Deep Learning for Robust Vehicle Identification

The level of precision of deep neural networks in visual perception tasks allows to capture crucial information from the environment for future projects, such as autonomous vehicles and smart cities. One possibility that this type of system would allow is the control and tracking of certain suspicious vehicles. Considering the use of this technology by police, it would facilitate the tracking of certain cars under investigation. With this vision, the objective of this work is the study of the current state-of-the-art of the methods and the development of a system that solves two tasks efficiently: the visual characterization and re-identification of vehicles and the license plates segmentation and character recognition. This dual identification can adapt to the environmental conditions, target distance and cameras capabilities and resolution. To test and validate this system, a custom dataset has been created to minimize the difference between lab and real environment.

Automated summary

The high precision of deep neural networks in visual perception tasks has the potential to extract important information from the environment, benefiting projects like autonomous vehicles and smart cities. This study aims to explore the latest methods and develop a system that efficiently solves two tasks: vehicle visual characterization and re-identification, and license plate segmentation and character recognition. A custom dataset is created to test and validate the system, bridging the gap between lab and real-world environments.