Vehicle Recognition Model Based on Their Characteristics Applying YOLO-v5

Cruz E.P.; Huamani M.I.; Flores R.M. · Proceedings - 2025 3rd International Conference on Intelligent Control and Computing, IC and C 2025 (2025) · DOI: 10.1109/ic-c68228.2025.00028

Resumen

Due to the deficiency in locating vehicles involved in incidents such as theft or accidents, we propose to use V-Localize, an artificial intelligence framework for continuous vehicle search and localization captured from live traffic videos based on inputted textual descriptions and rely on the YOLO-v5 deep learning model, thus overcoming the license plate dependency of vehicles, which could be altered, damaged, or missing. The methodology includes: (i) data collection, (ii) processing, (iii) training, and (iv) data evaluation. We use the VCoR dataset with 15 color classes and VRiV to evaluate the performance on live videos. The results show that YOLO-v5 has 9% higher accuracy than other algorithms, reaching a confidence level of 0.946. Experimental results show that one of the main advantages of the proposed continuous vehicle search and location framework is that it could be integrated into the ITS software solution to assist law enforcement, especially in critical cases, such as amber alerts or hit-and-run incidents.

Instituciones: Universidad Peruana de Ciencias Aplicadas