IMLP: Iranian Motorcycle License Plates

The Iranian Motorcycle License Plates (IMLP) dataset includes 1089 different images of motorcycles, containing in total 1191 Iranian motorcycle license plates. These images were captured and collected from the cities of Islamabad Gharb, Kermanshah, Karaj, and Tehran, and some were also obtained from the Instagram social network. Initially, motorcycles were filmed, and then one or two frames were selected for the dataset. The final images were randomly chosen from over 50,000 frames.The dataset was built to ensure diversity in conditions:Captured using different cameras (Samsung S23 Ultra, Xiaomi Note 10 5G, Xiaomi Note 11s, Samsung J6, and several other mobile phones, as well as social network images).Collected from different angles, distances, and lighting conditions (daytime and nighttime).For annotation and benchmarking, deep learning networks YOLOv8, SSD, and Faster-RCNN were employed. The motorcycle plates were identified in a two-stage process:Detection of the plate region.Recognition (OCR) of the digits.The dataset is organized into two main parts: Detect and OCR, each containing train, test, and validation splits. Every folder includes images and annotations compatible with YOLOv8, SSD, and Faster-RCNN. In addition, implementation files for these networks are provided to facilitate reproducibility.This dataset and the associated Python code are linked to the following peer-reviewed publication:Article: Deep Learning Based System for Automatic Motorcycle License Plates Detection and Recognition Authors: Abdolhossein Fathi, Babak Moradi, Iman Zarei, Afshin Shirbandi Journal: Signal, Image and Video Processing (Springer), Volume 18, Pages 8869–8879, 2024 DOI: https://doi.org/10.1007/s11760-024-03514-5In this article, the IMLP dataset was introduced and evaluated using YOLOv8, SSD, and Faster-RCNN. The results demonstrated that YOLOv8 achieved the best performance, with 98.5% accuracy in the detection stage and 99% accuracy in the recognition stage. The dataset and source code are made publicly available to support further research in intelligent transportation systems (ITS), particularly in the context of motorcycle license plate detection and recognition.

伊朗摩托车车牌（Iranian Motorcycle License Plates, IMLP）数据集包含1089张摩托车图像，总计包含1191枚伊朗摩托车车牌。该数据集的图像采集自伊朗伊斯兰阿巴德加尔布（Islamabad Gharb）、克尔曼沙赫（Kermanshah）、卡拉季（Karaj）与德黑兰（Tehran）等城市，另有部分样本来源于Instagram社交平台。原始采集素材为摩托车行驶视频，随后从中选取1至2帧用于数据集构建，最终数据集图像从超5万帧原始素材中随机遴选而出。

本数据集旨在保障场景多样性，其采集条件覆盖多维度差异：采用不同设备拍摄（包括三星S23 Ultra、小米Note 10 5G、小米Note 11s、三星J6等多款移动设备，以及社交平台来源图像）；采集角度、拍摄距离与光照条件（昼夜时段）均存在差异。

在标注与基准测试环节，本数据集采用了深度学习网络YOLOv8、SSD与Faster-RCNN。摩托车车牌的识别分为两阶段完成：首先检测车牌区域，随后对车牌字符进行光学字符识别（OCR）。

数据集分为检测（Detect）与OCR两大核心模块，每个模块均划分有训练集、测试集与验证集。所有文件夹均附带适配YOLOv8、SSD与Faster-RCNN格式的图像与标注文件，同时提供对应网络的实现代码以保障研究可复现性。

本数据集与配套Python代码关联如下同行评议学术论文：
论文标题：《基于深度学习的自动摩托车车牌检测与识别系统》
作者：Abdolhossein Fathi、Babak Moradi、Iman Zarei、Afshin Shirbandi
期刊：《Signal, Image and Video Processing》（Springer旗下），第18卷，第8869–8879页，2024年
DOI：https://doi.org/10.1007/s11760-024-03514-5

在该论文中，研究团队对IMLP数据集进行了介绍与基准测试，结果显示YOLOv8表现最优：检测阶段准确率达98.5%，字符识别阶段准确率达99%。本数据集与源代码已公开发布，旨在为智能交通系统（intelligent transportation systems, ITS）领域，尤其是摩托车车牌检测与识别方向的后续研究提供支撑。