Export tool memory optimization

[matlabbe]

Refactoring:

• Optimize memory usage to be able to export large maps:
  • Data disappears from point cloud when exporting at low decimation value #1368
  • When exporting large point clouds, RAM usage can be too high and app exits #1375
• Add support to export landmark poses. See Difference in poses export between GUI and rtabmap-export script. #1387 (comment)
• Add option to export Full iterative poses (like in DbViewer) See rtabmap-export (Global Full vs Global iterative) #1398

New options:

--opt #               Optimization approach:
                            0=Iterative Global Optimization (default)
                            1=Full Global Optimization
                            2=Use optimized poses already computed in the database instead
                              of re-computing them (fallback to default if optimized poses don't exist).
                            3=No optimization, use odometry poses directly.
--density_radius #    Filter poses in a fixed radius (m) to keep only one to be exported in the assembled cloud.
--density_angle #     Filter poses up to angle (deg) in the --density_radius.

Example the export a large map in multiple files.

1. Make sure the database has optimized poses saved in the database with:

rtabmap-export --poses --save_in_db rtabmap.db
Opening database "rtabmap.db"...
Opening database "rtabmap.db"... done (0.000244s).
Optimizing the map (Iterative global optimization)...
Optimizing the map (Iterative global optimization)... done (6.975819s, poses=85934).
Saved 85934 poses to database! (min=[-105.723869,-15.019793,-2.311474] max=[167.144485,251.769775,14.714050])

2. Look at the min/max of the poses, we will use them to adjust the ranges in that script:

#!/bin/bash

# Define ranges
min_x=-110
max_x=170
min_y=-20
max_y=260

divide=4

if [ $# -ne 1 ]; then
  echo "Usage: $0 DATABASE_PATH"
  exit
fi

database_path=$1

interval_x=$(( (max_x - min_x) / divide ))
interval_y=$(( (max_y - min_y) / divide ))

echo "min_x=$min_x max_x=$max_x interval=$interval_x meters"
echo "min_y=$min_y max_y=$max_y interval=$interval_y meters"

db_name=$(basename "$database_path")
db_name="${db_name%.*}"
        
for (( x=$min_x; x<$max_x; x+=$interval_x )); do
    for (( y=$min_y; y<$max_y; y+=$interval_y )); do
        x_end=$(( x + interval_x ))
        y_end=$(( y + interval_y ))
        echo "Export Range x=($min_x, $x_end) y=($min_y, $y_end)..."
        output_name=${db_name}_${x}_${x_end}_${y}_${y_end}
        rtabmap-export --cloud --xmin $x --xmax $x_end --ymin $y --ymax $y_end --voxel 0.05 --decimation 4 --ascii --opt 2 --output $output_name $database_path
    done
done

The script uses --opt 2 to make sure we use always the same poses between the exports. The resulting files would look like this with the ranges in the file name:

rtabmap_-110_-40_-20_50_cloud.ply
rtabmap_-110_-40_120_190_cloud.ply

rtabmap_-40_30_-20_50_cloud.ply
rtabmap_-40_30_50_120_cloud.ply
rtabmap_-40_30_120_190_cloud.ply
rtabmap_-40_30_190_260_cloud.ply

rtabmap_30_100_-20_50_cloud.ply 
rtabmap_30_100_50_120_cloud.ply
rtabmap_30_100_120_190_cloud.ply
rtabmap_30_100_190_260_cloud.ply
    
rtabmap_100_170_50_120_cloud.ply  
rtabmap_100_170_120_190_cloud.ply      
rtabmap_100_170_190_260_cloud.ply