Data Augmentation Method for Improving Vehicle Detection and Recognition Performance

Vehicle detection and recognition are now implemented through powerful machine learning methods, those methods are not only relying on clever learning strategies, but also require high quality datasets. To obtain high quality datasets, time-consuming and grueling processing is needed. In this research, we proposed a data augmentation concept that is able to prepare high quality datasets for vehicle detection and recognition training in a more efficient approach. The effectiveness of our proposed data augmentation concept has been proved by applying it on the training data preparing process of YOLOv4 model. The result shows that the mean average precision (mAP) had increased 1.93% comparing to the YOLOv4 model which was trained without data augmentation.