Unlocking the smartphone's senses for smart city parking

Studies have shown that in places like New York City drivers often spend over 20 min looking for parking, contributing to as much as 30% of the total traffic. In response, cities like San Francisco have deployed systems capable of pointing drivers to the closest available parking spot. Unfortunately, such systems have gained little traction as they rely on specialized infrastructure that is expensive to build and maintain. We present SmartPark, a smartphone based system that relaxes the requirement for specialized infrastructure by relying on the smartphone's sensors and ubiquitous Wi-Fi and cellular infrastructure. To accomplish this, SmartPark addresses two major challenges, under the constraint of minimum impact on battery life: automatic transportation mode detection and location matching. Solved together, they enable SmartPark to automatically detect when a user pulls out of a parking spot, making it available again. It addresses each challenge using a combination of thorough statistical analysis of the sensor readings and a novel Random Forest based classification algorithm. Experimental results from 12 volunteers, using 7 different smartphones, in 3 different cities show that SmartPark can distinguish 9 different transportation modes with accuracy between 95.57-100%, enabling it to correctly detect unparking events virtually 100% of the time. This is accomplished with a minimum impact on battery life-running SmartPark on a fully charged LG Google Nexus 5 for 5 h straight caused the battery level to drop only about 4%.