Deep Learning Based Place Recognition for Challenging Environments

Deep learning;place recognition;CNN

Kumar, Devinder

University of Waterloo

关键词: Deep learning;    place recognition;    CNN;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/10691/3/Kumar_Devinder.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

Visual based place recognition involves recognising familiar locations despite changes inenvironment or view-point of the camera(s) at the locations. There are existing methodsthat deal with these seasonal changes or view-point changes separately, but few methodsexist that deal with these kind of changes simultaneously. Such robust place recognitionsystems are essential to long term localization and autonomy. Such systems should beable to deal both with conditional and viewpoint changes simultaneously. In recent timesConvolutional Neural Networks (CNNs) have shown to outperform other state-of-the artmethod in task related to classi cation and recognition including place recognition. In thisthesis, we present a deep learning based planar omni-directional place recognition approachthat can deal with conditional and viewpoint variations together. The proposed methodis able to deal with large viewpoint changes, where current methods fail. We evaluate theproposed method on two real world datasets dealing with four di erent seasons through outthe year along with illumination changes and changes occurred in the environment acrossa period of 1 year respectively. We provide both quantitative (recall at 100% precision)and qualitative (confusion matrices) comparison of the basic pipeline for place recognitionfor the omni-directional approach with single-view and side-view camera approaches. Theproposed approach is also shown to work very well across di erent seasons. The resultsprove the e cacy of the proposed method over the single-view and side-view camerasin dealing with conditional and large viewpoint changes in di erent conditions includingillumination, weather, structural changes etc.