Besides, analysing the pixels directly, maybe their algorithm take into account the vibration/movement of the live video to detect depth.