A Parking Monitoring System Using FMCW Radars

Parking monitoring systems provide the availability and statistics of parking space in an indoor or outdoor parking lot. Image recognition-based parking monitoring systems are popular since their strong link to billing systems. With privacy concern, those camera-based systems are not suitable for parking spaces monitoring. Hence, the radar-based parking monitoring system is built in this study. By combining the TI 77GHz FMCW radar, Raspberry PI (RPI) single board computer, and image-based auto-labeling service, an automatic radar heatmap collector is constructed in the data collection phase. Those long term collected data is then employed to train and a CNN-based model is developed. Due to the coupling effects of radar signals on two adjacent parking space, the availability of six adjacent parking spaces is defined to be the parking status and denoted by a 6-digit binary code. The trained mode is then ported to several GPU-based edge computing platforms to infer the parking status. The main advantages of radar-based parking monitoring system are excellent inference results in dimmed environment, privacy-protected and lightweight computing load. Even the on-device inference using RPI platform gives a prompt response with very satisfied results.