Open zhshe opened 4 years ago
If you meet a slow running, my suggestions are below:
Hi bro! I do as you say, but it's not any changes. And i find for only one gaussian process regression including 168 trainning samples, the TrainData() function run 22ms, and the prediction run 0.5ms only for one value. So, for 180 or 360 sectors and so many regression in one sectors, it take almost 8s for one cloud. What's more, i run on intel core i5 10th, 16g.
I would say that I have never seen this (8s) before. For a 15,000 points per frame point clouds, its running time is between 0.1s and 0.3s, and there is less packet loss in 10HZ frequency. My laptop is thinkpad, I7, 16G. This method is just a block in a large system https://github.com/alualu628628/Husky_Simulation.git. It also run well in this system, maybe you can find something useful in it. Besides, to algorithm complexity, LOAM is also much more complicated. I suggest you reduce the number of intervals or sample points if it is still slow. Please make sure the commands: rm -rf /build and catkin_make -DCMake=Release are completely implemented.
thanks for your answer! But i just play a ros-bag and run this node on my pc when rectifing the lidar_topic into kitti lidar topic and transforming car coordinate system to ground coordinate system. Also, i run rm -rf /build and catkin_make -DCMAKE_BUILD_TYPE=Release, it's not work. i will reading https://github.com/alualu628628/Husky_Simulation.git.
How long does it take to run with a point cloud