In this paper, we present an object detection based algorithm to automatically map the parking spaces in a parking lot, instead of manually mapping them. The work addresses an important gap in the recent computer vision based artificial intelligence techniques to build smart parking systems. We test our approach with two of the most popular object detectors, Faster R-CNN and YOLOv4. Our results show that our approach decreases the human effort needed by up to a compelling 90%. We show that the percentage of the available parking spots that are automatically detected through our approach accumulates over time and, in theory, can approach a 100%, on a day when all the parking spots are fully occupied. In other words, the approach is designed to have its highest performance over a busy parking lot during the busiest time.