BDD100K dataset exposes drivers to surveillance, data misuse

The BDD100K self-driving video dataset raises significant privacy concerns, according to researchers.

The dataset was found to contain footage of individuals (pedestrians) and vehicles (licence plates) captured in public spaces without their explicit consent, raising ethical issues around privacy violations and the potential misuse of personal data.

The high-resolution videos and images in the dataset could potentially enable identification of specific individuals or vehicles, exposing them to risks like surveillance, tracking, or other privacy intrusions, thereby compromising privacy and civil liberties.

While efforts were made to blur faces and license plates, the researchers showed many instances where these privacy protection measures failed, leaving identifiable personal information exposed. Larger faces and plates were especially prone to being inadequately anonymised.

The finding highlighted the need for robust privacy-preserving techniques and guidelines when collecting and using datasets incorporating sensitive personal information.

Research, advocacy 🧮

• Wilson B., Hoffman J., Morgenstern J. (2019). Predictive Inequity in Object Detection (pdf)