Page - 11 - in Joint Austrian Computer Vision and Robotics Workshop 2020

tematic error inx could not be determined as of yet but considering the flipped signs for almost all esti- mates made by the iterative method, numeric insta- bility is likely to contribute to the fragile nature of the solvePnPclass. 5.SUMMARYANDOUTLOOK In thisworkanovel framework fordockingamo- bile robot using only vision-based sensors and algo- rithms was developed. A CNN based object detector yielded bounding boxes of logos with high accuracy and confidence. Measurements of the logos were taken and related in a coordinate system. The fam- ily of solvePnP algorithms implemented in OpenCV was used to estimate the camera pose using the de- tector results and intrinsic parameters. All methods consistently estimated wrong distances in one of the directions,namely thex-axis. Followingpreliminary experiments, changes were made, in particular the coplanarity of the object points was removed and re- calibration of the camera undertaken, and the same experiments run again. Unfortunately the errors per- sisted, although improvements regarding the scatter- ness of the pose estimates could be made. Conse- quently, no control commands were generated and docking of the robot could not take place in this in- stance. For future reference, it is important to note the fragility of the solvePnP algorithms. The source of the errors is unclear and while additional point pairs could improve results regarding compactness, it seemsunlikely theycouldalleviate the largeerrors inpredicting thexcoordinates. References [1] F. Alijani. Autonomous vision-based docking of a mobile robot with four omnidirectional wheels, 01 2017. Master’s thesis. [2] H. Bay, T. Tuytelaars, and L. Van Gool. Surf: Speededuprobust features. InEuropeanconference oncomputer vision. Springer, 2006. [3] D. Burschka and E. Mair. Direct pose estimation with a monocular camera. In International Work- shoponRobotVision. Springer, 2008. [4] K. Chen, J. Wang, J. Pang, Y. Cao, Y. Xiong, X. Li, S. Sun, W.Feng, Z. Liu, J. Xu, Z. Zhang, D.Cheng, C. Zhu, T. Cheng, Q. Zhao, B. Li, X. Lu, R. Zhu, Y. Wu, J. Dai, J. Wang, J. Shi, W. Ouyang, C. C. Loy, and D. Lin. MMDetection: Open mmlab detection toolbox and benchmark. arXiv preprint arXiv:1906.07155, 2019. [5] M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisserman. The pascal visual ob- jectclasses (voc)challenge. International journalof computervision, 88(2), 2010. [6] M. Fichtner and A. Grobmann. A probabilistic vi- sual sensor model for mobile robot localisation in structured environments. In 2004 IEEE/RSJ Inter- national Conference on Intelligent Robots and Sys- tems(IROS)(IEEECat.No.04CH37566), volume2, pages 1890–1895. IEEE,2004. [7] X.-S. Gao, X.-R. Hou, J. Tang, and H.-F. Cheng. Complete solution classification for the perspective- three-point problem. IEEE transactions on pattern analysisand machine intelligence, 25(8), 2003. [8] S. Garrido-Jurado, R. Munoz-Salinas, F. J. Madrid- Cuevas, and R. Medina-Carnicer. Generation of fiducial marker dictionaries using mixed integer lin- earprogramming. PatternRecognition, 51,2016. [9] R. Hartley and A. Zisserman. Multiple view geom- etry in computer vision second edition. Cambridge UniversityPress, 2000. [10] K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of theIEEEconferenceoncomputervisionandpattern recognition. IEEE,2016. [11] R. E. Kalman. A New Approach to Linear Filtering and Prediction Problems. Journal of Fluids Engi- neering, 82(1), 031960. [12] U. Kartoun, H. Stern, Y. Edan, C. Feied, J. Handler, M.Smith,andM.Gillam. Vision-basedautonomous robot self-docking and recharging. In 2006 World Automation Congress. IEEE,2006. [13] A.KendallandR.Cipolla. Geometric loss functions for camera pose regression with deep learning. In Proceedings of the IEEE Conference on Computer VisionandPatternRecognition. IEEE,2017. [14] V.Lepetit,F.Moreno-Noguer,andP.Fua. Epnp: An accurate o (n) solution to the pnp problem. Interna- tional journalof computervision, 81(2), 2009. [15] K. Levenberg. A method for the solution of certain non-linear problems in least squares. Quarterly of applied mathematics, 2(2), 1944. [16] D. G. Lowe. Distinctive image features from scale- invariant keypoints. International journal of com- putervision, 60(2), 2004. [17] J. Y. Lu and X. Li. Robot indoor location modeling andsimulationbasedonkalmanfiltering. EURASIP Journal on Wireless Communications and Network- ing, 2019(1), 2019. [18] B. D. Lucas, T. Kanade, et al. An iterative image registration technique with an application to stereo vision. In Proceedings of the 7th international joint conference on Artificial intelligence, volume 2. IJ- CAI,1981. [19] R. C. Luo, C. T. Liao, K. L. Su, and K. C. Lin. Automatic docking and recharging system for au- tonomous security robot. In 2005 IEEE/RSJ Inter- 11