菜鸟看论文——U-V-Disparity与地面检测、相机姿态估计

阳光穿透心脏的1/2处 2021-12-24 08:27 318阅读 0赞

# 菜鸟看论文——U-V-Disparity与地面检测、相机姿态估计 #

本文转载自：[https://www.oipapio.com/cn/article-4924327][https_www.oipapio.com_cn_article-4924327]  
部分PPT地址：[https://download.csdn.net/download/qq\_326324545/11260394][https_download.csdn.net_download_qq_326324545_11260394]  
首先，本部分ppt是由我的师妹王童鞋制作，此处应有掌声~![face0.gif][]

## 一、U-V-Disparity基础知识 ##

![1][]  
![2][]  
![3][]  
![4][]  
![5][]  
![6][]  
![7][]  
![8][]  
![9][]  
![10][]  
![11][]  
注意，此处的（x,y,z）没有实际意义，其实是(u*Zc，v*Zc,Zc）的简写而已，所以求(u,v)时候需要经过下一步才行：  
![12][]  
其中z=Zc。

## 二、地面估计 ##

参考：02.Real time obstacle detection on non flat road geometry through v-disparity representation  
![13][]  
![14][]  
![15][]  
![16][]  
![17][]  
参考：  
15.Ground Segmentation and Occupancy Grid Generation Using Probability Fields  
05.U-V-Disparity: An efficient algorithm for Stereovision Based Scene Analysis  
示例2：  
![18][]  
![19][]  
An assumption in our proposed system is that the stereo rig can only rotate in the pitch and yaw angle, q;f,and as such no angular rotation occurs in the roll direction  
![20][]  
注意和上面的区别在于没有高度h的位移。  
Expanding (1) the image coordinates of a point P are:  
![21][]  
The epipolar constraint reduces the search for corresponding pixels in the two images to a 1-D search problem along the epipolar line. The disparity d of the matched feature is thus calculated as:  
![22][]  
Following the method presented in \[4\] Xp, Yp, Zp can be derived from ul , v and d as follows to get:  
![23][]  
![24][]  
![25][]  
![26][]  
![27][]  
![28][]  
参考：03. A single framework for vehicle roll, pitch, yaw estimation and obstacles detection by stereovision  
![29][]  
![30][]  
![31][]  
![32][]  
![33][]  
![34][]

## 三、 相机姿态与几何 ##

参考：  
07.Global Correlation Based Ground Plane Estimation Using V-Disparity Image  
05.Obstacle Detection with Stereo Vision for Off-Road Vehicle Navigation

### 1.Pitch估计 ###

The behavior of the ground correlation line during a pitch variation is to oscillate, parallel to itself. Experimentally, we found out that the cameras height variation due to oscillation has neglectable effects.  
![35][]  
![36][]  
其中，y，z都是是点（x,y,z）的相机坐标，而不是世界坐标，（u，v）是相机物理坐标，而不是相机像素坐标。高度是相机垂直于地面的高度，角度是相机与地面的夹角，高度和角度都是变化的。w是视差。从而得到v（图像物理坐标，而不是像素坐标）和视差（图像物理坐标，而不是像素坐标）的斜率公式：  
![37][]  
![38][]  
Actually, for the ground correlation lines to be exactly parallel to each other, it should satisfy the conditions shown in Fig. 2. In this figure,the y axis intersects the ground plane through the point G, the distance(a) between camera and G is fixed, and all different ground planes pass through point G. In this situation, the ground plane can be represented as  
![39][]  
![40][]  
![41][]  
At different pitch angles shown in Fig. 3, the slope of ground plane in V-disparity image will be different（此时，y不是恒定的，而h是恒定的）.  
![42][]  
![43][]  
The changing ratio of g  
![44][]  
approaches 0 when is small, and  
![45][]  
![46][]  
![47][]  
For the ground correlation lines to be parallel with each other at different pitch angles, it only need to satisfy Eq. 11.Thus for a given 4w and certain pitch oscillation, we need to reduce s. In \[10\], the tilt angle is 8.5. In \[1\], the tilt angle is 7.73.  
![48][]  
The change of slope is very small, the difference between the greatest slope and smallest slope is merely 0.0652, which is 1.17%. Thus the assumption that the ground correlation lines under different pitch angles are parallel, is valid.  
![49][]  
![50][]  
可看出是三维空间的直线（不易求解）。

### 2.Roll估计 ###

#### （1）在v-disparity图中利用线性相关系数判定是否是直线，进而判定roll是否为0，如果接近于1，那么roll就是0，否则不为0 ####

![51][]  
![52][]  
![53][]

#### （2）利用上述公式，将pitch和roll的估计转换为p，q的估计，归根到底是p的估计 ####

![54][]  
![55][]  
![56][]

### 3.Yaw估计 ###

![57][]  
![58][]  
![59][]  
![60][]  
![61][]  
![62][]  
![63][]

## 参考 ##

02.Real time obstacle detection on non flat road geometry through v-disparity representation  
03.In-vehicle obstacles detection and characterization by stereovision  
03.A single framework for vehicle roll, pitch, yaw estimation and obstacles detection by stereovision  
13.Complex ground plane detection based on V-disparity map in off-road environment

[https_www.oipapio.com_cn_article-4924327]: https://www.oipapio.com/cn/article-4924327
[https_download.csdn.net_download_qq_326324545_11260394]: https://download.csdn.net/download/qq_326324545/11260394
[face0.gif]: /images/20211223/e633fcef3da84f859532e80fd21ecc4f.png
[1]: /images/20211223/31f6b4056d4648a2a722a84f7dcac91b.png
[2]: /images/20211223/2dee3a6dbfd84b418a5d279484c9494e.png
[3]: /images/20211223/573fd223cc704ce78956f57d2a3a744e.png
[4]: /images/20211223/fe6567b10c154bffbd1312a141b4baff.png
[5]: /images/20211223/d2fd46c0534e4d4caa5a15f6359876df.png
[6]: /images/20211223/086eb513c5e94fc39c33761963773b6d.png
[7]: /images/20211223/d75fb1cf94f54acb818b34dedd271ca6.png
[8]: /images/20211223/282fcef5affa4a389d2893ad4b2635db.png
[9]: /images/20211223/b2cfe12015a141aeaf53c56930e55077.png
[10]: /images/20211223/d7a218b047a64209b39423924a4ad6c9.png
[11]: /images/20211223/1dbe14549575491eaaacd583d75ca4b4.png
[12]: /images/20211223/3c92e7f759bd4485bac34bd13c6d57fa.png
[13]: /images/20211223/5d1de625b0a24bf0a23b67d994e2c1cd.png
[14]: /images/20211223/e67dbd90ef954e80a0440c7334353779.png
[15]: /images/20211223/1c87e48ee6f74262bbb07bf96f8aa5ca.png
[16]: /images/20211223/5102abaf4d6145fb91a4e6297f354786.png
[17]: /images/20211223/f1d6117e2fa247318a25e267c670caa7.png
[18]: /images/20211223/523ce69201ec4869b53270764ff896d4.png
[19]: /images/20211223/0ae76e8a5c034c42a5b0d296a39764f6.png
[20]: /images/20211223/7962496bc6d243b888d2998926596ebf.png
[21]: /images/20211223/6fbc45bd66994cc395f724ae179e26f8.png
[22]: /images/20211223/4f09d78f3ecb490b9fab5126b87584cd.png
[23]: /images/20211223/9f00a25bf91949ff8c8e729f4f409772.png
[24]: /images/20211223/afdcf5941c264022acb71cae11d304ea.png
[25]: /images/20211223/1e79e24e2f8249ca84be1650146ed12f.png
[26]: /images/20211223/2c87e75e74ef4148a8c3dda60fb98d7a.png
[27]: /images/20211223/e1402095a29c475faa00804e32689f68.png
[28]: /images/20211223/1aa32d15905248f39244a0adb312d67b.png
[29]: /images/20211223/20b40e7329b245888dd830a1daeeb5b6.png
[30]: /images/20211223/e2b307a881e04f409b82f0f8995adedc.png
[31]: /images/20211223/c95d15f877104f5986462e903f701d0a.png
[32]: /images/20211223/569b05158fb64031bdbaa9e17aa84312.png
[33]: /images/20211223/60eb697e206c4d7dbf1e14a4e82ba1ae.png
[34]: /images/20211223/65a5da13859841fbb68b2441b815e6a6.png
[35]: /images/20211223/e3530c5de450406dbadd95dace6633da.png
[36]: /images/20211223/1deb8e6558b342b284575bb5352df495.png
[37]: /images/20211223/cd901850065e4e7da65d058109303200.png
[38]: /images/20211223/69274e3f512640f9b756e267b12c60b4.png
[39]: /images/20211223/369802a49d7a4a179504b59d42aef350.png
[40]: /images/20211223/d757f8276b134a628df31900fb6ba860.png
[41]: /images/20211223/7e327fb0b9604625a6f913a95f6dfe44.png
[42]: /images/20211223/23fcef25e3c74b6dbda052a1362f12a7.png
[43]: /images/20211223/efc67fc1c4314319a5e4940840ce6daf.png
[44]: /images/20211223/f76569356aa64ac29ad1dca1e6b57c7f.png
[45]: /images/20211223/ab66a417c0734e91a2309167f6211c19.png
[46]: /images/20211223/0f724b3d395f4771a9c01cbdfa114eca.png
[47]: /images/20211223/fd3465946503420aa86bbecbd50ad465.png
[48]: /images/20211223/0ff115cbad57438a9e9d080bfe6109e3.png
[49]: /images/20211223/562de9f468e14c319ab949c003032b9a.png
[50]: /images/20211223/82c3629ac93a42aa941d30c6fffc4ef2.png
[51]: /images/20211223/043b0d6f1c1c4a05a1a84c40f1a43f1b.png
[52]: /images/20211223/d08f9b7a7fdc40ff89d8342b31ec8ea0.png
[53]: /images/20211223/0d4ace6c29b44ce6a4870a1fcebc27b5.png
[54]: /images/20211223/d4818de13a0746f3ae2a5031a52d4a33.png
[55]: /images/20211223/5597730de7894a799c7ddc901ad79910.png
[56]: /images/20211223/00331dd8787b45009a4a9ddac04c8d05.png
[57]: /images/20211223/e02d9597ef73414c89b742fe10b68e0f.png
[58]: /images/20211223/8067ed0b26254459b4269c8042826af0.png
[59]: /images/20211223/6acd819170074926bb4ad9f466a8a29c.png
[60]: /images/20211223/f29afb88697641229de5a1b97d35b542.png
[61]: /images/20211223/0886f4c3c5e24e2d900acd9390a3f3a0.png
[62]: /images/20211223/1ba537891dc84a998ba65d68aa286a3a.png
[63]: /images/20211223/443cea8861a74b2885609cd2b53a5b54.png