A global leader in automotive technology wanted to anonymize personal information such as faces and license plates in videos to be used for improving vehicle safety and autonomy. Though the company had several existing video processing techniques to choose from, dealing with large amounts of video data with information in multiple frames or at different resolutions was challenging. In addition, they wanted to strike a balance by not anonymizing too much to the point of data loss or too little resulting in identification and loss of privacy.

The product we implemented started with preprocessing videos to resize or crop as needed and converting them into a suitable format. Object detection used multiple techniques to reduce false negatives, while pixelization or blurring of the faces and license plates used image processing techniques. Postprocess video ensured that the faces and license plates were effectively blurred without data loss. The engagement also included a rigorous QA process to check for identifiable information, adjust the pixelization techniques, and convert the processed video back into its original format.