kitti数据集介绍论文Vision meets Robotics: The KITTI Dataset

下载链接：http://www.cvlibs.net/publications/Geiger2013IJRR.pdf
摘要：
使用灰度摄像机、彩色摄像机、雷达扫描器、高精度GPS/IMU惯性导航
系统 记录了6h的交通情况，记录频率10-100 Hz。图片经过calibration(校准)、同步并且有时间戳（calibrated, synchronized and timestamped），有原始数据集和经过修正后的数据集（rectified），标注文件包括 3D tracklets形式的标签。
一：介绍
二：传感器设置
2个灰度camera
2个彩色camera
一个OXTS和GPS导航系统
一个雷达扫描器
注意他们的XYZ轴

三：数据集
The raw data set is divided into the categories ’Road’,
’City’, ’Residential’, ’Campus’ and ’Person。
For each sequence, we provide the raw
data, object annotations in form of 3D bounding box tracklets
and a calibration file, as illustrated in Fig. 4.

A.数据描述
date_drive.zip 存放的是整个的视频序列，除了raw data，也提供了 synced data（rectified and synchronized处理后的）。
image、雷达等都有自己的timestamps，雷达的timestamps有三个，雷达起始旋转时间(timestamps_start.txt)，spin end时间（旋转一周采集） (timestamps_end.txt)， facing forward and triggering the cameras (timestamps.txt)。
数据这样存储：
1.image:包括raw data 和 calibrati之后的图片。
2. 每帧包括30个值： The geographic coordinates
including altitude, global orientation, velocities, accelerations,
angular rates, accuracies and satellite information
3.点云数据：Each point is stored with
its (x, y, z) coordinate and an additional reflectance value ®
B.数据标注
3Dbbox的坐标系方向和雷达的坐标系的相对关系看下图，种类 ’Car’, ’Van’, ’Truck’, ’Pedestrian’, ’Person (sitting)’, ’Cyclist’, ’Tram’ and ’Misc’
bbox信息在date_drive_tracklets.xml
For each frame, we provide the object’s translation
and rotation in 3D, as illustrated in Fig. 7.
Note that we only provide the yaw angle, while the other two angles
are assumed to be close to zero

本文也提供了The total number of objects and the object orientations
for the two predominant classes ’Car’ and ’Pedestrian’ are
shown in Fig. 8.
For each object class, the number of object
labels per image and the length of the captured sequences is
shown in Fig. 9.
The egomotion of our platform recorded by
the GPS/IMU system as well as statistics about the sequence
length and the number of objects are shown in Fig. 10 for the
whole dataset and in Fig. 11 per street category.
C.开发组件
reading tracklets ：readTrackletsMex.cpp
run_demoTracklets.m：demonstrates how
3D bounding box tracklets can be read from the XML files and projected onto the image plane of the cameras.
run_demoVelodyne.m： The projection of 3D Velodyne point clouds into the image plane，See Fig. 6 for an illustration.

run_demoVehiclePath.m：shows how to
read and display the 3D vehicle trajectory using the GPS/IMU data.
从当前车辆所在的经纬信息 、海拔信息以及 偏航角、俯仰角等，可以使用墨卡托投影计算出车辆的6D信息（Rx Ry Rz tx ty tz）

D. Benchmarks

四. 传感器标定
we write scalars in lower-case
letters (a), vectors in bold lower-case (a) and matrices using
bold-face capitals (A). 3D rigid-body transformations which
take points from coordinate system a to coordinate system b
will be denoted by Tba
, with T for ’transformation’
a. 同步：雷达扫到前面的时候，相机与其同步数据
b. 相机标定：相机中心是在同一个x-y平面的，标定参数在calib_cam_to_cam.txt
Here, i ∈ {0, 1, 2, 3} is the camera index, where 0 represents
the left grayscale, 1 the right grayscale, 2 the left color and
3 the right color camera. Note that the variable definitions
are compliant with the OpenCV library, which we used for
warping the images. When working with the synchronized
and rectified datasets only the variables with rect-subscript are
relevant. Note that due to the 枕形失真效应 the images have been cropped such that the size of the rectified images is smaller than the original size of 1392 × 512 pixels.


. Note that in order to project a 3D point x in reference camera coordinates to apoint y on the i’th image plane, the rectifying rotation matrix of the reference camera R(0)rect must be considered as well.

A 3D point x in IMU/GPS
coordinates gets projected to a point y in the i’th image as

这篇文章中出现很多的
calibration 标定
synchronized 同步
rectified 修正相机坐标系位置
这是我的理解