Vision-based localization in campus with a multi-stage framework

In this paper, a novel vision-based multi-stage localization framework is proposed. Our framework makes rough classification scenes in a wide range of Campus, using the powerful classification ability of the convolutional neural network for multi class objects. Then, refined classification is implemented based on SIFT and GIST features, which achieves a good scene classification effect and completes the scene localization. We also establish a campus scene database with several categories of scene images collected from different road sections. Finally, experimental results tested on our campus dataset are presented to demonstrate the effectiveness and competitiveness of the proposed localization framework.