Anonymous Video Processing for Live Street View

We propose a method for hiding the identity of persons and objects in video sequences. This type of video sequence will be used in applications where the scene is important but the identity of persons should be unrecognizable, e.g., street view applications. Other parts, such as company logos, names on house doors or license plates on parked cars should not be recognizable as well. This is today solved in street view by blurring out parts of images that are compressed with standard encoding techniques. We propose a method that automatically will blur out parts that are moving but still show the motion. Our technique requires very low bit rates for video transmission, almost as low as for a still image. The low bit rate and the low computational complexity will enable video views of streets at the same costs as for still images.