Deep learning and mobile-based smart parking system approach in IoT-based smart cities

In the era of Internet of Things (IoT) and smart city ecosystems, innovative smart parking systems are needed for more sustainable cities. With the increasing number of vehicles, smart parking systems are among the important issues in smart cities. The reason for this is that the search for insufficient parking places brings with it serious cost, air pollution and stress problems. The solutions that have been researched on smart parking systems are no longer sufficient. In this thesis, a new deep learning and cloud-based mobile smart parking application design has been developed to minimize the problem of drivers searching for parking spaces. Within the application, a service based on deep learning with LSTM has been developed to prediction the parking location. Here, dynamic access to the previously created LSTM-based model is provided via the user's mobile device, and the related parameters are entered and the occupancy rates of the parks are displayed on the mobile device. In this way, both energy and time savings were achieved. With real-time parking data collected in Istanbul, Turkey, results with an accuracy of 99.57% were obtained. To demonstrate the effectiveness of the proposed model, it was compared with machine learning models, SVM, RF and ARIMA methods. In addition, the weather data, which has a significant impact on the use of the parking lot, was taken from AKOM and combined with the parking data. Experiments were carried out on this data set with 27 different models created from RNN, LSTM and GRU methods, which are widely used in time series prediction problems and have proven their success. GRU deep learning model gave the best results with 99.11% accuracy and 0.90 MAE, 2.35 MSE and 1.53 RMSE metric values. Experimental results with various hyper parameters clearly demonstrate the success of the GRU deep learning model in predicting park occupancy rates. When the results obtained are examined, it has been shown that deep learning models can be used in smart parking systems for more sustainable smart cities, with high accuracy in parking loT prediction.