UFPR-ALPR

该数据集称为ufpr-alpr数据集，包括在车辆和摄像机 (在另一车辆内) 都在移动的现实场景中来自150车辆的4,500完全注释图像 (超过30,000个LP字符)。它已经在我们的IJCNN论文 [PDF] 中进行了介绍。
这些图像是用三个不同的相机获取的，并且以便携式网络图形 (PNG) 格式提供，尺寸为1,920 × 1,080像素。使用的相机是: GoPro Hero4 Silver，华为P9 Lite和iPhone 7 Plus。
我们用每个相机收集了1,500张图像，如下所示:
灰色LP的汽车900；
与红色LP的汽车300；
灰色LP的摩托车300。
数据集拆分如下: 40% 用于训练，40% 用于测试，20% 用于验证。每张图像在文本文件中都有以下注释: 拍摄图像的相机，车辆的位置和信息，例如类型 (汽车或摩托车)，制造商，型号和年份; LP的标识和位置，以及字符的位置。

This dataset is named ufpr-alpr, which contains 4,500 fully annotated images from 150 vehicles (over 30,000 License Plate (LP) characters) captured in real-world scenarios where both the vehicles and the cameras (mounted inside another vehicle) are in motion. It was introduced in our IJCNN paper [PDF].

These images were acquired using three different cameras, and are provided in Portable Network Graphics (PNG) format with a resolution of 1,920 × 1,080 pixels. The cameras used are: GoPro Hero4 Silver, Huawei P9 Lite, and iPhone 7 Plus.

We collected 1,500 images per camera, as detailed below:
- 900 cars with gray license plates (LP);
- 300 cars with red license plates (LP);
- 300 motorcycles with gray license plates (LP).

The dataset is split as follows: 40% for training, 40% for testing, and 20% for validation. Each image is paired with annotations stored in a text file, which include: the camera used to capture the image; the position and detailed information of the vehicle, such as its type (car or motorcycle), manufacturer, model and production year; the identification and bounding box of the LP, as well as the positions of each individual character on the LP.

搜集汇总

背景与挑战

背景概述

UFPR-ALPR是一个用于自动车牌识别（ALPR）的数据集，包含4,500张在移动场景中采集的完全注释图像，覆盖150辆车和超过30,000个车牌字符。图像由三种不同相机获取，尺寸为1920x1080像素，并按车辆类型和车牌颜色分布，数据集拆分用于训练、测试和验证。该数据集适用于车牌检测和字符识别研究，基于现实世界条件，支持计算机视觉和机器学习应用。