You can use ML to summarize the video and you can store the most interesting bits in low quality - instead of actual video you can just store something like "red car crossed intersection going northbound, male with blue cap stopped at light, ..."