Abstract: A method and apparatus are provided for capturing video so that compression and quality can be optimized. During operation, a video recording system will employ a learning algorithm to determine periods when a light bar is on, or active. Reference Intra frames are then stored and used for the subsequent creation of predictive frames. More particularly, at least a first Intra frame is stored and used for creating predictive frames during periods of light bar activity. In a similar manner, a second Intra frame is stored and used for creating predictive frames during periods of light bar inactivity. By learning the light bar pattern, Intra frames can be more intelligently selected, resulting in optimized compression and quality.