Faster R-CNN based autonomous navigation for vehicles in warehouse

Abstract

The robots in warehouse have boosted the efficiency and economic benefits of the logistic industrial chain. This paper provides a deep learning, single-camera-based solution to navigate vehicles in warehouse. Firstly we train a Faster R-CNN model to detect shelf-legs and tags in the captured image. To position the localized objects into the world coordinate, we then present a precise Inverse Perspective Mapping (IPM) algorithm. Finally, an unsupervised Support Vector Machine (SVM) algorithm is utilized to enumerate all possible paths and derive a best guiding line to navigate vehicles. The proposed solution is evaluated on real world warehouse images including various intricate situations. The experimental results prove the robustness and reliability of our work.