Image Processing Versus Deep Learning for People Counting

People counting is important for crowd control in various kinds of facilities. Sensor devices are useful; however, multiple devices are needed when people overlap. On the other hand, one camera can obtain multiple types of information, such as gender, age, and clothing. Then, image processing-based methods were traditionally used, while deep learning-based approaches are now leveraged, and several methods are competing with each other. Deep learning-based methods, however, depend on the training data, and the results are affected by the properties of the database, whereas image processing-based approaches can be controlled by the algorithm, and few studies have compared the two methods. In this paper, image processing- and deep learning-based approaches for people counting are proposed and compared in terms of precision and processing time.