Leon1da / elective_artificial_intelligence_1

This repository contains the project developed for the Elective in Artificial Intelligence 1 class.

Elective in Artificial Intelligence: AI for Visual Perception in HCI & HRI

This repo contains the code for the project Beacon-based Scale Estimation for Monocular Structure from Motion in Robotics Systems

Installation

pip install -r requirements.txt

Tested on Ubuntu 20.04 LTS + Python 3.8

Optional:

Install COLMAP (https://github.com/colmap/colmap) and hloc (https://github.com/cvg/Hierarchical-Localization) to start the mapping phase. Note that we already provide some maps.

Data

Unzip dataset.zip inside data/icra_data/

Run the complete pipelines

There are two ways to run the pipeline:

Incrementally adjust the scale using the available data coming from the beacons:

python tests/test_incremental_pipeline.py --input_model reconstruction_outputs/<recostruction_sequence>/sfm/

Compute the scale correction oneshot:

python tests/test_oneshot_pipeline.py --input_model reconstruction_outputs/<recostruction_sequence>/sfm/

Test building block

test_scale_estimator_module.py

1. generate a point clouds (blue)
2. generate a sets of measurements for the point cloud (red)
3. estimate the Similarity transformation that aligns the two point clouds (cyan)
4. refine the obtained solution estimating a Rigid transformation between the points and the measurements corrected with the Similarity above (green)

python tests/test_scale_estimator_module.py

test_segmentation_module.py

Segment the given images.

python tests/test_segmentation_module.py

test_sfm_module.py

given a set of images

1. extract features
2. compute matches
3. run a reconstruction

python tests/test_sfm_module.py

Note:

• hloc required
• in order to run a custom reconstruction the file test_sfm_module.py should be properly cofigurated.

test_sfm_visualization.py

1. visualize a reconstruction

python tests/test_sfm_visualization.py --input_model reconstruction_outputs/<recostruction_sequence>/sfm/

Validation

test_simulator_generation.py

generate 3 point clouds:

• a torus
• a sphere
• a cuboid with different level of noise

python tests/test_simulator_generation.py

Note:

• in order to generate a torus, a sphere or a cuboid the file test_simulator_generation.py should be properly modified.

test_trajectory_alignment.py

given a reconstruction

1. perform Iterative Closest Point optimization using the ground-truth poses and estimated poses

python tests/test_trajectory_alignment.py --input_model reconstruction_outputs/<recostruction_sequence>/sfm/