Cockpit Video Coding with Temporal Prediction

Recording an airplane cockpit screen is a challenging task since video codecs hardly preserve text details at the low bitrates required by avionic applications. We recently proposed a scheme for semantic compression of airplane cockpit video that preserves the readability of text while meeting bitrate and encoder complexity constraints. Within each frame, text is segmented from the video and encoded as character strings rather than as pixels. Text in the screen is then inpainted, producing a residual video with few high frequency components easily encodable with standard codecs. The residual video is transmitted with the encoded text as side-information. At the receiver side, characters are synthesized atop the decoded residual video, leaving the text unaffected by compression artefacts. In this work, we evaluate our scheme with multiple video codecs with different prediction schemes, producing novel experimental evidence in terms of attainable ratedistortion performance and highlighting directions for future research.