Blurring the border between real and virtual parking environments

Modern multi-level indoor parking environments promise to alleviate the parking problems in modern cities but they are oftentimes stressful for human drivers. Increasing automation of the parking process has the potential for significant gains in efficiency, safety and comfort but requires highly accurate sensing and monitoring of the environment. Another challenge is the appropriate visualization of large amounts of sensor data from disparate sources, in an intuitively understandable way. We address these challenges with our platform VPIPE for realistic visualization of 3D parking environments, parking lots and sensor data of vehicles. As central building block for this platform, we propose a cost-effective camera-based parking lot monitoring system that uses a cascade of Random Forest and Artificial Neural Network classifiers. The achieved detection accuracy in our parking testbed is 94.98%.