Abstract | ||
---|---|---|
Semantic maps play a key role in tasks such as navigation of mobile robots. However, the visual SLAM algorithm based on multi-objective geometry does not make full use of the rich semantic information in space. The map point information retained in the map is just a spatial geometric point without semantics. Since the algorithm based on convolutional neural network has achieved breakthroughs in the field of target detection, the target segmentation algorithm MASK-RCNN is combined with the SLAM algorithm to construct the semantic map. However, the MASK-RCNN algorithm easily treats part of the background in the image as foreground, which results in inaccuracy of target segmentation. Moreover, Grubcut segmentation algorithm is time-consuming, but it's easy to take foreground as background, which leads to the excessive edge segmentation. Based on these, our paper proposes a novel algorithm which combines MASK-RCNN and Grubcut segmentation. By comparing the experimental results of MASK-Rcnn, Grubcut and the improved algorithm on the data set, it is obvious that the improved algorithm has the best segmentation effect and the accuracy of image target segmentation is significantly improved. These phenomenons demonstrate the effectiveness our proposed algorithm. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1007/978-3-030-27526-6_51 | INTELLIGENT ROBOTICS AND APPLICATIONS, ICIRA 2019, PT I |
Keywords | DocType | Volume |
Semantic map, Robot positioning and navigation, Scene segmentation, Deep learning | Conference | 11740 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xin Wu | 1 | 0 | 0.34 |
Shiguang Wen | 2 | 0 | 0.34 |
Yuan’ai Xie | 3 | 3 | 2.40 |