Highway Emergency Lane Obstacle Detection & V2I Dataset

This dataset collects data from an autonomous vehicle in a highway scenario, focusing on the detection of stationary vehicles in the emergency lane and the subsequent Vehicle-to-Infrastructure (V2I) communication. AUTOMATED VEHICLE DATA COLLECTOR The vehicle used for data collection is an autonomous car prototype retrofitted with the following sensor suite: Inertial Navigation System: OxTS GNSS/INS solution providing global position, IMU data, and odometry. LiDAR: 1x Ouster 360° LiDAR. Cameras: 4x Lucid Vision cameras (Front Center, Front Left, Front Right, and Rear recorded). Radars: 4x Continental ARS430 radars providing objects list, near and far range detections. Vehicle Sensors: Wheel speed sensors and internal vehicle feedback. and controllable actuators: Steering & Braking: Wire-controlled actuators for lateral and longitudinal control. The system is designed to detect the position, velocity, and type of other road users. For this specific task, the perception stack utilizes neural networks for image segmentation to build a local map of the lanes. An algorithm processes this data to identify vehicles specifically stopped in the emergency lane. DATASET CONTENT The dataset includes a ~22-second recording in a real highway environment. It contains: ROS2 Bag (.mcap): Contains all raw sensor data, system status, perception outputs and the message that will be sent to the Infrastructure. Visualization Video: Shows the algorithm output on BEV (bounding boxes of detected vehicles; red highlighting for the vehicle detected outside the carriageway) + raw cameras. To replicate the video you need to run the ros2 bag and launch tviz2. On rviz2 open the provided config 'lane_keeping.rviz'. Key ROS2 Topics included (relevant list): Sensors (Raw Data): /ouster_sensor_center/points (LiDAR Pointcloud) /lucid_vision/*/image_rect/compressed (Camera images) /ces_ars430di/sensor_*/objects_list (Radar object list detections) /oxts/nav_sat_fix & /oxts/imu (GNSS and Inertial data) State Estimation & Localization: /estimation/ekf/vehicle_state (Fused vehicle state) Perception & Scene Understanding: /perception/target_tracking/obstacles_list (Tracked object list) /estimation/multilane_prediction/bev_image/compressed (Bird's Eye View representation of lanes) /perception/positioning_algorithm/denm_stationary_vehicles (V2I Alert Message) SCENARIOS Highway: The ego-vehicle travels on the highway while a target vehicle is stationary in the emergency lane. The system detects the stationary vehicle, classifies its position relative to the lanes, and triggers a V2I warning message. Location: A50 Tangenziale Ovest di Milano (Rozzano/Assago area), Italy. DATASET ACCESS To request access to the dataset please contact: Giulio Panzani, Politecnico di Milano, [e-mail] Riccardo Pieroni, Politecnico di Milano, [e-mail]