Copyright (c) 2018, All rights reserved.

If you use or publish results using this data, please cite the URL link-source:
https://doi.org/10.17862/cranfield.rd.6270233

We have developed a new dataset (VOIDataset: Vehicle-Obstacle Interaction Dataset) for vehicle-obstacle interaction recognition task.  The dataset includes 277 trajectories (sequences of x,y positions of the vehicle and the obstacle) of three different scenarios (67 crash, 106 left-pass, and 104 right-pass trajectories). The distance between the vehicle and the obstacle (length of the trajectory) is 50 meters. The trajectories were manually annotated, and used to evaluate our activity recognition method in our paper 'Vehicle Activity Recognition Using Transfer Learning with Deep Convolutional Neural Networks'. 

Data was gathered using a simulation environment developed in Virtual Battlespace 3 (VBS3), with the Logitech G29 Driving Force Racing Wheel and pedals. Here a model of a Dubai highway was used. We consider a six lane road with an obstacle in the centre lane. The experiment consisted of 40 participants, all of varying ages, genders and driving experiences. Participants were asked to use their driving experience to avoid the obstacle. A koda Octavia was used in all trails, and with maximum speed 50KPH. We recorded the obstacle and ego-vehicle's coordinates (the centre position of the vehicle), velocity, heading angle, and distance from each other. The generated trajectories were recorded at 10Hz.

Description of our VOIDataset.

Scenario 		Description
*************************************
1. Left-Pass: The ego-vehicle successfully passes the object one the
left.
2. Right-Pass: The ego-vehicle successfully passes the object one the
right.
3. Crash: The ego-vehicle and the obstacle collide.

If you are not able to download the dataset, please send a request to Dr. Alaa AlZoubi (a.s.alzoubi@cranfield.ac.uk)