An Estimating Floor Region Algorithm based on Image Segmentation using FPGA for Smart Rovers

Chi-Chia Sun^{1, 2}, Nian-Jyun Yang¹

 

¹Department of Electrical Engineering, National Formosa University, Huwei, Taiwan

²Smart Machine and Intelligent Manufacturing Research Center, National Formosa University, Huwei, Yunlin,Taiwan

ccsun@nfu.edu.tw

 

Abstract: 

By far most of the floor regions detection algorithms are still dependent on the depth or other additional sensors that consume more power relatively and increase the total weigh for smart land rovers. In this paper, a new framework that enable to estimate floor regions in single image for Unmanned Ground Vehicle (UGV) robots is presented. In general case, the cluttered indoor surroundings such as patterned floors, shadows and reflections, those surroundings are very difficult to differentiate floor regions. The proposed algorithm combines extracting surface texture characteristic with specific geometric area is able to find out from object boundary, and through SVM to distinguish between floor and non-floor regions. In experimental results, public MIT Scene dataset and indoor database were selected to verify the detection accuracy. The proposed algorithm accuracy can reach up to 94.72% in average without any other sensors for assistant.

 

Keywords: 

FPGA, UGV, Floor Estimation, Image Processing, Heterogeneous Computing

pages: 

103-106

DOI: 

10.35745/icice2018v2.026

Year: 

2018

Published in: 

2nd International Conference on Information, Communication and Engineering (ICICE 2018)

Full paper download: 

103-106.pdf