传统回环检测算法大都采用手工特征来表征图像,在应对光照、视角等环境变化时的鲁棒性较差,并且特征提取比较耗时,很难满足实时性要求。针对上述问题,提出了一种改进的基于深度神经网络提取图像特征的回环检测算法。具体地,在经典NetVLAD算法中引入空洞空间金字塔池化模块,通过多尺度特征融合,在降维的同时提高了特征图的分辨率,得到更加鲁棒且紧凑的图像特征描述。在公共数据集上的实验结果表明,所提出的算法在精确率和召回率上均有一定的提升,可以较好地应对环境变化,图像特征提取耗时也有明显改善。
Traditional loop closure detection algorithms mostly utilize handcrafted features to represent images. The robustness of dealing environmental changes such as illumination and viewpoint changes is vulnerable, and the features extraction is time-consuming, which cannot meet the real-time requirements. To address these problems, we propose an improved algorithm which uses deep neural network to extract more robust image features. Specifically, the atrous spatial pyramid pooling (ASPP) module is introduced into the classic NetVLAD model to characterize the image. By the fusion of multi-scale features, the feature maps have fewer dimension and higher resolution, and thus, more accurate and compact image features are obtained. Experiments on public datasets show that the proposed algorithm has higher precision and recall rate. It can deal with the changes of illumination and viewpoint to a certain extent, and has less time cost in extracting image features.
[1] 周彦, 李雅芳, 王冬丽, 等. 视觉同时定位与地图创建综述[J]. 智能系统学报, 2018, 13(1): 97-106.
ZHOU Y, LI Y F, WANG D L, et al. A survey of VSLAM[J]. CAAI Transactions on Intelligent Systems, 2018, 13(1): 97-106.
[2] Zhang X, Wang L, Su Y. Visual place recognition: a survey from deep learningperspective[J].Pattern Recognition, 2021, 113: 107760.
[3] Tsintotas K A, Bampis L, Rallis S, et al.SeqSLAM with bag of visual words for appearance based loop closure detection[C]//International Conference on Robotics in Alpe-Adria Danube Region.Saint Paul, Minnesota, USA: Springer, Cham, 2018: 580-587.
[4] Sánchez J,Perronnin F, Mensink T, et al.Image classification with the fisher vector: theory and practice[J].International Journal of Computer Vision, 2013, 105(3): 222-245.
[5] Jégou H, Douze M, Schmid C, et al.Aggregating local descriptors into a compact image representation[C]//2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.San Francisco, California, USA: IEEE, 2010: 3304-3311.
[6] 卢宏涛, 张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理, 2016, 31(1): 1-17.
LU H T, ZHANG Q C.Applications of deep convolutional neural network in computer vision[J].Journal of Data Acquisition and Processing, 2016, 31(1): 1-17.
[7] Sünderhauf N, Shirazi S, Dayoub F, et al.On the performance of convnet features for place recognition[C]//2015 IEEE/RSJ international conference on intelligent robots and systems (IROS). Hamburger,Germany: IEEE, 2015: 4297-4304.
[8] Chen B, Yuan D, Liu C, et al.Loop closure detection based on multi-scale deep featurefusion[J].Applied Sciences, 2019, 9(6): 1120.
[9] Arandjelovic R, Gronat P, Torii A, et al.NetVLAD: CNN architecture for weakly supervised place recognition[C]//The IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, Nevada, USA, 2016: 5297-5307.
[10] Chen L C, Papandreou G, Kokkinos I, et al.Deeplab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40(4): 834-848.
[11] He K, Zhang X, Ren S, et al.Deep residual learning for image recognition[C]//The IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, Nevada, USA, 2016: 770-778.
[12] Milford M J, Wyeth G F.SeqSLAM: Visual route-based navigation for sunny summer days and stormy winter nights[C]//2012 IEEE International Conference on Robotics and Automation. Saint Paul, Minnesota, USA: IEEE, 2012: 1643-1649.
[13] Bentley J L.Multidimensional binary search trees used for associativesearching[J].Communications of the ACM, 1975, 18(9): 509-517.
[14] Malkov Y A, Yashunin D A.Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 42(4): 824-836.
