manufacturing work cells. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG. |
