The other way around may be easier, i.e. excluding known datacenter ranges. There are some commercial databases for that, i'm not sure if there are any free ones. But you can also do this manually by running a whois on an IP and then extracting the ranges from the whois response and caching then. Then you can look at the orgname or something like that. You can also download the whois databases from the RIRs, but they don't contain the information what kind of entities they are.
