2021-10-19 00:52:28 +00:00
|
|
|
|
# 基于scan context和CNN的重定位研究
|
2021-10-19 00:52:07 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
#1-Day Learning, 1-Year Localization: Long-Term LiDAR Localization Using Scan Context Image
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
2019 ICRA
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
##SCI localization framework
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
##SCI
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
通过实验验证了与单通道图像训练相比的一点改 进。提出的sci比sc具有更高的分辨能力，是一种更适合cnn输入的格式。这个过程如图所示。我们注意到，进一步研究单色图像或彩色地图选择的网络调谐可以提高定位性能。
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
Matlab使用colormap Jet 可以将灰度图像生成彩色的热度图，灰度值越高，色彩偏向暖色调。相反亦然。
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
##CNN选择
|
|
|
|
|
|
cnn网络
|
|
|
|
|
|
|
|
|
|
|
|
输入：
|
|
|
|
|
|
1.N-way SCI Augmentation
|
|
|
|
|
|
2.热编码向量指示类别
|
|
|
|
|
|
输出：
|
|
|
|
|
|
scorevector
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
##Unseen place
|
2021-10-19 00:52:12 +00:00
|
|
|
|
|
2021-10-19 00:52:28 +00:00
|
|
|
|
|
