A Unified Framework for Vehicle Recognition

Abstract

This thesis addresses the need for a unified vehicle recognition system that integrates Vehicle Make and Model Recognition (VMMR), Automatic Number Plate Recognition (ANPR), and Car Color Detection. The existing conventional vehicle recognition systems suffer from fragmentation in application and cause inefficiency in traffic management, law enforcement, and access control. In this study, innovative convolution neural network (CNN) architectures such as MobileNetV2 for fine-grained vehicle make/model detection have been proposed. In the proposed framework, license plate detection is carried out using the YOLO approach before detection and color classification under different conditions, with advanced algorithms applied using histogram-based grouping methods and HSV color space analysis. The rich dataset of car manufacturers and models helps VMMR generalize but be accurate, while license plate design variations and ambient conditions are accounted for in ANPR to make it more efficient. The integrated modules shall then provide the unified system with full vehicle identification to complement security assessments and operation decisions on registered vehicles. The techniques of adaptive image calibration constitute another focal point of the proposed framework since consistency from the varying lighting and weather conditions improves reliability in the recognition system. Through extensive testing, validation, and integration, this unified framework is proved practical and effective, setting up the next generation of AI-driven vehicle recognition systems able to meet, at the best level, the most complex objectives of traffic management, law enforcement, and security. This research contributes to the development of unified approach of detecting vehicles, which brings forward a solution to current limitations in intelligent transportation systems and pushes the frontier of vehicle recognition technology to a new state-of-the-art.