AROA-Trajectory-Analysis

This repository holds the data and analysis code for "Using augmented reality to cue obstacles for people with low vision."

It has two sections:

• Matlab Analysis uses Matlab R2022B and holds the raw data, the code for examining individual trials, and code for summarization.
• R Analysis uses R 4.2.0 and holds the summative data and most of the plots used in the paper.

Note that any reference to "Collocated" indicates world-locked cues, and to "HUD" indicates heads-up cues.

---

Matlab Analysis

Top Level Breakdown

• Plotting Scripts - holds Matlab scripts for data analysis. See "Data Plotting" and "Data Processing" below.
• Raw Data - holds raw data for each participant as text files. Each subfolder contains one text file per trial. See "Raw Data Breakdown" for more information.
• Processed Data - holds processed data for each participant as CSV files. Each subfolder contains one CSV per trial. See "Data Processing" for more information.
• Plots - holds graphs of participant paths during each trial as PNGs and Matlab figures. Each participant has four images per trial: PNG and Matlab figure of side-on view, and the same for top-down view. The participant's paths are marked with dots indicating position at every 0.5 seconds. This folder also holds an "Empty Layout" folder showing layouts with no participant.
• Summary Statistics - holds summary stats for each participant in one CSV.

Raw Data Breakdown

File structure for raw data should be as follows, from top-level to bottom:

1. Raw data folder
2. Folder for each participant, titled [yy-mm-dd OA##] (e.g. "22-02-16 OA01").
3. One text file per trial, with titles as generated by Experiment Logger (e.g. "2022-02-16_03-15-26-PM_No Cues_Layout 2_Forward").

Trial data is separated by a semicolon (and a space for most strings). Columns are as follows:

• Log Type - Whether logging was done in Update or Fixed Update in Unity. All data uses Fixed.
• Cue Condition - The cue condition: No Cues, Collocated (world-locked), HUD (head-locked), or Combined.
• Layout - The layout number used. Integer between 1 and 8.
• Direction - Whether the trial was Forward or Backward
• Time - Time since start of trial. Note that raw data starts at the first recorded time after 0, whereas fixed data is set to startL at 0.
• Position X/Y/Z - Location coordinates of HoloLens in meters
• Rotation X/Y/Z - Rotation of HoloLens in degrees
• Eye Tracking Enabled - Boolean of whether eye tracking is enabled. Not used.
• Eye Tracking Data Valid - Boolean of whether eye tracking data is valid. Not used.
• Gaze Direction X/Y/Z, Eye Movement X/Y/Z - data related to eye movement and total gaze vector. Not used.
• HUD Cue Up/Right/Down/Left - Boolean of whether each heads-up cue was enabled.

Raw data files also have "Stopped logging at [end time]" at the bottom.

Data Processing

First, run raw_data_processing to turn the raw data into processed CSVs. This includes setting initial X, Y, and Z values to 0, realigning along the axis of maximal variance so that Z always represents moving down the hallway, removing duplicate entries, deleting data after Z = 15 (when the participant has crossed the finish line), and removing OA07 and OA11 which had corrupted data.

Second, summary_stats_exporter can summarize key statistics for each participant. It iterates through all participant's trials and creates a CSV with the following attributes for each trial:

• Condition - as per Cue Condition above
• Direction - as per Direction above
• Total Distance - distance travelled per trial
• Total Time - time per trial
• Average Speed - average speed per trial
• X/Y/Z Rotation - the absolute sum of rotation in the X, Y, and Z directions. (e.g. if a participant turned their head right 45 degrees and then back to the left 45 degrees, the total rotation would be 90 degrees.)
• Total Rotation - the sum of X, Y, and Z rotation
• Percent Stopped - the percentage of the trial for which a participant was moving < 0.1 m/s
• Percent Slowed - the percentage of the trial for which a participant was moving < 0.3 m/s
• Speed Variance - the variance of the participant's speed
• Speed Variance Corrected - the variance of the participant's speed, divided by their average velocity to account for increased variance at higher speeds

The Summary Statistics document was copied to the R Analysis folder for ease of access, and used in combination with Supplementary Data (see R Analysis) for final analyses.

Data Plotting

Plotting is done via plotByName and its helper functions. This function takes in a filename and a boolean for top-down or side-on and plots the layout and participant's path. Helper functions include lookupHeight to look up a participant's height in inches (hardcoded); overlayCircles to add a position marker every 0.5 seconds along the path; and overlayObstacles to add the appropriate obstacles based on layout and orientation.

Example position plot:

Note that plotAll can be used to call plotByName repeatedly to iterate through all participants; while plotEmpty can be used to draw an empty layout, with no participant data.

---

---

R Analysis

Folder Breakdown

Here is a breakdown of the files and subfolders in R Analysis.

• Folders
  • Plots - a folder to hold all plots generated via R, sorted by type.
  • Tests - a folder to hold all tests generated via R. Additional sub-analyses of "Preferred" conditions can be found in the Preferred subfolder.
• Data
  • Summary Statistics - Data generated via the HoloLens and processed with Matlab.
  • Supplementary Data - Additional data, such as white cane usage and control condition times.
  • Data Combined - Summary Statistics and Supplementary Data combined into one CSV.
  • Data_avg - Cleaned data, with forward and backward runs combined.
  • Data_noErr - Cleaned data, but with the two participants excluded from quantitative trials (07 and 11) filtered out.
  • Data_avg_noErr - Cleaned data, with forward and backward runs combined and disqualified participants excluded.
• Scripts
  • AROA_Control_Script - Master script. Creates above datasets from the raw data, calls below functions to create all plots and tests.
  • friedmanTests - Runs Friedman and/or Wilcox tests on likert rating answers, and exports plots.
  • friedmanControl - As above, but for control condition only.
  • plotAndAnova - Creates a plot and conducts ANOVA analysis for quantitative elements (e.g. time, distance) against cue condition.

Raw Data Breakdown

Here are the columns in Data Combined. There's one row per trial. Note that columns marked with an asterisk are not available for the Control condition.

• Participant ID - Participant ID string, e.g. "OA01."
• Eye calibration success - Boolean indicating whether eye calibration worked for that participant.
• HUD adjustment - String indicating whether the heads-up cues needed any adjustment to accommodate a lack of vision in one eye. Either "none" or number of times the cues were condensed and moved left or right.
• White Cane - Boolean indicating whether the participant used a white cane.
• PWS Average Time - Double indicating preferred walking speed average time in seconds.
• Preferred Walking Speed - Double indicating the average preferred walking speed in meters per second.
• Condition - String indicating cue condition. Control (first), Control (last), No Cues, Collocated, HUD, or Combined.
• Direction - String indicating direction. "Forward" or "Backward".
• Layout - Integer indicating layout (1-8).
• Order - Integer indicating what order the trials ocurred in for each participant. (1-6)
• PPWS - Double indicating percentage of preferred walking speed for each trial. (Format: 50%)
• Errors - Integer indicating number of errors committed per trial.
• Median Rating - Double indicating the median of the Likert ratings for each trial.
• Confidence/Obstacle location/Obstacle size/Awareness - Integers indicating participant's Likert rating for each trial.
• Preferred Condition - String indicating whether each condition was the participant's preferred condition. "Yes" or "No" for world-locked, heads-up, and combined; "N/A" for control and no cues.
• Calibrated Distance* - Double indicating true distance walked by the participant in each trial between the starting and ending lines, in meters.
• Average Speed* - Double indicating average speed of the participant based on calibrated distance and time, in meters per second.
• X Pitch/Y Yaw/Z Roll* - Double indicating total degrees of rotation in X/Y/Z directions.
• Total Rotation* - Sum of X, Y, and Z rotation.
• rotSpeedX/Y/Z* - Average rotation speed in the X/Y/Z direction, in degrees per second.
• Percent Stopped/Slowed* - Double indicating percentage of time each participant was moving less than 0.1 m/s or 0.3 m/s, respectively. (Format: 0.50)
• Speed Variance* - Double indicating the variance of the participant's speed in each trial.
• Speed Variance Corrected* - As above but divided by the participant's average speed.
• Calibrated Time - Double indicating time taken to walk from starting to ending line, in seconds.

Data Processing

Here are the steps taken to process the raw data as given above. This is done in AROA Control Script.

1. Data are joined by unique IDs and extra columns cleaned.
2. Non-numerical data such as percentage signs and commas are removed.
3. The "N/As" in data columns are replaced with 9999 to keep them numerical.
4. The Likert scores for all non-control conditions are changed from 1 to 7 to -3 to 3 by subtracting 4. This is because the non-control conditions are compared against the control condition (with 0 = same as the control).
5. Data_avg is created by averaging forward and backward runs for each trial, then averaging the Control (first) and Control (last) trials.
6. Data_avg_noErr is created by removing any participants with NaN marked (indicating that the HoloLens did not record properly for one or more of their trials).
7. Data_noErr is created by removing any participants with NaN marked, but leaving forward and backward/Control (first) and Control (last) intact.

For information on how Friedman and ANOVA tests were run, see the relevant scripts.

Note that the code also includes means of analyzing the data based on each participant's cue preference, but these are not used in the paper.

Data Plotting

Plotting is done primarily in the friedmanTests, friedmanControl, and plotAndAnova scripts.

Example Likert plot (in this case, Confidence ratings by condition):

Example quantitative plot (Percentage of Preferred Walking Speed by condition):