With the proliferation of sensors it is becoming increasingly difficult to discover and access sensor data of interest. Currently, most researchers and sensor data users access data from sensors that they build by themselves or from known sensor network run by their friends; rarely do they try to find sensor data of interest that are maintained by other users and groups. Even if they are able to find these data streams, accessing them is rather difficult because of login requirements in remote systems as well as diversity in protocols needed to access and interpret them. In our project, ROADNet, we have defined a novel way of discovering and accessing real-time data from sensor networks. Our approach is to abstract sensor streams and define systemic, descriptive and structural meta data properties for them in a catalog. Then a user can easily discover sensors of interest from the catalog, and get access to their streams. We describe a system that has more than 20,000 multi-disciplinary sensor data streams that are discovered/accessible using this approach.