perf(autoware_detected_object_validation): reduce lanelet_filter processing time

[a-maumau]

Description

This PR is for speeding up the lanelet filtering process.
This PR will reduce the number of loops by using the inner R-tree of intersected lanelets.

Using the TimeKeeper, I checked the bottleneck of lanelet filter node. The main bottleneck was the for-loop for the filtering for the objects that have no bounding box (unknown object).

processing times

Example of worst cases.

• following values are different run from the figure

no bbox filter is the part taking time.

Before this PR

objectCallback: 136.19 [ms]
    ├── getConvexHull: 0.35 [ms]
    ├── getIntersectedLanelets: 1.62 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 2.24 [ms]
    │   └── isObjectOverlapLanelets: 2.24 [ms]
    │       └── no bbox filter: 2.24 [ms]
    ├── filterObject: 1.77 [ms]
    │   └── isObjectOverlapLanelets: 1.68 [ms]
    │       └── no bbox filter: 1.68 [ms]
    ├── filterObject: 1.60 [ms]
    │   └── isObjectOverlapLanelets: 1.60 [ms]
    │       └── no bbox filter: 1.60 [ms]
    ├── filterObject: 0.75 [ms]
    │   └── isObjectOverlapLanelets: 0.74 [ms]
    │       └── no bbox filter: 0.74 [ms]
    ├── filterObject: 1.11 [ms]
    │   └── isObjectOverlapLanelets: 1.11 [ms]
    │       └── no bbox filter: 1.11 [ms]
    ├── filterObject: 3.11 [ms]
    │   └── isObjectOverlapLanelets: 3.11 [ms]no bbox filter: pre process: 0.23 [ms]
    │       └── no bbox filter: main loop: 0.22 [ms]
    │       └── no bbox filter: 3.10 [ms]
    ├── filterObject: 1.29 [ms]
    │   └── isObjectOverlapLanelets: 1.29 [ms]
    │       └── no bbox filter: 1.28 [ms]
    ├── filterObject: 2.51 [ms]
    │   └── isObjectOverlapLanelets: 2.51 [ms]
    │       └── no bbox filter: 2.51 [ms]
    ├── filterObject: 1.55 [ms]
    │   └── isObjectOverlapLanelets: 1.55 [ms]
    │       └── no bbox filter: 1.55 [ms]
    ├── filterObject: 1.11 [ms]
    │   └── isObjectOverlapLanelets: 1.10 [ms]
    │       └── no bbox filter: 1.10 [ms]
    ├── filterObject: 0.54 [ms]
    │   └── isObjectOverlapLanelets: 0.54 [ms]
    │       └── no bbox filter: 0.54 [ms]
    ├── filterObject: 2.34 [ms]
    │   └── isObjectOverlapLanelets: 2.34 [ms]
    │       └── no bbox filter: 2.33 [ms]
   	...

After this PR

objectCallback: 12.90 [ms]
    ├── getConvexHull: 0.09 [ms]
    ├── getIntersectedLanelets: 0.32 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.00 [ms]
    ├── filterObject: 0.22 [ms]
    │   └── no bbox filter: pre process: 0.21 [ms]
    │       └── no bbox filter: main loop: 0.21 [ms]
    ├── filterObject: 0.58 [ms]
    │   └── no bbox filter: pre process: 0.57 [ms]
    │       └── no bbox filter: main loop: 0.57 [ms]
    ├── filterObject: 0.19 [ms]
    │   └── no bbox filter: pre process: 0.18 [ms]
    │       └── no bbox filter: main loop: 0.18 [ms]
    ├── filterObject: 0.01 [ms]
    │   └── no bbox filter: pre process: 0.01 [ms]
    │       └── no bbox filter: main loop: 0.00 [ms]
    ├── filterObject: 0.23 [ms]
    │   └── no bbox filter: pre process: 0.22 [ms]
    │       └── no bbox filter: main loop: 0.22 [ms]
    ├── filterObject: 0.20 [ms]
    │   └── no bbox filter: pre process: 0.20 [ms]
    │       └── no bbox filter: main loop: 0.20 [ms]
    ├── filterObject: 0.23 [ms]
    │   └── no bbox filter: pre process: 0.23 [ms]
    │       └── no bbox filter: main loop: 0.22 [ms]
    ├── filterObject: 0.18 [ms]
    │   └── no bbox filter: pre process: 0.17 [ms]
    │       └── no bbox filter: main loop: 0.17 [ms]
    ├── filterObject: 0.20 [ms]
    │   └── no bbox filter: pre process: 0.19 [ms]
    │       └── no bbox filter: main loop: 0.19 [ms]
    ├── filterObject: 0.12 [ms]
    │   └── no bbox filter: pre process: 0.12 [ms]
    │       └── no bbox filter: main loop: 0.12 [ms]
    ├── filterObject: 0.53 [ms]
    │   └── no bbox filter: pre process: 0.53 [ms]
    │       └── no bbox filter: main loop: 0.52 [ms]
    ├── filterObject: 0.17 [ms]
    │   └── no bbox filter: pre process: 0.16 [ms]
    │       └── no bbox filter: main loop: 0.16 [ms]
    ├── filterObject: 0.16 [ms]
    │   └── no bbox filter: pre process: 0.15 [ms]
    │       └── no bbox filter: main loop: 0.15 [ms]
    ├── filterObject: 0.05 [ms]
    │   └── no bbox filter: pre process: 0.04 [ms]
    │       └── no bbox filter: main loop: 0.04 [ms]
    ├── filterObject: 0.00 [ms]
    │   └── no bbox filter: pre process: 0.00 [ms]
    │       └── no bbox filter: main loop: 0.00 [ms]
    ├── filterObject: 0.13 [ms]
    │   └── no bbox filter: pre process: 0.13 [ms]
    │       └── no bbox filter: main loop: 0.13 [ms]
    ...

no bbox filter: pre process: from R-tree search to the end
no bbox filter: main loop: from for-loop checking to the end

VS. previous implementations

There is no guarantee the time is matching the actual time line or correspond to each other.

original: before #8109
current: after #8109 (before this PR)
optimized: after this PR

play in x0.5 means, bag play with rate of 0.5.

mean processing time

original mean: 34.401 ms
current mean: 30.980 ms
optimized best comb. mean: 3.899 ms

(mean processing time was calculated with records that have more than 1 ms)

Related links

Parent Issue:

• Link

How was this PR tested?

Processing time

Processing time was gathered from TimeKeeper: https://github.com/autowarefoundation/autoware.universe/tree/main/common/autoware_universe_utils

Degradation

Qualitative test is done by manually checking and comparing with the original implementation's detected and filtered unknown objects.

No quantitative evaluation was done.

Notes for reviewers

None.

Interface changes

None.

Effects on system behavior

None.

[github-actions]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[technolojin]

TEST under TIER IV cloud

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 39 lines in your changes missing coverage. Please review.

Project coverage is 29.24%. Comparing base (6719f5f) to head (d617ffb).
Report is 15 commits behind head on main.

Files	Patch %	Lines
...t_validation/src/lanelet_filter/lanelet_filter.cpp	0.00%	39 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8240      +/-   ##
==========================================
- Coverage   29.25%   29.24%   -0.01%     
==========================================
  Files        1600     1600              
  Lines      117744   117775      +31     
  Branches    50716    50723       +7     
==========================================
  Hits        34447    34447              
- Misses      74072    74103      +31     
  Partials     9225     9225              

Flag	Coverage Δ		*Carryforward flag
differential	2.75% <0.00%> (?)
total	29.26% <ø> (+<0.01%)	⬆️	Carriedforward from 6719f5f

*This pull request uses carry forward flags. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[technolojin]

LGTM! Thank you for your contribution.

I did not confirmed that the Autoware coding guideline allows to use abbreviated namespace.
namespace bg = boost::geometry;

[YoshiRi]

LGTM

[YoshiRi]

LGTM! Thank you for your contribution.

I did not confirmed that the Autoware coding guideline allows to use abbreviated namespace. namespace bg = boost::geometry;

It seems no problem because some planning code use this expression.