A novel database, BehavePassDB, is proposed, which includes several touch gestures (free-text keystroke, swipe, tap dynamics), and background sensors (accelerometer, gyroscope, magnetometer, lin. accelerometer, gravity sensor).

Each subject’s data are structured into 4 acquisition sessions. Each acquisition session contains different tasks (texting, text reading, gallery swiping, tapping). During each of the tasks, touchscreen and background sensor data are acquired.

The BehavePass database will be divided into two subsets:

• Development: BehavePassDB DevSet (51 users).

• Validation and Evaluation: BehavePassDB ValSet and BehavePassDB EvalSet (respectively 10 and 20 different users), including the skilled impostor case, i.e., along the 4 genuine sessions, 2 extra sessions carried out by a different user on the same mobile device are included. Such extra sessions will be used to verify a different user on the same mobile device as the genuine one.

Each of the participant teams is free and encouraged to use any additional databases for development, such as the public HuMIdb (mobile user interaction data), the Aalto DB (mobile keystroke dynamics), etc., or private self-collected ones.

Starting from the acquired raw data, to simplify the participation, both the development (BehavePassDB DevSet) and the evaluation (BehavePassDB EvalSet) databases will be provided in the form of JSON files structured as follows:

• DevSet.json, 51 users: Genuine sessions (“g1”-“g4”) : Tasks (“keystroke”, “readtext”, “gallery”, “tap”) : Modality (“touch”, “sensor_acc”, “sensor_accl”, “sensor_gyro”, “sensor_grav”, “sensor_magn”) : Time-Series Data

• ValSet_Task_enrolment.json: Pseudonymized Enrolment Sessions : Task : Modality (“touch”, “sensor_acc”, “sensor_accl”, “sensor_gyro”, “sensor_grav”, “sensor_magn”) : Time-Series Data

• ValSet_Task_verification.json: Pseudonymized Verification Sessions : Task : Modality (“touch”, “sensor_acc”, “sensor_accl”, “sensor_gyro”, “sensor_grav”, “sensor_magn”) : Time-Series Data.

• EvalSet_Task_enrolment.json: Pseudonymized Enrolment Sessions : Task : Modality (“touch”, “sensor_acc”, “sensor_accl”, “sensor_gyro”, “sensor_grav”, “sensor_magn”) : Time-Series Data

• EvalSet_Task_verification.json: Pseudonymized Verification Sessions : Task : Modality (“touch”, “sensor_acc”, “sensor_accl”, “sensor_gyro”, “sensor_grav”, “sensor_magn”) : Time-Series Data.

The Time-Series Data will be provided into a different form depending on the modality. Specifically, each acquired sample is structured as follows:

• Keystroke data: [timestamp, ascii_code]

• All other touch data: [timestamp, x_coordinate / screen_width, y_coordinate / screen_width, action_type]. action_type refers to the touch."0" corresponds to laying the finger, "1" to lifting the finger, "2" to moving the finger on the screen.

• Background sensor data: [timestamp, x_coordinate, y_coordinate, z_coordinate].

The participants can implement any further processing step to extract additional features.

A graphical visualization of the Tasks is reported below.