Training with my VLP 16 lidar

[minghaohsu410168]

Hello,

I have some question about training my own model, could you give me some afvice

1.how can I create the test.dat and train.dat with my LiDAR?
2.which point cloud feature you use?

thank you so much

[bellonemauro]

Hi,

your question is a very relevant one.
This code only gives you a way to train with "any" possible feature, generating those features is a relevant research task, though mainstream research aims to automate this task.

For learning purposes, the features can be R G B colors, corners (xyz) coordinates, 3D sift, or any other feature generated using the standard pcl library.
This code is only a part of a bigger code used to generate the features for this paper https://ieeexplore.ieee.org/abstract/document/8168386
in which we had both color features and geometric features.

I am sure that this does not solve your issue, but it might help you to follow a research line.
M.

[minghaohsu410168]

@bellonemauro,
Thank you for your reply, I am truely thankful.

I have read some paper that I know I need to define the feature by my self to train the model, like number of cluster points, intensity and so on...

I need to train the libsvm for point cloud data, but I am not very familiar with it.
If I choose number of cluster point (N>300) to be one of feature that classify human or non human, I just need to calculate the training data with the libsvm format : :, like this?
1 1:350
1 1:400
0 1:200
0 1:150

Is this understanding wrong? If there is an error, please correct me. I am new to learning machine learning.

[bellonemauro]

yes the format should be correct, just check the "0".
If I remember well the label cannot be 0, for binary you can use -1,1 (if I remember it correctly)
BTW you can see the example data in the data folder in this repo, here, https://github.com/bellonemauro/SVM_pcl_demo/blob/master/Data/example1/test.dat