Framework for Vehicle Make and Model Recognition-A New Large-Scale Dataset and an Efficient Two-Branch-Two-Stage Deep...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems (ITS)

Background:

• Vehicle Make and Model Recognition (VMMR) is vital for Intelligent Transportation Systems (ITS), enabling applications like intelligent surveillance and autonomous driving.
• Existing VMMR systems require accurate and efficient performance in real-world scenarios.

Purpose of the Study:

• To introduce a new large-scale dataset, Diverse large-scale VMM (DVMM), featuring popular European vehicle brands.
• To propose a novel two-branch deep learning framework for enhanced VMMR.
• To develop a new metric for evaluating classification confusion in VMMR.

Main Methods:

• A novel two-branch deep learning architecture was designed for separate make and model recognition.
• A two-stage training procedure and a unique decision module were implemented for processing predictions.
• A new metric based on the true positive rate was introduced to assess classification confusion.

Main Results:

• The proposed framework achieved 93.95% accuracy on the DVMM dataset and 95.85% on traditional datasets.
• The two-branch approach demonstrated superior performance over one-branch methods across various dataset scales.
• The method significantly reduced vehicle model confusion and inter-make ambiguity.

Conclusions:

• The proposed DVMM dataset is general, diverse, and practical for VMMR research.
• The novel two-branch VMMR paradigm offers improved robustness and reduced confusion compared to single-branch designs.
• The study highlights the effectiveness of the proposed deep learning framework for accurate vehicle recognition in ITS.