Exwayz Map Cleaner algorithm performs a processing pipeline for classifying points belong to mobile objects. The method is called DOC for Dynamic Object Classification.

1. The ground is classified on each frame of the sequence
2. A free space check is done using adjacent frame of the query frame to estimate if points have caused occlusion, indicating that they could belong to mobile objects.

<aside> 💡

The method is described in this paper we published on ArXiv: https://arxiv.org/pdf/2502.02144

</aside>

Below is a sample DOC configuration file.

__profile_name__: "default"

buffer:
    length: 200.0
    extra_length: 20.0
    d_update: 0.50
    use_trajectory_rotation: false
    exhaustive_reprojection: true
	

ground_classification:
    radius_seeds: 2.0
    knn_propagation: 30
    max_normal_angle: 15.0
    max_ptopl_dist: 0.02
    subsampling: 0.04

mobile_object_classification:
    num_threads: 4

    dtheta: 0.3
    dphi: 0.8

    radius_search_short: 20.0
    d_step_short: 1.0

    radius_search_long: 100.0
    d_step_long: 10.0

    window_size: 5
    threshold: 0.2

buffer

ground_classification

mobile_object_classification