FHWA Ear Project: Outside Vehicle and Interior Analysis Set
收藏资源简介：
Project Description This dataset was created as a follow-on to doi:10.15787/VTT1/93LFTS. Data Request Scope/Specification This dataset consists of the following six sections. 1. Blurred interior cabin snapshot images from 701 designated events. ORNL selected these events based on seatbelt usage and the number of people in the image based on the ground truth in the Event Detail Table (EDT) provided in the original study. Since the blurred images were obtained at long intervals, the images in this dataset are not necessarily concurrent with the actual 20-30 second front video clip described in the EDT, but they are from the same trip. Based on cabin snapshot availability, images were selected in the following order of preference: during the event, before the event, and after the event. 2. Dash camera video from 1186 events selected by ORNL based on hands-on-wheel criteria, turns, and hand-reaching as specified in the EDT. These videos are time-concurrent with the corresponding 20-30 second front video clip denoted in the EDT. 3. Face data, radar time series, extended time series, extended GPS data, and front camera video for selected cases of red light violations from the EDT at the time of May 2015. This consisted of 141 events which ORNL identified by filtering Driver Behavior 1, 2, and 3 to include the following signal violations: "Apparently did not see signal;" "Intentionally disregarded signal;" and "Tried to beat signal change." The face data and GPS data for crashes was only accessed in VTTI's secure enclave. The time series and GPS data for near crashes included data 2 minutes prior to event completion. 4. Radar and lane tracker time series data from 1997 designated events. ORNL selected these events based on vehicle behavior, both the driver and the outside vehicles, for research in detecting and characterizing baseline behavior, aggressive behavior by the driver, aggressive behavior by the surrounding vehicles, distracted driving, and the relative traffic density based on events from the EDT. These time series data were concurrent with the corresponding 20-30 second front video clip denoted in the EDT. 5. Rear camera videos for 198 designated events. ORNL selected these events based on rear-end collisions, tailgating, and baseline behavior observed on events found in the EDT. These videos corresponded to the 20-30 second front video clip denoted in the EDT. 6. Face camera videos for 1998 designated events. ORNL selected these events based on distraction, drowsiness, looking about the vehicle, eating, erratic or non-standard behavior, and other face-related data noted in the EDT. ORNL accessed these videos in VTTI's secure data enclave and processed the data to extract facial landmark points and non-identifying motion data. This data set supplements doi:10.15787/VTT1/93LFTS .
项目说明 本数据集为doi:10.15787/VTT1/93LFTS的后续研究数据集。 数据请求范围与规范 本数据集包含以下六个部分： 1. 701个指定事件的模糊车内舱室快照图像。ORNL根据原始研究中提供的事件详情表（Event Detail Table, EDT）中的地面真值（ground truth），按照安全带使用情况以及图像内的人员数量筛选了上述事件。由于模糊图像的采集间隔较长，本数据集内的图像未必与EDT中描述的实际20-30秒前置视频片段同步，但均来自同一次行程。基于车内快照的可获取性，图像选择优先级依次为：事件发生期间、事件发生前、事件发生后。 2. 1186个指定事件的行车记录仪（dash camera）视频。ORNL根据EDT中规定的手握方向盘标准、转弯动作以及伸手动作筛选了这些事件。上述视频与EDT中标记的对应20-30秒前置视频片段时间完全同步。 3. 针对2015年5月时EDT中记录的闯红灯违规案例，包含人脸数据、雷达时间序列、扩展时间序列、扩展GPS数据以及前置摄像头视频。ORNL通过筛选驾驶员行为1、2、3，纳入了以下三类信号违规行为：“显然未看到信号灯”“故意无视信号灯”以及“试图抢在信号灯切换前通过”，最终共得到141个事件。碰撞事故相关的人脸数据与GPS数据仅可在VTTI的安全隔离区（secure enclave）中访问。近碰撞事件的时间序列与GPS数据包含了事件结束前2分钟的全部相关数据。 4. 1997个指定事件的雷达与车道跟踪器时间序列数据。ORNL根据驾驶员与外部车辆的行驶行为筛选了上述事件，用于研究检测并刻画基线行驶行为、驾驶员激进驾驶行为、周边车辆激进驾驶行为、分心驾驶，以及基于EDT中事件的相对交通密度。上述时间序列数据与EDT中标记的对应20-30秒前置视频片段完全同步。 5. 198个指定事件的后置摄像头视频。ORNL根据EDT中记录的追尾碰撞、跟车过近以及基线行驶行为筛选了这些事件。上述视频与EDT中标记的20-30秒前置视频片段一一对应。 6. 1998个指定事件的面部摄像头视频。ORNL根据EDT中记录的分心、困倦、四处张望车内、进食、不稳定或非标准行为以及其他与面部相关的数据筛选了上述事件。ORNL在VTTI的安全数据隔离区中获取了这些视频，并对数据进行处理以提取面部特征点（facial landmark points）与非识别性运动数据。 本数据集为doi:10.15787/VTT1/93LFTS的补充数据集。
