[1]ZHANG Jinyi,LIANG Bin,TANG Dikai,et al.Fast ICP-SLAM with rough alignment and local scale-compressed searching[J].CAAI Transactions on Intelligent Systems,2017,12(3):413-421.[doi:10.11992/tis.201605029]
Copy
CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume:
12
Number of periods:
2017 3
Page number:
413-421
Column:
学术论文—智能系统
Public date:
2017-06-25
- Title:
-
Fast ICP-SLAM with rough alignment and local scale-compressed searching
- Author(s):
-
ZHANG Jinyi1; 2; LIANG Bin1; TANG Dikai1; YAO Weiqiang2; BAO Shen2
-
1. School of Communication and Information Engineering, Shanghai University, Shanghai 200010, China;
2. Microelectronic Research and Development Center, Shanghai University, Shanghai 200010, China
-
- Keywords:
-
ICP-SLAM; rough alignment; initial pose matrix; local searching; dynamic threshold; real-time performance; cloud point; iteration
- CLC:
-
TP11
- DOI:
-
10.11992/tis.201605029
- Abstract:
-
ICP-SLAM has received much attention in the field of autonomous robots and unmanned cars. However, two deficiencies in traditional ICP-SLAM usually result in poor real-time performance. The first is the fact that the relative position between the current scan frame and the global map is not previously known. As a result, the ICP algorithm takes a large number of iterations to reach convergence. The second is that the establishment of correspondence is carried out by global searching and this requires an enormous amount of computational time. To overcome these problems, a fast ICP-SLAM is proposed. To decrease the number of iterations a rough alignment, based on an initial pose matrix, is proposed. In detail, the initial pose matrix is computed using a MEMS magnetometer and global landmarks. Then, a rough alignment is applied between the current scan frame and the global map at the beginning of the ICP algorithm with an initial pose matrix. To accelerate the establishment of correspondence, local scale-compressed searching with a dynamic threshold is proposed where match-points are found within a progressively constrictive range.Compared to traditional ICP-SLAM, under ideal stable conditions, the best experimental results show amount of iteration for ICP algorithm to reach convergence reduces 92.34% and ICP algorithm runtime reduces 98.86%. In addition, computational cost is kept at a stable level due to the elimination of accumulated computational consumption. Moreover, great improvement is observed in the quality and robustness of SLAM