基于RGBD数据车厢安全防控数据集

该数据集面向城市交通营运安全防控核心研究，针对现有数据集缺乏车厢环境多样化、真实行驶数据的痛点建设。数据来源于广州市番禺城市道路行驶中的车厢环境，由华南理工大学及相关合作单位基于RGBD传感器生成，通过“模拟场景 + 真实采集”相结合的方式，组织专业人员模拟司乘不安全行为及设施异常情况，实时拍摄采集不同行驶状态、光照条件、乘客数量下的场景数据。数据集具有重要实践意义，可为车厢场景司乘人员不安全行为识别、重要设施异常检测等技术研究提供核心数据支撑，助力提升城市交通系统运营安全水平与智能化治理能力。数据集包含17个类别，涵盖视频、图片、实验数据、统计数据及支撑材料等。视频长度2秒-20分钟、帧率30帧/秒，图片400多张，实验数据含多个模型权重文件，统计数据包含人物检测精度、关键点定位精度等评估图表，配套5篇高影响因子论文和3项专利。数据划分可参考训练与测试文件列表，模型训练依托配套权重文件及算法包开展，性能评估基于训练轮数、准确率等指标。数据集总计约8.91GB。

This dataset is developed for core research on urban traffic operation safety prevention and control, addressing the pain point that existing datasets lack diversified in-car environments and real driving data. The data is collected from the in-car environment during driving on urban roads in Panyu District, Guangzhou, and generated by South China University of Technology and its cooperating partners based on RGBD sensors. Adopting the combined approach of "simulated scenarios + real collection", professional personnel were organized to simulate unsafe behaviors of drivers and passengers and abnormal facility conditions, and real-time shooting and collection of scene data under different driving states, lighting conditions and passenger numbers were carried out. This dataset has important practical significance, providing core data support for technical research such as unsafe behavior recognition of drivers and passengers in in-car scenarios and abnormal facility detection, helping to improve the operational safety level and intelligent governance capability of urban traffic systems. The dataset contains 17 categories, covering videos, images, experimental data, statistical data and supporting materials, etc. The videos range from 2 seconds to 20 minutes in length with a frame rate of 30 frames per second; there are more than 400 images; the experimental data includes multiple model weight files; the statistical data contains evaluation charts such as human detection accuracy and key point positioning accuracy; and it is accompanied by 5 papers with high impact factors and 3 patents. The data division can refer to the training and test file lists; model training is carried out relying on the supporting weight files and algorithm packages; performance evaluation is based on indicators such as training epochs and accuracy. The total size of the dataset is approximately 8.91 GB.

搜集汇总

背景与挑战

背景概述

该数据集面向城市交通营运安全防控研究，针对车厢环境数据缺乏多样性和真实性的问题，由华南理工大学及相关单位基于RGBD传感器在广州市番禺城市道路行驶环境中采集生成。它采用'模拟场景+真实采集'方式，模拟司乘不安全行为和设施异常，涵盖不同行驶状态、光照和乘客数量下的场景数据，为不安全行为识别和异常检测提供核心支撑。数据集包含视频、图片、实验数据等17个类别，总计约8.91GB，旨在提升交通运营安全和智能化治理水平。