Can we draw AOIs?

[Nitro60zeus]

Is it possible to have certain areas of interest? For example, an AOI on the rearview mirror, so that we can have the count for the number of times the participant looks at the rearview mirror.

[ajdroid]

Not exactly by drawing an AOI but we do provide the tickfocus_hitpt in the recording which will tell you what 3D mesh the gaze was intersecting with at each point. The logic uses a collision channel to figure out what to include and/or ignore (we ignore the ego vehicle typically). Relevant code is here: https://github.com/HARPLab/DReyeVR/blob/57109001edcfe0d91a6f814cf9c3287ff2bb57e4/DReyeVR/EgoSensor.cpp#L287

[Nitro60zeus]

Oh, that's great! Thanks for your response. But how do I obtain that data for analysis? For example, how do I know if the gaze intersected with the traffic lights or not? Or let's say I have a sign board, and I want to know if the gaze intersects with that sign board or not.

[Nitro60zeus]

Ok, I see this. I assume this is what you meant... (got this txt file from /show_recorder_file_info.py -a -f /PATH/TO/RECORDER-FILE > recorder.txt )

Wrote a code to extract the names. Hope this helps anyone looking for the same cause:

import re
import csv

# Open the input text file
with open('4a.txt', 'r') as file:
    content = file.read()

# Use regular expressions to find all relevant values
timestamp_carla_values = re.findall(r'TimestampCarla:(\d+)', content)
pupil_diameter_values = re.findall(r'PupilDiameter:(-?[\d.]+)', content)
eye_openness_values = re.findall(r'EyeOpenness:(-?[\d.]+)', content)
hit_values = re.findall(r'FocusInfo:Hit:(\d+)', content)
actor_name_values = re.findall(r'ActorName:([\w]+)', content)

# Open a CSV file for writing
with open('z.csv', 'w', newline='') as csvfile:
    writer = csv.writer(csvfile)

    # Write the header row to the CSV file
    writer.writerow(['TimestampCarla', 'PupilDiameter A', 'PupilDiameter B', 'EyeOpenness A', 'EyeOpenness B', 'Hit', 'ActorName'])

    # Determine the minimum length of the lists to avoid index errors
    min_length = min(len(timestamp_carla_values), len(pupil_diameter_values), len(eye_openness_values), len(hit_values), len(actor_name_values))

    # Write the values to the CSV file, alternating EyeOpenness values
    for i in range(min_length):
        row = [
            timestamp_carla_values[i],
            pupil_diameter_values[i * 2] if i * 2 < len(pupil_diameter_values) else '',
            pupil_diameter_values[i * 2 + 1] if i * 2 + 1 < len(pupil_diameter_values) else '',
            eye_openness_values[i * 2] if i * 2 < len(eye_openness_values) else '',
            eye_openness_values[i * 2 + 1] if i * 2 + 1 < len(eye_openness_values) else '',
            hit_values[i] if i < len(hit_values) else '',
            actor_name_values[i] if i < len(actor_name_values) else ''
        ]
        writer.writerow(row)

print("CSV file 'z.csv' has been created with the specified values.")

[Nitro60zeus]

Not exactly by drawing an AOI but we do provide the tickfocus_hitpt in the recording which will tell you what 3D mesh the gaze was intersecting with at each point. The logic uses a collision channel to figure out what to include and/or ignore (we ignore the ego vehicle typically). Relevant code is here:

https://github.com/HARPLab/DReyeVR/blob/57109001edcfe0d91a6f814cf9c3287ff2bb57e4/DReyeVR/EgoSensor.cpp#L287

Hey, I have a small follow-up question. For how many seconds or milliseconds did you guys consider for the duration of collision of the gaze with the object mesh (eg: cyclist), to count as a successful fixation. (As in, when I am moving my eyes, i may glide over a number of objects, but the moment i stop and look at something (that duration, let's say 100 milliseconds) is when I actually looked at the object (or registered the object in my brain) or fixated my vision on it.)

[ajdroid]

Hi, we don't consider fixations in this part of the pipeline -- it just gives you instantaneous gaze hit locations. We filter for fixations post-parsing

[ajdroid]

We use this code for gaze event classification: https://github.com/HARPLab/ibmmpy

[Nitro60zeus]

Hi, we don't consider fixations in this part of the pipeline -- it just gives you instantaneous gaze hit locations. We filter for fixations post-parsing

I am extremely interested in the fixations. Can you tell me how do you filter for fixations post-parsing? Will be extremely helpful. I tried the DReyeVR parser, which is really a great tool. But I don't see any output related to fixations. Thanks a lot btw!

[ajdroid]

As I said above, we use the ibmmpy gaze event classifier to do fixation detection (essentially uses instantaneous velocity of gaze to form 2 clusters -- slow fixations and fast saccades)

[StijnOosterlinck]

@Nitro60zeus have you been able to detect gazes at the rearview mirror?