Towards Real-time Anomaly Detection and Calibration for Large Parking Lot in Urban Complex

With the development of low-cost, low-power sensing and communication technologies, there has been growing interest in the IoT for realizing smart cities, in order to maximize the productivity and reliability of urban infrastructure. One of the most representative examples is smart parking system, which consists of parking sensors and backend servers. In existing systems, parking sensors are placed in each parking spot to monitor the status of the parking spot, and then the sensing data is send to backend servers for further processing. Based on these sensor data, the smart parking system can provide some smart services, such as parking spot availability prediction, remote booking and parking guidance. The premise of all these smart services is that the sensor data is accurate and reliable. However, as the sensor ages, the sensor may produce some unreliable data, which brings great challenges to our smart services. In this paper, we present a novel anomaly detection and calibration system in "Suzhou Center", one of the largest most advanced urban complexes in China. Our system uses supervised machine learning algorithms, focusing on capturing spatio-temporal features. The experimental results show that our system can identify most anomalies and improve the accuracy of the sensor.