A Novel Part-Based Model for Fine-Grained Vehicle Recognition

Abstract

In recent years, fine-grained vehicle recognition has been one of the essential tasks in Intelligent Traffic System (ITS) and has a multitude of applications, such as highway toll, parking intelligent management and vehicle safety monitoring. Fine-grained vehicle recognition is a challenging problem because of small inter-class distance and substantial sub-classes. To tackle this task, we propose a part-based model for fine-grained vehicle recognition in a weakly unsupervised manner. We also provide a part location method that locates the discriminative parts based on saliency maps which can be easily obtained by a single back-propagation pass. The advantage of the method is that the resolution of saliency maps is the same as the resolution of input images. Thus, we can locate discriminative parts efficiently and accurately. Additionally, we combine the whole-level features and part-level features and improve the accuracy of recognition up to 98.41% over 281 vehicle models in the large-scale dataset CompCars.