Ceiling based SLAM for Mobile Service Robot

• Main Authors: Guan-Wei Xie, 謝冠緯
• Other Authors: Kao-Shing Hwang
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/3fd9ce

碩士 === 國立中山大學 === 電機工程學系研究所 === 107 === The main objective of this thesis is to implement CV-SLAM (ceiling vision-based simultaneous localization and mapping) technique by direct visual SLAM. By using an image filter processing through images to reduce the error conduce to fluorescent lamps. This research uses the LSD-SLAM to implement CV-SLAM. LSD-SLAM is a monocular SLAM. Use two images taken from different locations and at different moments to estimate depth maps. The interference from fluorescent lamps will decrease the similarity of two pictures. We design a images kernel filter based on the idea of image gradient. Image intensity value of a pixel will be replaced by the difference of image intensity value surrounded it. Therefore, we can get a much more consistent data. So, we can overcome the flickering image’s interference while using the direct methods. We also reduce the error caused by the angle difference between images and image filter. The proposed method are tested in two different experimental environments. One is a usual office ceiling and the other is a factory ceiling. The depth variation of two environments is different. The proposed method can walk through a Circular path under the disturbance from fluorescent lamps in both environments.