Skip to content

Latest commit

 

History

History
344 lines (258 loc) · 14.1 KB

README.md

File metadata and controls

344 lines (258 loc) · 14.1 KB
Raw

Dungeon Assistant

Solution for indoor localization using WiFi signal and LiDAR.

Architecture

.
├─android          # Android Demo App
├─capturer         # Android WiFi signal capturer App
├─dataset          # Dataset (huggingface submodule)
├─docs             # README assets
├─multiscan        # multiscan-modified submodule
└─utils            # utils for python scripts

Technical details can be found in the report.

Clone

To clone all submodules, use

git clone --recurse-submodules -j8 git://github.com/JeffersionQin/DungeonAssistant.git

Note: the dataset is quite large, you might not want to clone it at first. Check out the usage of git submodule for more details.

Scripts

ply file and trajectory viewer

$ python ply_viewer.py --help
usage: ply_viewer.py [-h] [--pointcloud POINTCLOUD] [--trajectory TRAJECTORY]

optional arguments:
  -h, --help            show this help message and exit
  --pointcloud POINTCLOUD
                        point cloud file path
  --trajectory TRAJECTORY
                        trajectory file path

Example,

Closure Optimization for multiscan's result

$ python closure_optimization.py --help                   
usage: closure_optimization.py [-h] [--pointcloud_base POINTCLOUD_BASE] [--pointcloud_prefix POINTCLOUD_PREFIX]
                               [--merge_cnt MERGE_CNT] [--output_dir OUTPUT_DIR]

optional arguments:
  -h, --help            show this help message and exit
  --pointcloud_base POINTCLOUD_BASE
                        base dir of point clouds
  --pointcloud_prefix POINTCLOUD_PREFIX
                        prefix of point cloud file name
  --merge_cnt MERGE_CNT
                        number of continuous (by folder names) point clouds to merge together
  --output_dir OUTPUT_DIR
                        output directory

Example,

Multiple point cloud and trajectory merger

$ python multi_merger_viewer.py --help
usage: multi_merger_viewer.py [-h] [--pointcloud_base POINTCLOUD_BASE] [--pointcloud_prefix POINTCLOUD_PREFIX]
                              [--merge_cnt MERGE_CNT] [--transformation_dir TRANSFORMATION_DIR]
                              [--overlap_discard_num OVERLAP_DISCARD_NUM] [--pointcloud_out POINTCLOUD_OUT]
                              [--trajectory_out TRAJECTORY_OUT]

optional arguments:
  -h, --help            show this help message and exit
  --pointcloud_base POINTCLOUD_BASE
                        base dir of point clouds
  --pointcloud_prefix POINTCLOUD_PREFIX
                        prefix of point cloud file name
  --merge_cnt MERGE_CNT
                        number of continuous (by folder names) point clouds to merge together
  --transformation_dir TRANSFORMATION_DIR
                        directory storing transformation matrix files, default to be empty
  --overlap_discard_num OVERLAP_DISCARD_NUM
                        number of overlap frames to discard
  --pointcloud_out POINTCLOUD_OUT
                        output point cloud file name
  --trajectory_out TRAJECTORY_OUT
                        output trajectory file name

Example,

Registration between two point clouds

$ python registration.py --help                           
usage: registration.py [-h] [--pointcloud1 POINTCLOUD1] [--pointcloud2 POINTCLOUD2] [--trajectory1 TRAJECTORY1]
                       [--trajectory2 TRAJECTORY2] [--fast_cache FAST_CACHE] [--icp_cache ICP_CACHE]
                       [--voxel_size_fgr VOXEL_SIZE_FGR] [--voxel_size_icp VOXEL_SIZE_ICP] [--skip_icp]

optional arguments:
  -h, --help            show this help message and exit
  --pointcloud1 POINTCLOUD1
                        first point cloud file path (1 --[transform]-> 2)
  --pointcloud2 POINTCLOUD2
                        second point cloud file path (1 --[transform]-> 2)
  --trajectory1 TRAJECTORY1
                        first trajectory file path
  --trajectory2 TRAJECTORY2
                        second trajectory file path
  --fast_cache FAST_CACHE
                        transformation cache of fast global registration if available. default is none
  --icp_cache ICP_CACHE
                        transformation cache of icp if available. default is none
  --voxel_size_fgr VOXEL_SIZE_FGR
                        voxel size for global fast registration downsampling. default is 0.05
  --voxel_size_icp VOXEL_SIZE_ICP
                        voxel size for icp downsampling. default is 0.05
  --skip_icp            skip icp and only run fgr
  --transformed_trajectory_out TRANSFORMED_TRAJECTORY_OUT
                        output trajectory of the transformed trajectory 1 (to trajectory 2)

Signal strength dataset construction and evaluation

Output file will be in the format of recording time and signal strength of each access point.

$  python dataset_construction.py --help  
usage: dataset_construction.py [-h] [--trajectory TRAJECTORY] [--wifi WIFI] [--output OUTPUT]

optional arguments:
  -h, --help            show this help message and exit
  --trajectory TRAJECTORY
                        trajectory file path
  --wifi WIFI           wifi file path
  --output OUTPUT       output file path

Evaluation script will also output the error statistics.

$  python evaluation.py --help                                                                                                                                 
usage: evaluation.py [-h] [--trajectory TRAJECTORY] [--wifi WIFI] [--dataset DATASET] [--output OUTPUT]

optional arguments:
  -h, --help            show this help message and exit
  --trajectory TRAJECTORY
                        trajectory file path
  --wifi WIFI           wifi file path
  --dataset DATASET     dataset file path
  --output OUTPUT       output file path

Note: on your own usage, please take care of your time zone settings (US/Eastern) and might need some modifications to the code. Also, current AP filtering is based on the setting of UPenn Engineering Quad (SSID == 'AirPennNet'), you might need to modify the code to fit your own environment.

Floor plan extraction script

$ python floorplan_extraction.py --help
usage: floorplan_extraction.py [-h] [--pointcloud POINTCLOUD] [--output OUTPUT] [--scale SCALE]

optional arguments:
  -h, --help            show this help message and exit
  --pointcloud POINTCLOUD
                        first point cloud file path (1 --[transform]-> 2)
  --output OUTPUT       output file path
  --scale SCALE         scale of the floor plan

The script will also output the size and scale information of the map generated.

minX: -812, minY: -1009
maxX: 294, maxY: 7
scale: 10
floor plan saved to: floorplan.png

This will be useful when configurating the Android Demo App.

Example floor plan,

Example: UPenn Engineering Quad Indoor Localization

The following is an example of using this project to conduct indoor localization in UPenn Engineering Quad.

Penn Engineering: https://www.seas.upenn.edu/

Data

Check the dataset folder, and the repository: https://huggingface.co/datasets/gyrojeff/DungeonAssistant

dataset
├─processed                               # processed data
│  ├─Equad-01
│  ├─Equad-02
│  └─Equad-04
└─raw                                     # raw data
    ├─231206 05 - Equad 01 Evening - 1F   # one trial of data collection
    │  └─Segments                         # overlapping sliding window segment
    │      ├─00
    |      ├─...
    │      └─18
    ├─231207 08 - Equad 02 Evening - 1F
    └─231208 10 - Equad 04 Evening - 1F

Data are scanned using MultiScanModified, cropped using https://github.com/JeffersonQin/multiscan-modified/blob/main/server/crop.py

For the file structure of the processed data, please refer to the command line arguments of the Reproduce section. It shows the complete structure of dataset/processed/Equad-01 as an example.

Reproduce

Closure optimization for Equad 01 (12/06 Evening [05])

python closure_optimization.py --pointcloud_base "dataset/raw/231206 05 - Equad 01 Evening - 1F/Segments" --pointcloud_prefix "20231206T222514-0500_2D50E931-A486-4EDE-A859-2982CCB91A95-" --merge_cnt 4 --output_dir dataset/processed/Equad-01/transform

Merge and transform for Equad 01 (12/06 Evening [05])

python multi_merger_viewer.py --pointcloud_base "dataset/raw/231206 05 - Equad 01 Evening - 1F/Segments" --pointcloud_prefix "20231206T222514-0500_2D50E931-A486-4EDE-A859-2982CCB91A95-" --merge_cnt 4 --transformation_dir dataset/processed/Equad-01/transform --overlap_discard_num 6000 --pointcloud_out dataset/processed/Equad-01/pc.ply --trajectory_out dataset/processed/Equad-01/trajectory.jsonl

Same for Equad 02 (12/07 Evening [08]), Equad 04 (12/08 Evening [10]).

Registration from 02 -> 01

python registration.py --pointcloud1 dataset/processed/Equad-02/pc.ply --trajectory1 dataset/processed/Equad-02/trajectory.jsonl --pointcloud2 dataset/processed/Equad-01/pc.ply --trajectory2 dataset/processed/Equad-01/trajectory.jsonl --voxel_size_fgr 0.5 --voxel_size_icp 0.1 --skip_icp --transformed_trajectory_out dataset/processed/Equad-02/trajectory_alignedto_01.jsonl

Construct dataset using WiFi signal data from 01.

python dataset_construction.py --trajectory dataset/processed/Equad-01/trajectory.jsonl --wifi "dataset/raw/231206 05 - Equad 01 Evening - 1F/WiFiCapture_20231206_222454.csv" --output dataset/processed/Equad-01/dataset.csv

Evaluation for position prediction from WiFi signal data from 02, using dataset constructed from 01.

python evaluation.py --trajectory dataset/processed/Equad-02/trajectory_alignedto_01.jsonl --wifi "dataset/raw/231207 08 - Equad 02 Evening - 1F/WiFiCapture_20231207_194322.csv" --dataset dataset/processed/Equad-01/dataset.csv --output dataset/processed/Equad-02/errors_to_01.npy
stats value (m)
mean 4.060915078446458
std 2.9731415933144287
median 3.479469846080446
max 17.191420152840074
min 0.1086243602998307
25th 1.762926267979433
75th 5.421255493003348

Other evaluations are similar.

One more example is between 01 and 05,

python closure_optimization.py --pointcloud_base "dataset/raw/231216 11 - Equad 05 Afternoon - 1F/Segments" --pointcloud_prefix "20231216T145533-0500_9D8445D3-B4CD-4264-B3C0-76DB1007D351-" --merge_cnt 2 --output_dir dataset/processed/Equad-05/transform

python multi_merger_viewer.py --pointcloud_base "dataset/raw/231216 11 - Equad 05 Afternoon - 1F/Segments" --pointcloud_prefix "20231216T145533-0500_9D8445D3-B4CD-4264-B3C0-76DB1007D351-" --merge_cnt 2 --transformation_dir dataset/processed/Equad-05/transform --overlap_discard_num 6000 --pointcloud_out dataset/processed/Equad-05/pc.ply --trajectory_out dataset/processed/Equad-05/trajectory.jsonl

python registration.py --pointcloud1 dataset/processed/Equad-05/pc.ply --trajectory1 dataset/processed/Equad-05/trajectory.jsonl --pointcloud2 dataset/processed/Equad-01/pc.ply --trajectory2 dataset/processed/Equad-01/trajectory.jsonl --voxel_size_fgr 0.5 --voxel_size_icp 0.1 --skip_icp --transformed_trajectory_out dataset/processed/Equad-05/trajectory_alignedto_01.jsonl

python evaluation.py --trajectory dataset/processed/Equad-05/trajectory_alignedto_01.jsonl --wifi "dataset/raw/231216 11 - Equad 05 Afternoon - 1F/WiFiCapture_20231216_145217.csv" --dataset dataset/processed/Equad-01/dataset.csv --output dataset/processed/Equad-05/errors_to_01.npy

More Results

Evaluation Mean (m) Std (m) Median (m) Max (m) Min (m) 25th 75th
Equad 01 -> 02 (~ 1 Day) 4.06 2.97 3.48 17.19 0.11 1.76 5.42
Equad 01 -> 04* (~ 2 Days) 4.28 3.45 3.24 20.85 0.05 1.69 6.33
Equad 01 -> 04 (~ 2 Days) 4.75 7.02 3.27 83.62 (outlier) 0.05 1.71 6.39
Equad 01 -> 05 (~ 10 Days) 12.05 8.07 11.02 41.23 0.15 5.65 17.37

*: Kick out outlier due to temporary WiFi collection system down.

Android Demo App

Usage / Configuration / Build

Notice that currently the Android Demo App is based on the previous example of UPenn Engineering Quad, and the configuration is hard-coded. You might need to modify the code to fit your own environment.

You would need to modify the minX, maxX, minY, maxY, scale data in android/app/src/main/java/moe/gyrojeff/dungeonassistant/MainActivity.java based on the output of floorplan_extraction.py.

Also, you would need to replace the file

  • android/app/src/main/res/mipmap-hdpi/floorplan.png with your own floor plan
  • android/app/src/main/res/raw/dataset.csv with your own processed signal dataset

Demo Video

You can find the demo video here: https://youtu.be/XT8v7n0ZuOk

Screenshot

TODO

  • Implement Kalman Filter for the Android Demo App
  • Implement the 2D error evaluation script
  • Implement k-NN for k > 1
  • Allow dynamic update of the map and metadata for the Android Demo App
    • Server based
    • UI based
  • Permission Request instead of manual grant for the Android Demo App