Localização e rastreamento de veículos através da identificação de padrões de imagens utilizando redes neurais

Abstract

The present work proposes a vehicle tracking and location system based on traffic images, using image processing techniques and neural networks. The project is based on the analysis and comparison of color histograms for the classification and identification of patterns in a sequence of detections performed by the YOLOv8 software (You Only Look Once, version 8). In the histogram comparison process, the Hellinger Distance method will be used, employing the Bhattacharyya coefficient to determine if the distance corresponds to the possibility of a vehicle being the same along a specific route. The main objective of the proposed work is to provide an alternative solution to existing ones in the market, such as the ByteTrack tracking method present in the YOLOv8 software. The proposed method demonstrated good performance under various test conditions, including diferente times of the day, weather conditions, and traffic intensities. The research highlighted the need for improvements in conflict management related to vehicle enumeration and performance optimization, with great potential to explore continuous learning techniques and integrations with other tracking frameworks. The approach was compared with YOLO’s ByteTrack algorithm, revealing a significant difference in performance and tracking accuracy, with ByteTrack outperforming in all cases. Vehicle tracking in traffic offers a range of benefits, from increased security, enabling quick location in case of theft, to the ability to monitor routes and delivery times, improving the efficiency of logistical operations.