Toward Enhancing Vehicle Color Recognition in Adverse Conditions: A Dataset and Benchmark

TL;DR

The results demonstrate that the proposed dataset poses greater difficulty for the tested models and highlights scenarios that require further exploration in VCR, while also offering valuable insights for the field of fine-grained vehicle classification.

Abstract

Vehicle information recognition is crucial in various practical domains, particularly in criminal investigations. Vehicle Color Recognition (VCR) has garnered significant research interest because color is a visually distinguishable attribute of vehicles and is less affected by partial occlusion and changes in viewpoint. Despite the success of existing methods for this task, the relatively low complexity of the datasets used in the literature has been largely overlooked. This research addresses this gap by compiling a new dataset representing a more challenging VCR scenario. The images - sourced from six license plate recognition datasets - are categorized into eleven colors, and their annotations were validated using official vehicle registration information. We evaluate the performance of four deep learning models on a widely adopted dataset and our proposed dataset to establish a benchmark. The results demonstrate that our dataset poses greater difficulty for the tested models and highlights scenarios that require further exploration in VCR. Remarkably, nighttime scenes account for a significant portion of the errors made by the best-performing model. This research provides a foundation for future studies on VCR, while also offering valuable insights for the field of fine-grained vehicle classification.

Figures (7)

• Figure 1: Examples of images from the datasets proposed in chen2014vehicle (a) and in this work (b), with the corresponding vehicle color annotation shown above each image. Observe that images in the proposed dataset (b) depict significantly more challenging scenes than those in (a), featuring adverse conditions such as nighttime settings and vehicles from various viewpoints.
• Figure 2: Distribution of vehicle colors in the UFPR-VCR dataset.
• Figure 3: Illustration depicting the process of extracting vehicle patches from images in the RodoSol-ALPR dataset, which lacks vehicle position labels.
• Figure 4: Examples of discarded motorcycle images: (a) and (c) were sourced from the RodoSol-ALPR dataset laroca2022cross, while (b) was extracted from the UFPR-ALPR dataset laroca2018robust. Below each image is the corresponding motorcycle's color. In this figure, the original images were slightly resized for better viewing.
• Figure 5: Examples of images excluded due to vehicles with multiple colors (a) and those partially outside the image frame (b, c). The colors of the vehicles in (b) and (c) cannot be visually determined solely from their front bumpers. Note that the vehicle images in this figure were resized for improved visibility.