Seite - 83 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Figure3: The twoexperimentsand thedistancesbetweenourcamerapositionsandOpenMVG’s. The bins of the histogram show the number of distances between camera centers in the respective range [cm]. Figure4: The reprojected laser point from thecomputedcameraposes. 6. ConclusionandOutlook In this paper, we presented a first evaluation of our idea to utilize a combination of RGB camera and LRF for sparse camera pose estimation, where we can select significant features during the image recording process with the LRF. We showed that metrically accurate camera pose estimation with just a few correspondences is possible. In future work, we plan a more sophisticated way to redetect significant features in the other views without the use of SIFT matches and homography estimation, which would also enable the reconstruction of more complex 3D structures. Additionally, we plan to improve theaccuracyby robustifyingourapproachagainstoutliersduringbundleadjustment. An interesting direction for future work is inspired by Li et al. [8], where they address the SfM problem by estimating up-to-scale edge lengths of a rigid graph constructed from 3D features and their respective image rays. We would like to investigate whether the laser range measurements can bedirectlyused as edge lengths for this approach. References [1] S.Agarwal,N.Snavely, I.Simon,S.M.Seitz,andR.Szeliski. Buildingromeinaday. In ICCV, pages72–79, Sept2009. [2] H. Aliakbarpour, P. Nunez, J. Prado, K. Khoshhal, and J. Dias. An efficient algorithm for extrinsic calibration between a 3d laser range finder and a stereo camera for surveillance. In ICAR, pages1–6, June2009. [3] H.Aliakbarpour,K.Palaniappan,andG.Seetharaman. Faststructurefrommotionforsequential andwideareamotion imagery. InComputerVisionWorkshop(ICCVW), pages1086–1093,Dec 2015. 83