Classification of Vehicle Types in Car Parks using Computer Vision Techniques

Abstract

The growing population of big cities has led to certain issues, such as overloaded car parks. Ubiquitous systems can help to increase the capacity through an efficient usage of existing parking slots. In this case, cars are recognized during the entrance phases in order to guide them automatically to a proper slot for space-saving reasons. Prior to this step, it is necessary to determine the size of vehicles. In this work, we analyze different methods for vehicle classification and size measurement using the existing hardware of car parks. Computer vision techniques are applied for extracting information out of video streams of existing security cameras. For streams with lower resolution, a method is introduced figuring out width and height of a car with the help of reference objects. For streams with a higher resolution, a second approach is applied using face recognition algorithms and a training database in order to classify car types. Our evaluation of a real-life scenario at a major German airport showed a small error deviation of just a few centimeters for the fist method. For the type classification approach, an applicable accuracy of over 80 percent with up to 100 percent in certain cases have been achieved. Given these results, the performed methods show high potentials for a suitable determination of vehicles based on installed security cameras.