Indoor Positioning System (GUI) - Razyapoo/Indoor-Positioning-System GitHub Wiki

Overview

This is a main GUI of our system. It solely serves as a tool, which enables to work offline with the collected data. It facilitates the user's interaction with our developed system, allowing for:

• Aligning (synchronizing) and visualizing video and UWB data streams recorded by Server
• Using the Pixel-To-Real and Optical methods for people localization
• Performing data analysis
• Exporting the data, e.g., for model training, but not restricted to it
• Loading / Playing / Pausing / Seeking the data streams

The application is built using the Qt 6.6.1 framework.

Installation

Installation process is detailed in the folder containing this application on GitHub.

Usage

Run the GUI and load the data

1. Open the Indoor Positioning System application:

   • Run the executable ./IndoorPositioningSystem in the build/ folder.

2. Navigate to the File in the top menu:

   

• This will provide you with four different options. As for now we are interested in:
  • Open video and UWB data: Select the folder with the video and UWB data as a single project / package from Data for Indoor Positioning System (GUI).
    • For example, select Train - Experiment 109. In the rest of this tutorial, this experiment will be used as an example.
  • Other options are not mandatory as for now. They will be detailed later in this tutorial.

When video and UWB data are selected:

• Enter the anchors' positions:

  

  • The GUI will require you to enter the anchors' positions
  • These positions can be found in the description in Data for Indoor Positioning System (GUI).
  • The GUI requires to set the origin anchor, which represents the origin of the UWB coordinate system.
    • This allows to extend the UWB coordinate system to match the actual size of the environmental area.
  • It is possible to add as many anchors as you wish.
  • It is allowed to overwrite an already existing anchor by writing its ID again with new coordinates.
  • Once all positions of anchors are entered, close the input window by pressing Finish adding.
    • After this the GUI will start playing the data stream.

Main view of the GUI

• Area 1: Represents the stream of video data, consisting of Frame ID and its timestamp
• Area 2: Represents the UWB data stream. Both Video and UWB streams are synchronized by looking for the closest timestamps.
• Area 3: Dedicated to the UWB data. It allows for correcting the data by analyzing them and showing the UWB top-down schema and estimated UWB coordinates.
• Area 4: Allows to navigate through the video. Frame seeking.
• Area 5: Play / pause data stream
• Area 6: Intended to run Pixel-to-Real and Optical methods and show their results.
• Area 7: Responsible for exporting the data. Note that this export process is very time-consuming. There is another option to export the data, which can be performed in the Data Analysis window, as detailed later.

Show UWB coordinates

• Coordinate window is shown automatically, once video and UWB data are loaded

• Top-down view (Localization schema)

• Schema allows to hover the mouse over the shapes to see their coordinates!

How to analyze UWB data (for its correction)

• Allows to correct the discrepancies in UWB distance measurements
• First, the time rage should be set
• Then, the pair of interest (tag and anchor) should be selected
• Then, standing periods are determined and mean value for each period is calculated.
• These mean values are then used to train polynomial regression capable of correcting the original UWB measurements.
• IMPORTANT: this analysis is intended only for tags, and only in pair with one anchor
• This allows for more detailed examination of measurements

The main process of analysis:

1. Set the time range of interest, then press Set Time Range button.

2. Select a tag and anchor pair

NOTE: In the current example, there will be only 1 tag and 2 anchors to choose from. In general, there can be more tags and anchors.

• Tag

• Anchor

This is how the choice will looks like in the common case:

• Tag

• Anchor

3. Input the rolling window size to determine standing periods

NOTE: Diagrams are interactive. Use the following combinations:

• ctrl + left mouse + "mouse move": allows to pan a diagram view
• ctrl + mouse wheel: allows to zoom in / out a diagram
• hovering the mouse over the lines: shows timestamp (in hour:min:sec for better analysis) + UWB measurement recorded at that time

4. Set a threshold to split the dataset into standing periods

• When Calculate is pressed, the mean values (segment / period representatives) are calculated for each standing period.

5. Assign each period with corresponding Ground Truth distance

• Ground Truth values are used to train polynomial regression function
  • There are two possibilities to enter the Ground Truth values:
    • (1.) Upload batch file containing GT values for all standing periods
    • (2.) Set Ground Truth for each standing period

6. Calculate the polynomial regression to correct distance measurements for the entire range of tag-anchor pair

7. Update UWB distances with corrected ones

• (1.) Update (adjust) UWB measurements "on-the-go" (in the main UWB stream in GUI) with correct ones using trained polynomial function. Subsequently, it will correct the UWB coordinates.
• (2.) Export segments' (standing periods) representatives for the selected tag-anchor pair.
  • This only exports the segment representatives from the defined range, and not the entire range of values one-by-one.
  • Export of segment representatives helps to reduce data export time and collect more diverse coordinates for machine learning training.

Run Pixel-to-Real and Optical methods

• These methods require additional configuration files to run. These files could be loaded either:

1. Go to the File in the top menu:

• Load pixel to real model params: Select the folder with the parameters for the Pixel-To-Real model (.json file).
• Load camera intrinsic params: Select the folder with the intrinsic parameters for the camera (.xml file).
• Load human detection model weights: Select the folder with the weights (both .weights and .cfg together as a single package) for the YOLOv4 people detection model.

NOTE: The model weights can be loaded only from File in the top menu for safety reasons.

2. Directly by running the methods (pressing buttons). It will show pop-up window and require to load configuration files.

This will start people detection and show dedicated windows with coordinates for each method

Export coordinates

• This feature allows to export "pixel coordinates - coordinates estimated by method" pairs. Methods: UWB, Pixel-to-Real, Optical.

• It is useful for model training and further analysis, but not restricted to it.

• The export can be performed by two ways:

  1. Frame-by-Frame: exports all frames within the defined time range (detailed in Area 7).

  2. Segment representatives: export only segment representatives (detailed in segment export).