Abstract：As the monocular vision can not estimate the target pose correctly, the picking objects from the targets stacked randomly is not feasible by monocular vision. Based on the point pair feature, an object recognition and pose estimation algorithm was proposed by point cloud. In the model training stage, the improved down sampling method was used to keep more distinctive point pairs, and the local reference frames of points were calculated as supplementary feature. In the scene matching stage, the distance weight was used to vote for poses, and the local reference frames of matching points were used to verify the poses. The multiple unoccluded candidate targets were selected by overlapping rate between model and scene. The results show that by the proposed method, the recognition rates can respectively reach 97% and 78% under Gaussian noise with variances of 3% and 5% times the size of model diameter. All experiments are completed within one second, which illuminates that the improved method can be used in real scenes.
YUAN H, DA F P, LIN T. Research on algorithm of point cloud coarse registration［C］∥Proceedings of the 2016 2nd IEEE International Conference on Computer and Communications. Piscataway:IEEE, 2016:1335-1339.
［2］
LI R Z, YANG M, TIAN Y, et al. Point cloud registration algorithm based on the ISS feature points combined with improved ICP algorithm［J］. Laser ＆ Optoelectro-nics Progress, 2017, 54(11):111503.
［3］
CHENG X, LI Z W, ZHONG K, et al. An automatic and robust point cloud registration framework based on view-invariant local feature descriptors and transformation consistency verification［J］. Optics and Lasers in Engineering, 2017, 98:1339-1351.
［4］
BUCH A G, KIFORENKO L, KRAFT D. Rotational subgroup voting and pose clustering for robust 3D object recognition［C］∥Proceedings of the IEEE International Conference on Computer Vision. Piscataway:IEEE, 2017: 4137-4145.
［5］
ZHOU X M, WANG K Y, FU J. A method of SIFT simplifying and matching algorithm improvement［C］∥Proceedings of the 2016 International Conference on Industrial Informatics-Computing Technology, Intelligent Technology, Industrial Information Integration. Pisca-taway:IEEE, 2016:73-77.
［6］
HINTERSTOISSER S, CAGNIART C, ILIC S, et al. Gradient response maps for real-time detection of textureless objects［J］. IEEE Transactions on Pattern Ana-lysis ＆ Machine Intelligence, 2012, 34(5):876-888.
［7］
RIOS-CABRERA R, TUYTELAARS T. Discriminatively trained templates for 3D object detection: a real time scalable approach［C］∥Proceedings of the 2013 IEEE International Conference on Computer Vision. Pisca-taway:IEEE Computer Society, 2013:2048-2055.
［8］
GUO Y L, SOHEL F, BENNAMOUN M, et al. Rotational projection statistics for 3D local surface description and object recognition［J］. International Journal of Computer Vision, 2013:105(1):63-86.
［9］
AKIZUKI S, HASHIMOTO M. DPN-LRF: a local refe-rence frame for robustly handling density differences and partial occlusions［C］∥Proceedings of the 11th International Symposium on Advances in Visual Computing. Heidelberg:Springer Verlag, 2015:878-887.
［10］
DROST B, ULRICH M, NAVAB N, et al. Model glo-bally, match locally: efficient and robust 3D object re-cognition［C］∥Proceedings of the 2010 IEEE Computer Society Conference on Computer Vision and Pattern Re-cognition. Piscataway:IEEE Computer Society, 2010:998-1005.
［11］
AKIZUKI S, HASHIMOTO M. Position and pose recognition of randomly stacked objects using highly observable 3D vector pairs［C］∥Proceedings of the IEEE Industrial Electronics Conference. Piscataway:IEEE, 2014:5266-5271.
［12］
HINTERSTOISSER S, LEPETIT V, RAJKUMAR N, et al. Going further with point pair features［C］∥Procee-dings of the 14th European Conference on Computer Vision.Heidelberg:Springer Verlag, 2016:834-848.
［13］
ABBELOOS W, GOEDEM T. Point pair feature based object detection for random bin picking［C］∥Procee-dings of the 2016 13th Conference on Computer and Robot Vision. Piscataway:IEEE, 2016:432-439.
2019, Vol. 40 
