Use realsense d435

[SeungRyeol]

I am using RealSense d435. So I changed code of the "darknet_3d.yaml" file as follows.

darknet_ros_topic: /darknet_ros/bounding_boxes
output_bbx3d_topic: /darknet_ros_3d/bounding_boxes
point_cloud_topic: /camera/depth/color/points
working_frame: camera_link
mininum_detection_thereshold: 0.3
interested_classes: ["person", "chair"]

Here is launch file.

<launch>

  <!-- Config camera image topic  -->
  <arg name="camera_rgb_topic" default="/camera/color/image_raw" />

  <!-- Console launch prefix -->
  <arg name="launch_prefix" default=""/>

  <!-- Config and weights folder. -->
  <arg name="yolo_weights_path"          default="$(find darknet_ros)/yolo_network_config/weights"/>
  <arg name="yolo_config_path"           default="$(find darknet_ros)/yolo_network_config/cfg"/>

  <!-- ROS and network parameter files -->
  <arg name="ros_param_file"             default="$(find darknet_ros)/config/ros.yaml"/>
  <arg name="network_param_file"         default="$(find darknet_ros)/config/yolov2.yaml"/>

  <!-- Load parameters -->
  <rosparam command="load" ns="darknet_ros" file="$(arg network_param_file)"/>
  <rosparam command="load" file="$(find darknet_ros)/config/ros.yaml"/>
  <param name="darknet_ros/subscribers/camera_reading/topic" type="string" value="$(arg camera_rgb_topic)" />

  <!-- Start darknet and ros wrapper -->
  <node pkg="darknet_ros" type="darknet_ros" name="darknet_ros" output="screen" launch-prefix="$(arg launch_prefix)">

    <param name="weights_path"          value="$(arg yolo_weights_path)" />
    <param name="config_path"           value="$(arg yolo_config_path)" />
  </node>

  <!-- Start darknet ros 3d -->
  <node pkg="darknet_ros_3d" type="darknet3d_node" name="darknet_3d" output="screen">
    <rosparam command="load" file="$(find darknet_ros_3d)/config/darknet_3d.yaml" />
  </node>
</launch>

This is the node graph(Nodes/Topics (all)).

darknet works fine, but the /darknet_ros_3d/bounding_boxes topic does not have any information.

How can I solve this problem?

[fgonzalezr1998]

I don't see the problem with the information provided. rqt_graph is ok and darknet_3d.yaml too. Have you changed something of darknet_ros_3d besides the yaml file?

[SeungRyeol]

No. I only modified the yaml config and launch files. I do not know what is the problem.

[SeungRyeol]

I found the following message in the middle of the terminal:

person: 60%
person: 57%
person: 32%
person: 57%
person: 32%
Failed to find match for field 'rgb'.
terminate called after throwing an instance of 'std::out_of_range'
  what():  vector::_M_range_check: __n (which is 65051840) >= this->size() (which is 271048)
person: 61%
person: 62%
person: 62%
person: 59%
person: 31%
person: 59%
person: 31%
person: 64%
person: 64%
[darknet_3d-4] process has died [pid 30971, exit code -6, cmd /home/shin/catkin_ws/devel/lib/darknet_ros_3d/darknet3d_node __name:=darknet_3d __log:=/home/shin/.ros/log/fcb614b8-2dbb-11ea-8605-4cedfbc5f8af/darknet_3d-4.log].
log file: /home/shin/.ros/log/fcb614b8-2dbb-11ea-8605-4cedfbc5f8af/darknet_3d-4*.log
person: 63%
cup: 34%
person: 57%
person: 57%

My guess is that I need to modify the code of the Darknet3D.cpp file for realsense, but I don't know how to fix it. How can I solve the problem??

[fgonzalezr1998]

You should not need to change Darknet3D.cpp code. Can you tell me which ros package are you using like realsense driver for see the messages, please?

[SeungRyeol]

https://github.com/IntelRealSense/realsense-ros

https://github.com/IntelRealSense/librealsense

here thank you

[fgonzalezr1998]

I've seen at realsense-ros documentation that if you want points cloud be published, you have to run rs_camera.launch with a specific parameter. I don't know if you did this. Probably, this is the mistake. darknet_ros works fine because it doesn't use pointcloud but darknet_ros_3d crashes because it need it.

[fmrico]

Hi @SeungRyeol ,

I have not a d435 right now. I have one in my office but until next week I can't access because of the holidays.

Meanwhile, you could check (use rviz), in this order:

1. Is the camera publishing images? check the topic name and the frame id (visualize It!!)
2. Is the point cloud being published? check the topic name and the frame id (visualize It!!)
3. Is darknet_ros publishing the bounding boxes? Check that the topic is /darknet_ros/bounding_boxes and the content with rostopic echo

Check darknet_ros/config/ros.yaml if something went wrong

If everything is ok until this point, the problem is in darknet_ros_3d. Then, check all topics in darknet_ros_3d/config/darknet_3d.yaml. Check also the classes. It should fit with the ones in /darknet_ros/bounding_boxes. Maybe you don't have base_footprint if you are not using a robot. If so, change it for camera_link.

In your conf, I have seen /camera/depth/color/points. Is this topic really correct?

Tell me if you found the error. In any case, I will try next week to reproduce your problem with the same camera.

Best Francisco

[SeungRyeol]

Running the launch file of the RealSense camera with rs_rgbd.launch instead of rs_camera.launch solved the problem. Really thank you. @fmrico @fgonzalezr1998

[fmrico]

Hi @SeungRyeol

Yes, I am testing it right now. with rs_camera.launch I am not able to see the point cloud. It seems not to be dept_registered.

I am happy to see that you made it work!!

Enjoy Darknet_ros_3d!!!

[hannes56a]

Hi all, i have currently nearly the same problem as SeungRyeol. I´m tring to run darknet_ros_3d with a Realsense D435. Darknet_ros finds objects and publishes the 2d boundingboxes. The Realsense publishes the depth_registered pointcloud, so this seems to be the correct pointcloud. But at least there are no data in the 3D boundingboxes :-(

When I have a look to the bounding_boxes in the rqt topic monitor I´m wondering why the empty 3D bounding_boxes needed the doubled bandwith than the 2D bounding_boxes... is that normal?

I started with ROS two weeks ago, so i´m not very confirm with it...

Can please someone help me finding my problem!? That would be great!

Here are some screenshots of my configuration. If there is something missing, please tell me!

THANKS A LOT!

Greetings, Hannes

[fgonzalezr1998]

Hi @hannes56a , rqt_graph seems correct. If you are using Intel realsense, make sure that you are launching rs_rgbd.launch. Also, darknet_ros_3d will only detect the objects that are in the objects list specified at darknet_3d.yaml. You can change it to detect a TV Monitor or any other object (make sure that objects names you put in the list are the same that YOLO detects). But, probably you have some type of problem in nodes communication because there are no TF messages and some warnings are advertising that Darknet ROS nodes are not connected.

One thing you can try is launching a rostopic echo to see if messages are being received correctly by darknet_ros_3d or publish darknet_ros bounding boxes manually using rostopic pub. In this way, you can try out if nodes are communicating with each other correctly. NOTE: All these trials you must do it with camera driver (rs_rgbd.launch) and darknet_ros_3d.launch running.

Respect to the Bandwith: Darknet ROS's Bounding Boxes are being published to 1 Hz approximately. However, Darknet ROS 3D's Bounding Boxes are published to 10 Hz and they contain more data, so even if I have never seen its bit rate, I think that it is not something abnormal that bit rate is greater.

[hannes56a]

hi @fgonzalezr1998, you are the man! Your hint with the detection objects was the solution. I always tested the 3d bounding_boxes with the tvmonitor. And this object was only in the ros.yaml and not in the darknet_3d.yaml. Now i added it, and it works fine.

Thank you!

[mhaboali]

Thanks, @fgonzalezr1998 for hints, they were very useful.

I encountered a strange issue here, the darknet_ros_3d one time works and 10 times doesn't work. Could you help with that?

I'm running the system exactly as you explained and I tried to publish some dummy 3d bounding boxes msg but didn't get a success.

[fgonzalezr1998]

Hello @mhaboali First of all, what I would do is to run Intel RealSense driver (using rs_rgbd.launch) and visualize with rviz if depth_registered_points are seen well Second: run darknet_ros. When you see that darknet_ros works well and bounding boxes are been received in the proper topic, you can kill it. Third: Review config/darknet_3d.yaml and sure that all parameters are correct (input topic, that is the darknet_ros output topic; point cloud topic that would be registered like I mentioned before; working frame, that would be camera_link or any other frame whose axes are equals; and minimum_probability, because if the probability is smaller than this value, darknet_ros_3d won't count it). Fourth: Open launch/darknet_ros_3d.launch and, in the fourth line (config camera image topic) maybe you have to change the RGB camera topic. This is the topic where darknet_ros are going to subscribe. I think that it is /camera/rgb/image_raw for Intel RealSense camera. Fifth: run launch/darknet_ros_3d.launch without run darknet_ros before. This launcher launches darknet_ros and darknet_ros_3d nodes.

• Arrived at this point, everything should work properly. If you have any problem in any of these steps, answer telling me what was the last step you checked successfully and what is the step where you had the problem to be able to trace better the mistake. However, I will record a video tutorial soon, showing all this configuration and how it works.

[mhaboali]

Hi @fgonzalezr1998 thanks so many for your clear steps, they were very useful for me. Now it works :D and the issue came when I used darknet_ros_3d_gdb.launch

I have one more question and please let me know if I have to open a new issue for it. How can I track those 3d boxes?

Your help is highly appreciated!

[fgonzalezr1998]

Yes @mhaboali , open a new issue and I will answer you there, please

[MyongjinKim]

Hello, @fgonzalezr1998 thanks to your code I could approach my goal some steps. But I faced some problems.

I'm trying to run darknet_ros_3d with stereo camera(ocams_1cgn) made by withrobot and I really want to get 3d coordinate values like xmin ymin or something. My camera finds some objects and shows me them with 2d boxes on the yolo window. Also, it has pointcloud info.

However, there are no 3d boxes and distance info as well.......

I tried to change some code to suit my camera path, but I think it is not perfect.

Could you tell me what I'm wrong? or missing? I'm a super beginner in ros and computer science.....

I'll upload some pics such as my rqt_graph, ros yalm and .....

THANK YOU VERY MUCH

-Myoungjin

[fgonzalezr1998]

Hi, @MyongjinKim I've seen that you have two different topics that publish sensor_msgs/PointCloud2. Try using stereo/point_cloud topic.

NOTE: You don't need to change parameters in the code because they are read from the YAML file. You only need to change darknet_3d.yaml and launcher file to change the RGB image topic as you do it.

Also, try to visualize point cloud with rviz. You can launch rviz typing rosrun rviz rviz to make sure it is working properly.

May useful to use the following commands for tracking: rosnode list, rosnode info, rostopic list, rostopic info and rostopic echo.

IMPORTANT!! You must launch darknet_ros_3d using roslaunch darknet_ros_3d.launch, not rosrun because otherwise config file will not used. If you use rosrun you have to use additional argument for passing the YAML file.

I'll be waiting for your feedback! ;)

[MyongjinKim]

Thanks for fast review :-) I tried to change cloud topic to stereo/point_cloud but it didn't work..... : ( And I also tried rostopic echo /darknet_ros_3bounding_boxes then I've got this error messge

• [ [ERROR] [1594374501.762555340]: "ocams_1cgn" passed to lookupTransform argument target_frame does not exist. Failed to find match for field 'x'. Failed to find match for field 'y'. Failed to find match for field 'z'. Failed to find match for field 'rgb'. ]

If have you ever seen the above message ???

I think the problem is that /darknet_3d can't get '/tf' info so it can't calculate the distance.... If you know any good idea please let me know.

Thanks a lot.

Myoungjin

[fgonzalezr1998]

@MyongjinKim exactly! It means that there is not tf between both frames. Without tf between point cloud frame and working frame, darknet_ros_3d can not calculate 3d coordinates.

you can see all available frames adding TF in rviz. If there is some TF whose axes are the same that camera_link (X-axis aiming to front, Y-axis aiming to left, and Z-axis aiming to top) you can use it changing the working frame in darknet_3d.yaml. In this case, the 3d coordinates will be calculated concerning this new frame.

However, if there is not any TF whose axes have these characteristics, you can publish by yourself the transform between two frames using static_transform_publisher where you indicate the translation and rotation that exists between both frames.

I recommend you that if you are going to use this camera more times, yo do you custom package with a launcher that includes static_transform_publisher and darknet_ros_3d launcher.

If you have any other hesitate, no doubt in ask it

[MyongjinKim]

Oh, finally I can catch the distance value!!!! I did rostopic echo /tf_ and I got a frame_id : "base_link" And I changed yaml file. Thank you so much @fgonzalezr1998

[MyongjinKim]

Lastly, may I ask you something? I got these values but I can't understand what exactly they mean.

It is difficult to understand them because they are not clearly described on your 'Read.me' page. Could you explain them in more detail?

THANKS A LOT

-Myoungjin

[fgonzalezr1998]

darknet_ros_3d takes as input the 2d bounding boxes and returns 3d bounding boxes. It means that each detected object will have an imaginary box around. Xmin is the nearest X coordinate and Xmax the farthest X coordinate. The same with Y and Z. So, you don't have (x, y, z) coordinates. You will have (x, y, z) maximum and (x, y, z) minimum. I represent that in the following way: (xmin, xmax, ymin, ymax, zmin, zmax). This is like that because these coordinates are not 3d coordinates of only one point but they are 3d coordinates of the volume around the object.

For a better understanding, you can visualize these 3d bounding boxes adding MarkerArray on rviz and subscribing to darknet_ros_3d/markers topic.

When you look output 3d bounding boxes what you can see is an objects array where each object has its 3d coordinates, its class name, and its probability of certainty.

I hope my explanation was useful @MyongjinKim

[MyongjinKim]

Thank you for your detailed explanation. Actually, I'd like to use my own weight and cfg files, so I put them in the cfg and weight folder, modified some files like darknet_ros_3d.launch or something but it didn't work. and I got below error message.

[ INFO] [1594611813.531696317]: [YoloObjectDetector] Node started. [ INFO] [1594611813.535463653]: [YoloObjectDetector] Xserver is running. [ INFO] [1594611813.536719577]: [YoloObjectDetector] init(). YOLO V3 First section must be [net] or [network]: Resource temporarily unavailable [darknet_ros-1] process has died [pid 18872, exit code 255, cmd /home/hotshot/catkin_ws/devel/lib/darknet_ros/darknet_ros __name:=darknet_ros __log:=/home/hotshot/.ros/log/f8c0a4b8-c4ba-11ea-9585-080027300d94/darknet_ros-1.log]. log file: /home/hotshot/.ros/log/f8c0a4b8-c4ba-11ea-9585-080027300d94/darknet_ros-1*.log

Could you tell me what I should fix???

I'm sorry to keep asking you about the little things. Thank you so much. @fgonzalezr1998

-Myoungjin

[fgonzalezr1998]

If you want to use other weights, you can change it between all weights you have available in darknet ROS Github page or you can re-train the neural network. This is something that I did some time ago and I recommend you follow these steps. @MyongjinKim

[MyongjinKim]

Hi @fgonzalezr1998 it's Myoungjin Finally I combined my weight file with darknet_ros_3d. The problem was cfg file. So now I can get these 3d box's information, but I am still wondering how long the object is far from my stereo camera. In order to calculate linear distance from the base, do I calculate sqrt((xmin+xmax)/2+(ymin+ymax)/2+(zmin+zmax)/2) ?? And do you think does it look right?

Thanks a lot

Myoungjin

[fgonzalezr1998]

First of all, 3D bounding boxes provide min and max coordinate in each axis. So, if yo want to get the Euclidean distance between the camera and the detected object, it is a good idea to take the center of the object.

In this way, the coordinates of the point would be X=(Xmin+Xmax)/2.0, Y=(Ymin+Ymax)/2.0, and Z=(Zmin+Zmax)/2.0. Once calculated this point, you know that the object distance vector is V = (X, Y, Z). So, you anly need to calculate the module of this vector: D = sqrt(X^2 + Y^2 + Z^2).

On the other hand... if you anly want to know how far is located the object in X axis, you anly have to do: D = (Xmin+Xmax) / 2.0

I want to make clear that 3D bounding boxes provides to you the min and max distances in the three coordinates where object is located. with this information, you know all you could need to know because you not only know the Euclidean distance. Youcan know the volume of the object and its position in the 3d space

@MyongjinKim I hope I solved your doubts

[MyongjinKim]

Hello @fgonzalezr1998 it's been a long time I just saw your readme file and I wanna ask you something Should I install ROS2 to use your package??? or is there any way to use it by using ROS melodic?? Regards Myoungjin :))

[fgonzalezr1998]

Hi @MyongjinKim we have 3 branches. melodic, ros2_eloquent and master. The last one contains the same that ros2_eloquent. So, if you use ROS2 (Eloquent), you must clone the ros2_eloquent branch; and if you use ROS1 (Melodic), you must clone the 'melodic' branch.

[MyongjinKim]

@fgonzalezr1998 Oh I c thanks a lot And could I ask one more thing about ROS ?? I really eager to use the calculated x,y,z, values using ur package and I'd like to show its center values on my terminal also I wanna send it different code as a string or float value. Can I make it ?? If yes could you give me some tips??? Super Regards Myoungjin Kim