About using KITTI data

[TT22TY]

Thank you for your sharing!

I try to use KITTI sequence 05 data for testing with lidar+camera version, however, the results is strange. I saw you said the demo you provided was recorded with PandarQT with x axis point to left and y axis point to back. Does it need to change the axis of the data, or just modify somewhere in the code? Thank you very much!

[MaverickPeter]

Thanks for following our work. Can you provide a screenshot or other detailed information about the strange result? About the axis, it's only related to the point cloud filter as mentioned in README, there is no extra modification in the code. The simplest way is to comment out that filter lines.

[TT22TY]

@MaverickPeter , Thank you for your replying! The problem I encountered was the road color I built is blue in the map, which should be grey. (I set the track_point_x =0, track_point_y = 0). I solved the problem by changing line 375 in ElevationMapping.cpp from:

if(midPoint.x > 0 && midPoint.x < img.size().width && midPoint.y > 0 && midPoint.y < img.size().height)

to:

if(midPoint.x > 0 && midPoint.x < img.size().width && midPoint.y > 0 && midPoint.y < img.size().height && P_img.z() >0)

Since I use the velodyne points which are behind the vehicle by setting "track_point_x =0, track_point_y = 0", while camera does not have the behind image of the vehicle. I guess this lead to the color projection error.

By the way, I was wondering what about your plan about stereo-only version.Thank you very much!

[MaverickPeter]

@TT22TY Nice work! I just filtered out the points behind the car, so I didn't encounter this problem. About the stereo version, we use the open-source libelas for the reconstruction of the dense stereo point cloud. As the input of the GEM is the point cloud, we don't need more coding for the stereo version, only minimal configuration changes are needed. Note that, we provide various point types in our code, and one can easily define their own point type with minimal change in code.

[TT22TY]

@MaverickPeter Sounds great! Looking forward to the stereo version~

[TT22TY]

@MaverickPeter , Hi, I was just wondering how will you use the libelas, and when will you release stereo-only version, thank you ~

[MaverickPeter]

@TT22TY, For the stereo-only version, first, you may refer to https://github.com/jeffdelmerico/cyphy-elas-ros and configure your stereo camera params and get a reconstructed point cloud, and then you can simply change the input point cloud topic to the reconstructed point cloud and enable GEM. We've already made some tests in simulation and everything goes fine. You can try this plan as a temporary solution. Since the code change is trivial to build a new branch and I'm in charge of other projects in recent days, there is no guarantee of releasing the stereo-only version, sorry about that.

[TT22TY]

@MaverickPeter , Thank you very much~