Cost effective and accurate vehicle make/model recognition method using YoloV5

Automatic meta-data extraction from images from highway cameras is a necessary component for intelligent transportation and smart city. Meta-data can include detailed information on vehicles, such as car make/model, car registration plate and drivers' behaviour, etc.. This paper focuses on real-time car make/model information extraction from highway cameras. As we have very limited access to the real world data due to data privacy and protection, we use open-source data (e.g. car selling websites) and transfer learning on open-source pre-trained models to build a model which is generic enough to be applied directly to similar data sets from other sources, (e.g. real-world highway cameras) without losing much accuracy. To achieve this, we propose applying the object detection method 'You Only Look Once' (Yolo) for classification problem of car make/model. The proposed method and trained model achieve an accuracy of 95.6% when applied directly to real-world highway cameras without using their data for training.