Apollo: Low cost perception in standard mode

[alexbuyval]

Hi all!

I am trying to test the perception module with only a camera. I am using the r3.0.0 branch. The YOLO detector detects the car very well. However, I found that car position are published in car coordinates system (not map coordinates system). I have disable flag use_navigation_mode, however it does not work properly anyway.

Here is an example.

The detected car is front me about 30 meters. I supposed that if I use the standard mode the perception module have to convert an obstacle position from ego car frame to map frame. However, I see in '/apollo/perception/obstacles/' topic the following:

perception_obstacle {
  id: 1
  position {
    x: 35.8058547974
    y: 0.589089155197
    z: 0.0
  }
  theta: 0.143582284451

It looks like ego car frame coordinates.

In addition, in the dreamviewer the position of obstacle is invalid:

Is it possible to use a low cost perception module in standard mode?

[CCodie]

Hi @alexbuyval , I have a question.
I'm trying to do something similar to you the lowcost perception.
Is the above picture using the Offline Perception Visualizer ?
If so, which dag streaming config file did you use? (e.g. dag_camera_obstacle_vis.config, ...)
Because I'm having problems with several dag streaming config files.
Thanks !

[alexbuyval]

Hi @CCodie

Is the above picture using the Offline Perception Visualizer ?

Yes, it is.

If so, which dag streaming config file did you use?

Yes, I use the dag_camera_obstacle_vis.config

Best regards
Alex

[alexbuyval]

Hi!

It seems I found a solution.

I found the following commit in master branch. There is a transformation from camera coordinates to world frame for standard mode.
So I merged the perception module from my branch with master branch. Now the position of detected car is right.
Here is an example

[xinwf]

@alexbuyval hey, can you tell me how do you get this result? Which modules do I need to enable, which script do I need to use and where to find the output picture?

[alexbuyval]

Hi @xinwf

You need to run:

./scripts/bootstrap.sh
./scripts/perception_lowcost_vis.sh

Last script run the perception module with the visualization window.
Also check that '/modules/perception/conf/perception_lowcost_vis.conf' file has the following settings:

--dag_config_path=modules/perception/conf/dag_camera_obstacle_vis.config
--use_navigation_mode=false

And, of course, you need corresponded data in camera topic and chassis topic.

[jinjinhong]

@alexbuyval hello,I have learn apollo these days,could tell me how to add a new modules in apollo ?If you are convenient ，can you give me some examples?

[alexbuyval]

Hi @jinjinhong

It is not difficult. You can copy exist module, for instance, 'routing' and then rename folder, rename files like routing.cc, routing.h and so on. Then change name into BUILD file. Then change class name into your files and replace the code on your.

I have example new module, however for r.3.0.0 version. As I understand, old modules are not suitable for master branch

[jinjinhong]

ok, thank you.

发自网易邮箱大师

On 01/8/2019 16:38，Alexandr [e-mail] wrote：

Hi @jinjinhong

It is not difficult. You can copy exist module, for instance, 'routing' and then rename folder, rename files like routing.cc, routing.h and so on. Then change name into BUILD file. Then change class name into your files and replace the code on your.

I have example new module, however for r.3.0.0 version. As I understand, old modules are not suitable for master branch

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the thread.

[louis016]

Hi @alexbuyval,

we are trying to use the low cost perception in standard mode in our car.

The yolo camera detector works well, but when I tried to transform the coordinates from camera coordinates to world frame for standard mode, if failed.

Can you tell me how do you get the transform matrix from camera to world?