An approach is provided for compressing video frames captured by a camera. Clusters of the video frames are determined, where the clusters have correlations to respective sets of values of meta data associated with the video frames. The meta data includes a location, focal length, and direction of the camera, and a time of day during a capture of the video frames by the camera. Based on multiple video frames within each of the clusters being compared with each other but not being compared with a video frame in another cluster, base and delta frames are determined and stored for each of the clusters. An amount of storage required for the base and delta frames is based in part on each of the correlations to the respective sets of values of the meta data.