Closed-loop detection of mobile robot based on quaternion convolutional neural network

被引:0
作者
Zhang X. [1 ]
Zhang Y. [1 ]
Su X. [1 ]
机构
[1] School of Mechanical Engineering, Shenyang University of Technology, Shenyang
来源
Zhongguo Guanxing Jishu Xuebao/Journal of Chinese Inertial Technology | 2019年 / 27卷 / 03期
关键词
Convolutional neural networks; Loop closure detection; Quaternion; Simultaneous localization and mapping; Superpixel segmentation;
D O I
10.13695/j.cnki.12-1222/o3.2019.03.012
中图分类号
学科分类号
摘要
To improve the accuracy of closed-loop detection for mobile robot in complex environments and reduce the cumulative error of visual odometry, a closed-loop detection method based on quaternion convolutional neural network is proposed. Firstly, the superpixel segmentation is used to extract multi-scale landmarks and improve the appearance invariance and viewpoint invariance of image description. Then, the convolutional layer of convolutional neural network is extended to the quaternion convolutional layer, which can effectively increase the correlation of red, green and blue channels, extract the deep information of color images, and reflect the integrity of color images. Finally, in image similarity measurement, not only the distance of road sign is calculated, but also the shape and spatial distribution information of landmarks are considered to improve the accuracy of similarity measurement. The effectiveness of the proposed method is experimentally verified on the public datasets of Carnegie Mellon university and the University of Queensland. Experimental results show that, compared with traditional closed-loop detection algorithm, the proposed algorithm not only guarantees a high recall rate, but also improves the accuracy of closed-loop detection, achieving an average accuracy of 87.64% and 90.12% respectively on the data set, significantly improving the precision of closed-loop detection of mobile robots in complex environments such as such as illumination variations, view variations, and seasonal variations. © 2019, Editorial Department of Journal of Chinese Inertial Technology. All right reserved.
引用
收藏
页码:357 / 365
页数:8
相关论文
共 23 条
  • [11] Gao X., Zhang T., Unsupervised learning to detect loops using deep neural networks for visual SLAM system, Autonomous Robots, 41, 1, pp. 1-18, (2017)
  • [12] Xia Y., Li J., Qi L., Et al., Loop closure detection for visual SLAM using PCANet features, Neural Networks, IEEE International Joint Conference, pp. 2274-2281, (2016)
  • [13] Hou Y., Zhang H., Zhou S., Convolutional neural network-based image representation for visual loop closure detec-tion, Information and Automation, IEEE International Conference, pp. 2238-2245, (2015)
  • [14] Sunderhauf N., Shirazi S., Dayoub F., Et al., On the performance of ConvNet features for place recognition, IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 4297-4304, (2015)
  • [15] Zhu X., Xu Y., Xu H., Et al., quaternion convolutional neural networks, Computer Vision European Conference on Computer Vision, pp. 645-661, (2018)
  • [16] Greenblatt A.B., Agaian S.S., Introducing quaternion multi-valued neural networks with numerical examples, Information Sciences, 423, pp. 326-342, (2018)
  • [17] Zeng R., Wu J., Shao Z., Et al., Color image classification via quaternion principal component analysis network, Neurocomputing, 216, pp. 416-428, (2016)
  • [18] Yin Q., Wang J., Luo X., Et al., Quaternion convolutional neural network for color image classification and forensics, IEEE Access, 7, 1, pp. 20293-20301, (2019)
  • [19] Meng B., Liu X., Wang X., Human action recognition based on quaternion spatial-temporal convolutional neural network and LSTM in RGB videos, Multimedia Tools and Applications, 38, 11, pp. 26901-26917, (2018)
  • [20] Jahani H.E., Habib A.H., Puig D., An optimized convolutional neural network with bottleneck and spatial pyramid pooling layers for classification of foods, Pattern Recognition Letters, 105, pp. 50-58, (2017)