Vacant Parking Space Detection Based on DenseNet-121 with Swin Transformer in Smart Cities

The development of urban mobility imposes severe challenges to smart cities' transportation infrastructure, such as controlling accidents, gridlocks, and Vacant parking spaces (VPS). The availability of parking lots has become a serious problem in many cities around the world. Lately, research has revealed that drivers searching for VPSs impact up to 40% of traffic gridlock. To overcome this issue, we propose a VPS detection model based on the improved hybrid Deep learning model. In this paper, the spatial attributes are captured using a Densely Connected Convolutional Network 121 (DenseNet121) and then the attention network method assigns varying weights to the features, decreasing the model's computational cost and enhancing the model's performance. A Swin Transformer captures long-distance dependent information features, capturing deep discriminative features. Finally, a SoftMax technique is employed for the model classification. The proposed model is evaluated using surveillance camera frames on the PKLot and CNRPark datasets. The results show that the proposed method can achieve the best accuracy of 99.85% and 98.34% for the PKLot and CNRPark datasets, respectively. The results obtained show that the proposed approach improved VPS prediction compared to other well-known state-of-the-art VPS methods.