在笔记本上安装CUDA8 & ZED Stereo Camera

ZED双目摄像头其实买到手的只是2个摄像头硬件，需要配合GPU才能运行使用SDK，记经过多次失败，成功在笔记本上运行了ZED的Demo。官方的安装步骤如下：

• Make sure the latest USB 3.0 drivers are installed.
• Install CUDA 8 and latest NVIDIA drivers.
• [Linux] Compile and install OpenCV 3.1, more information here.
• Run the ZED SDK setup tool to install the ZED driver, tools and samples.

  比较麻烦的是CUDA8的安装，另外在wiki上有详细的版本要求，支持Ubuntu16和CUDA8也是最近才有的事。

  1. 安装CUDA8

  目标系统是Ubuntu16-64bit，经过多次尝试，只有deb安装方式可行，其他的方式没有成功过。注意：最好是新系统安装，如果以前安装过NIVIDA笔记本显卡的驱动就要先卸载，安装CUDA时会自动安装相应的驱动，更方便可靠。首先是下载相应的deb文件，然后是参考安装手册安装deb文件：

  

  第一步命令补全自己下载的deb文件名称，第三步添加路径时可以在home文件夹隐藏文件.bashrc中添加下列语句（“Ctrl+H可以显示隐藏文件”），直接修改即可，需要在文件最后添加的语句，然后重启一下：

  <span style="font-size:14px;">$ export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}
  $ export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64\{LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}</span>

  之后就是按照第4步编译一个例子，看能不能运行，如果可以，表示CUDA的环境安装好了。

  
  

  2. 安装OpenCV3.1

  从OpenCV官网下载源码，一定要是opencv3.1的，先修改源码的一个BUG：opencv3中有个模块在CUDA8.0中被抛弃了，编译会不通过，下载源码后，需要在opencv / modules / cudalegacy / src / graphcuts.cpp 中修改：（也就是在红色部分添加 || (CUDART_VERSION >=8000)    ）

  

  编译方法：

  先更新系统，安装依赖项：

  <span style="font-size:14px;">sudo apt-get update
  sudo apt-get upgrade
  sudo apt-get install build-essential 
  sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev
  sudo apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev
  </span>

  然后解压下载的opencv文件，进入目录后编译并安装：

  <span style="font-size:14px;">cmake .
  sudo make -j4
  sudo make install
  </span>

  最后是构建路径：

  <span style="font-size:14px;">sudo /bin/bash -c 'echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf'  
  sudo ldconfig</span>

  编译opencv自带的samples：

  <span style="font-size:14px;">cd  xxxxxxxxx(你的opencv文件目录)/samples/
  sudo cmake .
  sudo make -j4
  </span>

  
  使用人脸检测测试一下安装的opencv，进入上述samples文件夹下的cpp文件夹：

  <span style="font-size:14px;">cd xxxxxxxxx/cpp/
  ./cpp-example-facedetect faces.jpg</span>

  
  

  成功安装，opencv！
  

  
  

  3. 安装ZED driver

  这部分很简单，下载对应的SDK，安装即可。ROS里面有ZED的wrapper：https://github.com/stereolabs/zed-ros-wrapper，安装方法见：http://wiki.ros.org/zed-ros-wrapper

  需要注意的是：安装包文件件的名字必须与节点一致（比如文件名字改成zed_wrapper），launch文件的参数列表如下：

  Parameter	Description	Value
  svo_file	SVO filename	path to an SVO file
  resolution	ZED Camera resolution	'0': HD2K
  _	_	'1': HD1080
  _	_	'2': HD720
  _	_	'3': VGA
  quality	Disparity Map quality	'0': NONE
  _	_	'1': PERFORMANCE
  _	_	'2': MEDIUM
  _	_	'3': QUALITY
  sensing_mode	Depth sensing mode	'0': FILL
  _	_	'1': STANDARD
  openni_depth_mode	Convert depth to 16bit in millimeters	'0': 32bit float meters
  _	_	'1': 16bit uchar millimeters
  zed_id	ZED Camera ID, ignore if SVO file given	int, default '0'
  gpu_id	GPU device ID, define which CUDA device will handle the computation	int, default '-1' (best device found)
  frame_rate	Rate at which images are published	int
  rgb_topic	Topic to which rgb==default==left images are published	string
  rgb_cam_info_topic	Topic to which rgb==default==left camera info are published	string
  rgb_frame_id	ID specified in the rgb==default==left image message header	string
  left_topic	Topic to which left images are published	string
  left_cam_info_topic	Topic to which left camera info are published	string
  left_frame_id	ID specified in the left image message header	string
  right_topic	Topic to which right images are published	string
  right_cam_info_topic	Topic to which right camera info are published	string
  right_frame_id	ID specified in the right image message header	string
  depth_topic	Topic to which depth map images are published	string
  depth_cam_info_topic	Topic to which depth camera info are published	string
  depth_frame_id	ID specified in the depth image message header	string
  point_cloud_topic	Topic to which point clouds are published	string
  cloud_frame_id	ID specified in the point cloud message header	string
  odometry_topic	Topic to which odometry is published	string
  odometry_frame_id	ID specified in the odometry message header	string
  odometry_transform_frame_id	Name of the transformation following the odometry	string

  最后一个ZED深度图效果图：