MSc Thesis project in collaboration with TU Delft and cogniBIT GmbH on human prediction of another vehicle at an intersection.
30 participants were shown 168 intersection scenarios to analyse the influence of heading angle, relative position, blinkers, deceleration and lateral offset on the predicted directions. The predicted direction response is an array of [Forward%, Left%, Right%], always summing to 100%. When the participant was unsure about the direction a [33.3%, 33.3%, 33.3%] response was given (also the default option), when the participant was very confident about a direction the response would be [100%, 0%, 0%] in the left/forward/right direction.
The human prediction responses are stored in the .h5 files. Please read the metadata.txt file for the contents of each file.
To recreate the experiment, the video and image files are to be used with the experiment_gui.py from the gitlab page below. The files need to be stored in a /data/experiment/ directory or the location in the python file needs to be changed.
https://gitlab.tudelft.nl/tud-cor-hri/student-projects/human-predictions-in-traffic-joop
Within the experiment_gui.py code, the participant_ID value needs to be changed to the pseudonymized ID for the corresponding participant. The ID is also responsible for the random seed, which affects the random order the videos are shown in.
The experiment was approved by the TU Delft Ethics committee, and requires participant of the experiment to sign an informed consent form prior to taking part in the experiment. No PII of the participant is stored within the dataset.
ER -