ROS visualization of 3D target detection

Laboratory has a radium godC16的激光雷达,Recently in my hand,Want to to play.Original intention is to do a real time detection,通过ROSLaser radar for laser point cloud,用pointpillarsModels to real-time 3 d target detection task.But god radiumc16That a laser radar,不太好处理,Now can only use their own drive,In a real-time display.So thought is at the bottom of the,Storage of radium god beforec16的点云数据,再用ros进行三维目标检测.

1. 思路：

  通过rosTo realize the point cloud news订阅和检测框bbox消息的发布,To achieve the function of a testing results visualization.

2. 实验环境：

• Ubuntu16.04
• cuda 10.1
• ros-kinetic
  Testing environment mainly depends on theOpenPCDet,Environmental structures, refer to this article I博客

3. 步骤：

1. 前提条件

已经搭建过了OpenPCDet.
已经搭建了ROS环境,我用的是kinetic版本,With other versions should also can.

2. 编译环境

Create a workspace and into the

mkdir -p ~/pointpillars_ros/src
cd pointpillars_ros/src

将ROSPackage copy or clone to the current folder

git clone https://github.com/BIT-DYN/pointpillars_ros
cd ..

编译

# To build a goodopenpcdet环境
conda activate openpcdet
pip install --user rospkg catkin_pkg
pip install pyquaternion
# 因为我用的是kinetic,所以需要安装pcl-ros,We don't have to install the above version
sudo apt-get install ros-kinetic-pcl-ros
sudo apt-get install ros-kinetic-jsk-recognition-msg
sudo apt-get install ros-kinetic-jsk-rviz-plugins
## 注意,执行catkin_make时,In the outermost folder,Also is my example of~/pointpillars_ros
catkin_make

迁移OpenPCDet中的一些文件

为保险起见,I will be here all these files into thesrc/pointpillars/tools文件夹下了.

修改代码

ros.py代码

# Below the path to your own
sys.path.append("/home/ubuntu/pointpillars_ros/src/pointpillars_ros") # Line 22
# 54行后边
""" Initialize ros parameters """
config_path = rospy.get_param("/config_path", "/home/ubuntu/pointpillars_ros/src/pointpillars_ros/tools/cfgs/kitti_models/pointpillar.yaml")
ckpt_path = rospy.get_param("/ckpt_path", "/home/ubuntu/pointpillars_ros/src/pointpillars_ros/tools/models/pointpillar.pth")
# Subscribe to the laser point cloud name,改成自己的,我用的是kitti的.bag文件
self.sub_velo = rospy.Subscriber("/kitti/velo/pointcloud", PointCloud2, self.lidar_callback, queue_size=1,  buff_size=2**12)

launch/pointpillars.launch代码

<launch>
# Major changes under the first line
  <node pkg="rosbag" type="play" name="player" output="log" args="-l /media/ubuntu/ximing/dataset/ros_kitti/bag/2011_10_03/kitti_2011_10_03_drive_0027_synced.bag" />
  <node name="pointpillars_ros" type="ros.py" pkg="pointpillars_ros" output="screen"/>
  <node type="rviz" name="rviz" pkg="rviz" args="-d $(find pointpillars_ros)/launch/pointpillars.rviz" />
</launch>

launch/pointpillars.rviz代码
Major changes topic name you use,I change the name of the point cloud and image

3. 运行

执行下边的命令

conda activate openpcdet
source ~/pointpillars_ros/devel/setup.bash
roslaunch pointpillars_ros pointpillars.launch

4. 图片效果：

我使用kittiThe original data of a certain period of time,转成.bag格式,用pointpillarsThe model test,效果也不行,猜测原因可能是1.Rotation matrix parameters didn't change;2.The show laser point cloud with test box delay,Not a good match;3.The third point may be not model(可能性不大).
Below is to use the thorc16Radar acquisition in the lab30秒的数据,不到1分钟,696MB.

目前用rosbag recordSome stores the thorC16激光雷达的数据,Also not too will use radar real time read,实时检测.Late again to find a way to.

5. 视频效果

哔哩哔哩

6. 主要参考资料:

  Thank you very much for the share of the open source data,Let a small white I,也能快速demoAn effect.

• ROS点云的Pointpillars实时目标检测
• Github资料

遇到的问题,解决办法：

• error while loading shared libraries: libopencv_core3.so.3.3
• ROS运行python报错：/usr/bin/env: ‘python\r’: No such file or directory