BMW is expanding its use of video and sensor data from customer cars to train and improve driver-assistance systems using real-world traffic situations.
The programme covers selected vehicles running BMW Operating System X, including the new iX3, i3, X5 and 7 Series. Collection began in Germany in April and has since expanded across the European Economic Area. Participation requires the customer’s prior consent.
Unlike continuous dashcam recording, BMW’s system is triggered only by specific driving events. These can include activation of the emergency braking system, hard braking, an evasive manoeuvre, an unexpected obstacle or particular lane-change situations.
Each event can generate up to 120 seconds of video, together with information from radar and ultrasonic sensors, vehicle speed, steering angle and location data. Exterior cameras can inevitably capture pedestrians, cyclists, other vehicles and number plates.
From September, BMW is gradually allowing relevant recordings to be transmitted to its back-end systems without first blurring faces and number plates. The company says this is necessary because details such as pedestrian gaze and head movement can be useful when training machine-learning systems.
BMW said the footage is analysed automatically in a secured environment and is not used to identify individual road users. The vehicle identification number is removed after transmission so the development data can no longer be linked to the recording vehicle.
If BMW engineers need to view an individual recording, faces and number plates are obscured beforehand as far as technically possible.
The real-world material supplements BMW’s simulation work and development-vehicle testing. Data gathered from unusual traffic situations can then be used to refine functions such as emergency braking, lane-change assistance, adaptive cruise control and automated parking.
