laxnpander / OpenREALM

OpenREALM is a pipeline for real-time aerial mapping utilizing visual SLAM and 3D reconstruction frameworks.
GNU Lesser General Public License v2.1
453 stars 117 forks source link

OpenVSlam config file #69

Closed roblab2011 closed 2 years ago

roblab2011 commented 2 years ago

Hi,

how can the parameters for OpenVSlam be changed (fx,fy,cx,cy....)? Could not find the config file. Output for the example is: Camera: model: perspective setup: monocular cx: 1150.4143848 color_order: RGB fx: 1680 fy: 1680 cy: 646.6886903999999 k1: -0.00209 k2: 0.014828 p1: 0.0006579999999999999 p2: 0.000242 k3: 0 Thx

laxnpander commented 2 years ago

@roblab-wh-ge Hey, config files are in the profiles folder of the ROS package e.g., https://github.com/laxnpander/OpenREALM_ROS1_Bridge/blob/65a86d465b5ba0b5256f5c8f1bb8090ec3a027f8/realm_ros/profiles/alexa_reco/camera/calib.yaml#L6

roblab2011 commented 2 years ago

Thx for the quick reply. I had changed the calib.yaml file but got the values i showed above. I think the differences came from the factor resizing (0.6) in open_vslam_settings.yaml. What is it good for ?

laxnpander commented 2 years ago

@roblab-wh-ge It defines on what resolution you want to run the vslam. Resizing of 0.5 means half width and height of the original image. The original camera calibration is then scaled appropriately. So in the calib.yaml you have to but the full resolution parameters and if vslam is too slow you only modify the resizing.

roblab2011 commented 2 years ago

Thx, I got it. I could build a Docker images and run the examples as well as with my own data. Unfortunately i could not get any 3D information neither with the examples nor with my own data with: roslaunch realm_ros alexa_reco.launch The rviz Pointclouds topic and Depthimages seems to be empty. Can you give me a hint?

laxnpander commented 2 years ago

@roblab-wh-ge Difficult to say without seeing anything. Can you post a log of the terminal? Do you see a map being built in the examples and your custom data? Custom data is definitely tricky, because you have to follow several rules. For example your setup must be able to properly run with visual SLAM algorithms. This typically requires frame rates of at least 10 FPS and high quality cameras as well as some parameter tuning. What is your setup?

Notes: Currently the publishing of 3D point clouds is disabled, I haven't maintained this functionality in a while. See here: https://github.com/laxnpander/OpenREALM/blob/e9dce8ba669b197f35da59398fad65ae421ade64/modules/realm_stages/src/mosaicing.cpp#L262 I simply didn't need it and quality of the point cloud is also rather bad in most cases. It is, however, internally still used to rectify the images for the ortho. Also you should get an elevation map if everything works will.