EDIS-Net for Lane Markings Detection

The dataset of around 15.3K image frames is created from Malaysia's different places to test the framework. For instance, the videos were taken at Melaka KL expressway, KL-KLIA, Melaka-JB expressway, Melaka-Muar, Sungai Udang, Jalan UPM and Jalan Ayer Keroh. A high-resolution mobile camera was mounted beside the steering to capture the videos of lane markings. It can record the video in different resolutions like 4k, 60/30 fps and 1080/720, 30/60/240 fps. The videos were taken in the mode of 1080, 30 fps, and converted into 1920 ×1080 image frames. Then, the image frames' dimension is changed into 224×224 to make it compatible with the framework. The dataset is classified into three categories: lane markings pattern, illumination variation, and natural climate. The lane markings pattern contains 7.9k image frames of the solid lane, double line lane, curve lane, and vehicle obstacle. Again, illumination variation includes 5k image frames of low light, dark light, and cloudy. And about 2.5k rainy image frames are arranged in the natural climate category. All the images are in jpg form.

本数据集包含约15300张图像帧，采集自马来西亚多个地点，用于测试该框架。采集场景涵盖马六甲-吉隆坡高速公路、吉隆坡-吉隆坡国际机场高速公路、马六甲-新山高速公路、马六甲-麻坡路段、双溪乌栈、UPM路（Jalan UPM）以及亚逸牙律路（Jalan Ayer Keroh）。研发团队采用安装于方向盘旁的高清移动摄像头采集车道标线视频，支持多种分辨率录制模式，包括4K、60/30 fps，以及1080p/720p、30/60/240 fps。本次采集的视频以1080p、30 fps模式录制，随后转换为1920×1080分辨率的图像帧，再将所有图像帧统一调整至224×224分辨率，以适配该框架。本数据集分为三大类别：车道标线模式、光照变化以及自然气候。其中车道标线模式类别包含7900张图像帧，涵盖实线车道、双白线车道、弯道车道以及车辆障碍物场景；光照变化类别包含5000张图像帧，涵盖弱光、暗光以及阴天场景；自然气候类别包含约2500张雨天场景的图像帧。所有图像均采用jpg格式存储。