Improved target pose estimation algorithm based on YOLO-6D

被引：0
|
作者：
Cong M. [1 ]
Zhang B. [1 ]
Du Y. [2 ]
Li J. [1 ]
机构：
[1] School of Mechanical Engineering, Dalian University of Technology, Liaoning, Dalian
[2] School of Mechanical Engineering, Dalian Jiaotong University, Liaoning, Dalian
来源：
Huazhong Keji Daxue Xuebao (Ziran Kexue Ban)/Journal of Huazhong University of Science and Technology (Natural Science Edition) | 2023年 / 51卷 / 12期
关键词：
attention mechanism; convolutional network; deep learning; pose estimation; target detection;
D O I：
10.13245/j.hust.238897
中图分类号：
学科分类号：
摘要：
Aiming at the traditional attitude estimation algorithm's weak anti-background interference ability and poor recognition accuracy of occluded targets，deep learning was combined to propose an improved target attitude estimation model based on the YOLO 6D algorithm．The YOLO V2 detection network in the original algorithm was changed to the YOLO V3 network，and an attention mechanism was added to enhance the model's ability to detect objects with complex backgrounds and occlusions．The pose estimation method was adjusted to improve the estimation accuracy by selecting the cell group for EPnP pose estimation based on random sample consensus (RANSAC) algorithm，which was trained on LineMod dataset and tested on Occlusion LineMod dataset．According to the 2D projection index，when the distance threshold is 30 pixels，the accuracy of the algorithm in this paper is 72.30% under the Occlusion LineMod dataset． It runs at 25 frame/s on GTX2080Ti GPU with real-time processing capabilities． Its comprehensive performance exceeds other convolutional neural network (CNN)-based algorithms. © 2023 Huazhong University of Science and Technology. All rights reserved.
引用
收藏
页码：8 / 13
页数：5
相关论文
共 17 条
  • [1] BIRDAL T, ILIC S．, Point pair features based object detection and pose estimation revisited[C], Proc of 2015 International Conference on 3D Vision, pp. 527-535, (2015)
  • [2] LIU D Y，, ARAI S, MIAO J Q, Point pair feature-based pose estimation with multiple edge appearance models (PPF-MEAM) for robotic bin picking[J], Sensors, 18, 8, (2018)
  • [3] VOCK R，, DIECKMANN A，, OCHMANN S, Fast template matching and pose estimation in 3D point clouds[J], Computers & Graphics, 79, pp. 36-45, (2019)
  • [4] ZENG A, SONG S, 3Dmatch: learning local geometric descriptors from RGB-D reconstructions[C], Proc of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 1802-1811, (2017)
  • [5] PAIS G D，, RAMALINGAM S，, GOVINDU V M, 3D RegNet：a deep neural network for 3D point registration[C], Proc of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 7193-7203, (2020)
  • [6] YANG J，, LI H，, CAMPBELL D, Go-ICP： a globally optimal solution to 3D ICP point-set registration [J], IEEE Transactions on Pattern Analysis and Machine Intelligence, 38, 11, pp. 2241-2254, (2015)
  • [7] KEHL W，, MANHARDT F，, TOMBARI F, SSD-6D：making RGB-based 3D detection and 6D pose estimation great again[C], Proc of the IEEE International Conference on Computer Vision, pp. 1521-1529, (2017)
  • [8] TEKIN B，, SINHA S N，, FUA P．, Real-time seamless single shot 6D object pose prediction[C], Proc of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 292-301, (2018)
  • [9] HINTERSTOISSER S，, LEPETIT V，, ILIC S, Model based training， detection and pose estimation of texture-less 3D objects in heavily cluttered scenes[C], Proc of Asian Conference on Computer Vision, pp. 548-562, (2012)
  • [10] BRACHMANN E，, KRULL A，, MICHEL F, Learning 6D object pose estimation using 3D object coordinates[C], Proc of European Conference on Computer Vision, pp. 536-551, (2014)
12