交通目标检测视频数据

交通目标检测视频数据是分别在2020年12月、2020年7月，采用卡口相机在贵阳市市内、武汉市环路高架/高速/大桥上等自然场景采集到的交通视频数据，以及通过算法对视频中的车辆、非机动车等检测数据，包括视频帧数位置、车框位置、车辆的类别（机动车、非机动车）、车辆的跟踪编号。数据分析是采用人工与算法相结合的数据清洗模式，提取了视频中多目标快速运动片段，确保数据交付质量。目标打标签过程，经过多轮人工审核，提升了学习质量。数据量428MB。

The traffic object detection video dataset was collected in December 2020 and July 2020 via traffic bayonet cameras in natural scenarios including urban areas of Guiyang, as well as ring elevated roads, expressways and bridges in Wuhan. It also contains algorithm-generated detection data for vehicles, non-motor vehicles and other targets in the videos, covering video frame indices, vehicle bounding box coordinates, vehicle categories (motor vehicles, non-motor vehicles), and vehicle tracking IDs. For data cleaning, a hybrid workflow combining manual labor and algorithms was adopted, and fast-moving multi-target segments in the videos were extracted to ensure the quality of the delivered dataset. The target annotation process underwent multiple rounds of manual reviews to improve the quality of model learning. The total size of this dataset is 428 MB.

搜集汇总

背景与挑战

背景概述

该数据集是2020年在贵阳市和武汉市自然场景中采集的交通视频数据，包含车辆和非机动车的检测信息，如视频帧位置、车框位置、车辆类别和跟踪编号。数据经过人工与算法结合的清洗和多轮人工审核，确保质量，总数据量为428MB。