Processing time is too long

[shanyuejun]

===STEP 578===Time used: 79.19370 ms===
Pose Odom:x: 241.04177 y: 13.40081 z: -2.28776
[ INFO] [1723014374.617594517]: PointCloud seq: [0]
after filter:27973 points. time: 26.56799
t_get_pcl:26.60267ms
t_gp:6.24446ms
t_gp:6.17847ms
t_gp:5.73657ms
t_gp:5.60209ms
t_gp:5.66493ms
t_gp:6.18845ms
num_cells: now:0 new:168 glb:45240
mesh_msg: point 81216 faces: 11771
t_visualize: 2.10524ms
t_overlap_region: 6.23964ms
t_match_points : 14.72014ms
t_gp : 35.69519ms
t_rt : 11.83215ms
t_update : 1.98488ms
t_draw_map : 2.11401ms
===STEP 579===Time used: 76.85505 ms===
Pose Odom:x: 241.20769 y: 13.71667 z: -2.31015
[ INFO] [1723014374.721056856]: PointCloud seq: [0]
after filter:28019 points. time: 27.95666
t_get_pcl:27.98500ms
t_gp:6.58582ms
t_gp:6.03398ms
t_gp:5.62858ms
t_gp:5.67413ms
t_gp:5.80328ms
t_gp:5.86475ms
num_cells: now:0 new:76 glb:45316
mesh_msg: point 81216 faces: 11986
t_visualize: 1.83716ms
t_overlap_region: 6.04469ms
t_match_points : 15.45551ms
t_gp : 35.64439ms
t_rt : 11.36213ms
t_update : 2.68575ms
t_draw_map : 1.84110ms

How to improve running speed?

[RuanJY]

The parameter will influence the runtime and accuracy. Please let me know your parameters.

Here are the influences of different parameters in KITTI. Usually, If you can increase the cell size grid or decrease the register_times, the run time will decrease significantly.

[RuanJY]

I abbreviated this part in the 6 pages conference paper.

[shanyuejun]

slamesher:
use_sim_time: true
grt_available: false
odom_available: false
read_offline_pcd: false
dataset: 1
max_steps: 10000
save_mesh_map: true
#register param
range_max: 60
range_min: 3.0
register_times: 5
cross_overlap: true
cross_cell_overlap_length: 1
num_margin_old_cell: 500
point2mesh: true
residual_combination: true
#visualize parameter
meshing_tsdf: false
full_cover: true
visualisation_type: 3
#gp param
num_thread: 8
grid: 1.2
min_points_num_to_gp: 3
num_test: 6

variance_register: 0.60
variance_map_update: 0.50
variance_map_show: 0.20
variance_min: 1.36
variance_sensor: 0.02

test_param: 0

[RuanJY]

register_times: 2
visualisation_type: 2
grid: 1.6

In this setup, the process will be much faster, sacrificing some pose robustness. SLAMesh is a pure-lidar odometry, the robustness is an issue if the movement is aggressive, and a tightly coupled IMU will be very helpful.

[shanyuejun]

register_times: 2
visualisation_type: 2
grid: 1.6

In this setup, the process will be much faster, sacrificing some pose robustness. SLAMesh is a pure-lidar odometry, the robustness is an issue if the movement is aggressive, and a tightly coupled IMU will be very helpful.

Thank you.