Data Acquisition

The data was collected in urban Montreal and spans 66 km of urban roads, amounting to 14 TB of storage. The dataset includes:

• 2.8 million images collected at 30 Hz by the four AR0231 cameras
• 42250 images at 25 Hz captured by the C920 camera during sound measurements
• 480240 64-channel lidar point clouds recorded at 20 Hz
• 81.1 billion sound pressure samples recorded from the 1024 microphones

All measurements were time-stamped and synchronized with GNSS as time reference. The microphone array signals were recorded at 10 second intervals at a sampling rate of 46.875 kHz. All acoustic captures are highlighted in the map below.

During the data acquisition, to minimize the effect of high winds on the sound readings, a speeds were maintained between 30km/h to 40km/h.

Map

Map displaying the data acquisition route. Measurements from all sensors are available for red sections. Blue sections are missing data from the microphone array and RGB camera.

Traffic Diversity

This graph highlights the instances of traffic agents in the dataset across 6 different classes, with cars being the vast majority.

Distribution of Data

The dataset covers both day and night scenes, variety of weather conditions, neighborhoods and scenery and includes capture sequences with our prototype capture vehicle moving as well as being stationary.