some words before...

semantics are necessary to build bigger and better SLAM systems.
Robotics is the killer application of SLAM
Will end-to-end learning soon replace the mostly manual labor involved in building today’s SLAM systems?.
-> Integrating semantics into SLAM is often talk about, but it is easier said than done. Moreno's PhD thesis: Dense Semantic SLAM

SLAM (construction) <---help---> Deep Learning (perception) large-scale "correspondence engines" -> large-scale datasets

调试过程： 1. 确认摄像头是或否工作，运用cheese 2. 安装ros下的库uvc_camera，确认摄像头所对应设备号/dev/video* 3. 在uvc_camera下新建launch文件夹，再新建uvc_camera_node.launch文件，复制以下代码 4. 运行roscore 5. 运行rosrun lsd_slam_viewer viewer 6. 运行roslaunch uvc_camera uvc_camera_node.launch 7. 运行rosrun lsd_slam_core live_slam /image:=image_raw _calib:=~/ROS_DEV/rosbuild_ws/package_dir/lsd_slam/lsd_slam_core/calib/FOVxxx.cfg