Searching for parking spots generates frustration and pollution. To address these parking problems, we present PocketParker, a crowdsourcing system using smart- phones to predict parking lot availability. PocketParker is an example of a subset of crowdsourcing we call pocket- sourcing. Pocketsourcing applications require no ex- plicit user input or additional infrastructure, running eectively without the phone leaving the user's pocket. PocketParker detects arrivals and departures by lever- Aging existing activity recognition algorithms. Detected events are used to maintain per-lot availability mod- els and respond to queries. By estimating the number of drivers not using PocketParker, a small fraction of drivers can generate accurate predictions. Our evalua- Tion shows that PocketParker quickly and correctly de- Tects parking events and is robust to the presence of hid- den drivers. Camera monitoring of several parking lots as 105 PocketParker users generated 10;827 events over 45 days shows that PocketParker was able to correctly predict lot availability 94% of the time.