1. Vehicle-Color-Identification (https://github.
com/jwhabi/Vehicle-Color-Identification )
This project aims to improve vehicle color recognition in Intelligent Traffic Systems (ITS)
by addressing challenges like weather conditions and light pollution. The approach
involves dehazing images for clarity and using a convolutional network for color
classification. The project utilizes a dataset of 15,601 vehicle images in eight colors,
captured under varying urban conditions, to enhance the accuracy of vehicle colour
identification.
My simple notebook tests the trained weight (predictions weren’t accurate):
vcol.ipynb
2. Vehicle Color-24: https://github.com/mendy-2013/Vehicle-Color-24-Dataset
3. DVM-CAR A Large-Scale Dataset for Automotive Applications
4. VCoR (Vehicle Color Recognition) Dataset:
Images in this dataset were sourced from google. The were further cleaned up,
post-processed, annotated and consolidated into a dataset. VCoR is a large scale and
most diverse Vehicle color recognition dataset. VCoR contains 10k+ image samples and
15 color classes which is almost twice as diverse as the largest existing dataset. The 15
color categories represent the most popular vehicle color models according to CarMax,
including: white, black, grey, silver, red, blue, brown, green, beige, orange, gold, yellow,
purple, pink, and tan.
5. Vehicle Color Classification Under DifferentLighting Conditions Through Color Correction
This paper introduces a novel method for classifying vehicle colors into seven
categories, especially under varying lighting conditions. The technique involves two main
steps:
a. Color Correction: To minimize the effects of lighting, a mapping function is
developed to reduce color distortions between frames. This helps in maintaining
consistency in vehicle colors across different lighting scenarios.
b. Handling Specular Highlights and Shade Colors: The method addresses the
issue of specular highlights (like reflections making a vehicle's window appear
white) by removing windows from the analysis. This ensures that the vehicle's
color is more uniformly represented. Additionally, the technique efficiently
differentiates between similar shades (like white vs silver or black vs navy), a
common issue in traditional methods.
To achieve this, a tree-based classifier is used, first separating vehicles into chromatic
and nonchromatic categories based on their shade intensity, and then further classifying
them into detailed color categories. This approach significantly enhances the accuracy of
vehicle color classification, even with vehicles in various shades and under different
lighting conditions. The paper utilizes methods like Support Vector Machine (SVM) and a
novel vehicle window removal task in its approach.
