Influence of Depth and Structure of Convolutional Neural Network on Loop Closure Detection

Dongcan Zhang ； Guoliang Zhang ； Junxue Li ； Yujie Chen

《World Scientific Research Journal》 7卷6期 (2021/06) Pp. 463-472

https://doi.org/10.6911/WSRJ.202106_7(6).0059

摘要

With the increasing popularity of computer vision and deep learning, the use of convolutional neural network in loop closure detection has become more and m‐ore popular In this paper, the influence of the structure and depth of Convoluti‐onal Neural Network on the performance of loop closure detection algorithm is verified by comparing the convolutional descriptors extracted from different laye‐rs of Convolutional Neural Network. Group‐based experiments were carried out on three open data sets respectively to draw the curves of precision‐recall rate, compare performance parameters, and obtain the robustness of the convolutiona‐l image descriptors of different layers for scenes of seasonal change, lighting ch‐ange and multi-interference, so as to find out the appropriate number of netwo‐rk layers to complete loop closure detection. According to the experimental res‐ults on the data set, the experimental conclusion is drawn.

關鍵字

VSLAM ； Loop Closure ； CNN

參考文獻

Mur-Artal R, Tardós J D. Orb-slam2: An open-source slam system for monocular, stereo, and rgb-d cameras[J]. IEEE Transactions on Robotics, 2017, 33(5): 1255-1262

Sünderhauf N, Shirazi S, Jacobson A, et al. Place recognition with convnet landmarks: Viewpoint-robust, condition-robust, training-free[J]. Robotics: Science and Systems XI:, 2015: 1-10.

Camara L G, Přeučil L. Spatio-semantic conv-net-based visual place recognition[C]//2019 Europ-ean Conference on Mobile Robots (ECMR). IEEE,2019: 1-8

Merrill N, Huang G. Lightweight unsupervised deep loop closure[J]. arXiv preprint arXiv:1805.07703, 2018

Zhou B, Lapedriza A, Khosla A, et al. Places: A 10 million image database for scene recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(6): 1452-146