Vehicle Type and Speed Detection on Android Devices Using YOLO V5 and MobileNet

Vehicle-type detection tool has many applications in transportation, traffic control, guiding and controlling unmanned vehicles, tolls and road taxes, traffic violations, smuggling detection, etc. In the proposed version, the MobileNet neural network and the YOLO V5 algorithm are integrated. In this integration, the YOLO V5 algorithm replaces the convolutional layers of the neural network and the neural network be used for the classification of vehicles. The Kivy library is employed to transform the developed algorithm into an Android application. The data used in this study consists of two datasets: The ImageNet database and a constructed database. The proposed method results show improvement in increasing the accuracy of vehicle detection, reducing the computational load, detection accuracy in different weather conditions, separating overlapping cars. Various methods are presented for better neural network training and reducing neural network size. The reason for these capabilities is the use of developed algorithms and the use of techniques such as data augmentation, spatial filtering, and distillation.