Fisheye projections from spherical maps

Written by Paul Bourke
May 2003, software updated January 2016

See also: Mapping a equirectangular projection to perspective

The source code implementing the projections below is only available
on request for a small fee. It includes a demo application and an invitation to
convert an image of your choice to verify the code does what you seek.
For more information please contact the author.

"sphere2fish" takes a full spherical map and generates fisheye views
given a center position for the fisheye, optionally an up vector, and an
aperture from 0 to 360 degrees. In addition, the user may choose the
level of antialiasing (supersampling) and an option to create a circularly
bounds fisheye (traditional) or a rectangularly bound fisheye (the later is
usually only appropriate for smaller aperture angles). The application is
in the form of a UNIX style command line interface, a TGA file is expected
for the input image (24 or 32 bit RGB), the output fisheye image is
written to standard output. The usage message is given below.

Usage: sphere2fish [options] sphereimage
Options
-w n width and height of the fisheye image, default = 512
-t n fisheye aperture (degrees), default = 180
maximum is 360 degrees
-c x y center longitude and latitude of fisheye, default = 0 0 degrees
-u x y z up vector, default = (0,0,1)
required for latitude = +- 90 degrees (poles)
-f full rectangular fisheye instead of circular crop
-a n antialiasing level, default = 2
-o s output file name, name derived from input filename
-bg r g b background image colour (outside fisheye circle), default = 0 0 0
-bf back fade to black, reduce interreflections

The conventions used with the program as drawn below. In particular, zero longitude and latitude are taken to be in the center of the spherical image.

The following spherical image will be used to illustrate (and test) various capabilities of the fisheye generator.

Default view conditions
sphere2fish lobby.tga > 1.tga

  From 90 degrees longitude
sphere2fish -c 90 0 lobby.tga > 2.tga

From 45 degrees latitude
sphere2fish -c 0 45 lobby.tga > 3.tga

  Fisheye from north pole
sphere2fish -c 0 90 lobby.tga > 4.tga

Fisheye from north pole and with different up vector
sphere2fish -c 0 90 -u 0 1 0 lobby.tga > 5.tga

  Extend to whole rectangular range
sphere2fish -f lobby.tga > 6.tga

Full 360 degree fisheye
sphere2fish -t 360 lobby.tga > 7.tga

  90 degree fisheye
sphere2fish -t 90 -f lobby.tga > 8.tga

Camera roll
sphere2fish -c -60 0 -u 0 1 1 lobby.tga > 9.tga

Photographic example

The following is a spherical projection stitched from 50 images using AutoPano Pro.
Only a 1/2 hemisphere is captured. The original spherical projection image is
22000x11000 pixels.

In this case only one fisheye can be captured. If the full spherical projection was captured
(only requiring twice the number of individual images) then a fisheye pointing in any
direction can be sampled. The original fisheye images can be up to 10,000 pixels square.

Converting Images from Panoramic to Dome

Written by Paul Bourke
February 2002

Example images courtesy (and copyright) Astro Copy Service,
Planetarium Augsburg, Germany

See also Angular fisheye projections.

The following will discuss the conversion of panoramic images into
those suitable for display onto a dome, in particular, angular
fisheye as used by the majority of planetarium domes. There are
a number of panoramic image formats, they all use the same horizontal
axis which ranges from 0 to 2 pi. There are however a number of
vertical distortions, the two most common are considered here.
One is often called a radial panoramic where the vertical axis
is considered to lie on the surface of a sphere, the other is a
linear panoramic where the vertical axis results from a standard
perspective projection. This second is the most common, the image
is obtained by projecting onto a cylinder about the camera.

As with most mappings the goal is to estimate the colour for each
pixel in the destination image. The destination image in this case
is the angular fisheye, the colour estimate will be determined by
the corresponding
pixels in the source (panoramic) image. In general there isn't a
single pixel in the input image corresponding to a particular pixel in the
destination image. For best results it is usual to find the closest
pixel in the input image for a range of positions within each pixel
in the destination image, these are averaged together to determine
the final estimate of the colour. This is commonly called antialiasing,
the simplest form of which is to estimate the colour by averaging over
a 2x2 or 3x3 grid within each pixel.

The basic idea is to find the mapping between
a coordinate system in the angular fisheye and a coordinate system in the
panoramic image. The convenient coordinate system in the angular fisheye
is "r" (the distance from the center of the pixel to the pixel in
question) and the angle "phi" (angle of the vector to the pixel).
The convenient coordinate system in the panoramic image is the same
angle "phi" (the vertical axis of the panoramic), in both images this
angle varies from 0 to 2 pi. The vertical axis of the panoramic is
proportional to the "r" in the angular fisheye. This is illustrated
in the following figure.

The conversion of the two types of panoramic image will be illustrated
by transforming the following test pattern. This will be done for three
values of thetamax, that is, the panoramic images will be assumed to
vary vertically from 0 to 30, 60, and 90 degrees.

Radial panoramic

The hole in the center for all but the 90 degree case reflects the lack
of image data from thetamax to 90 degrees which is the central
portion of the fisheye image.

Linear panoramic

The most obvious feature of this case is the extreme distortion
that occurs as the fisheye tends to 90 degrees. Of course, while
the distortion seems extreme below that is because the test image
has a equal spacing grid, something that would not normally occur in a
90 degree panoramic. A normal 90 degree panoramic would appear very
distorted itself and the process of turning that into a dome image
would correct for the distortion.

The following is a
linear 45 degree angular fisheye of the panoramic shown at the top
of this page.

Notes

  • It has been assumed here that the panoramic extends from 0 degrees
    upwards along the vertical axis. This is because the discussion here
    has been targeted towards turning panoramas into images for
    projection in planetarium domes where one only sees above the
    horizon in both the panoramic and fisheye.

  • Radius of the dome image is related to the length of the
    panoramic, namely, the diameter of the dome image is the
    width of the panoramic divided by pi. Other dome dimensions
    will result in a radial compression or stretching.

Further Examples

pan2sph panfilename [options]
Options:
-t n set max theta on vertical axis of panoramic, 0...90 (default: 45)
-a n set antialias level, 1 upwards, 2 or 3 typical (default: 2)
-w n width of the dome image, height = width (default: 512)
-s use sine function correction (default: off)
-r n rotation angle, 0...nx (default: 0)
-f flip insideout (default: off)
-bg r g b set background colour, 0...255 each (default: 0 0 0)

Diagram illustrating panorama within a circular fisheye for planetarium use

Further exercise
December 2004

Usage: pano2fish [options] fisheyeimage
Options
-w n fisheye image width and height, default = -1
-r r1 r2 inner and outer radius of the fisheye image
-h h1 h2 top and bottom panoramic edges
-a n antialiasing level, default = 1 (no antialising)
-d change direction of panoramic
-p n rotate by n degrees
-c clear region outside fisheye radius

Example

Input image
Default settings

pano2fish -w 480 pano.tga
  Set panoramic height range, measured from bottom of the image

pano2fish -a 2 -h 172 50 -w 480 pano.tga
Set the fisheye radius range, measured from the center of the fisheye

pano2fish -a 2 -h 172 50 -r 208 60 -w 480 pano.tga
  Reverse panoramic direction

pano2fish -a 2 -h 172 50 -r 208 60 -w 480 -d pano.tga
Change phase

pano2fish -a 2 -h 172 50 -r 208 60 -w 480 -d -p 180 pano.tga
  Invert mapping by swapping radius bounds

pano2fish -a 2 -h 172 50 -r 60 208 -w 480 pano.tga
Clear region outside radius range

pano2fish -a 2 -h 172 50 -r 208 60 -w 480 -c pano.tga

Rendering software solution.
Update: February 2016

There are now a number of plug-ins for compositing packages that allow one
to work in polar coordinates, rather than Cartesian coordinates and also allow
one to place various image projections (perspective, cylindrical, equirectangular)
into the fisheye image space.
But if one is used to a 3D modelling/rendering package or if one is creating
realtime content then an alternative approach is to compose the non-fisheye
elements in the 3D scene and then simple render the fisheye. For plane perspective
images these may be placed on a billboard in the scene, at the desired position
and angle. The image will appear on a flat surface in the dome oriented, positioned
and sized as set in the model.

In the case relevant here, one could place a cylinder in the 3D model world with
the panorama on the surface as a texture. This cylinder should have the same horizontal
angular extent as the panorama, not necessarily a full 360 degrees. The height of the cylinder
would normally be fixed to the right aspect ratio of the panorama but the position and
angle of the panorama could be varied as desired. Indeed, rotating and tilting the cylinder is
a standard type of dome transition or way of enlivening what is otherwise a static image (of
course the texture could also be a video sequence). Again, rendering the fisheye of the
textured cylinder will ensure the geometric effect in the dome is the same as within
the 3D model.

Some examples are shown below, on the left is a view of the cylinder in relation to
a polar grid representing the dome. On the right is the fisheye rendered from the
middle of the dome.
The examples here are created using PovRay but any 3D modelling/rendering package
that supports a circular fisheye lens could be chosen.

Raising the cylinder above the spring line of the dome.

Rotating the cylinder with respect to the dome axis.

网址:

http://paulbourke.net/dome/2fish/

Fisheye projections from spherical maps [转]的更多相关文章

  1. Computer Generated Angular Fisheye Projections [转]

    Computer GeneratedAngular Fisheye Projections Written by Paul Bourke May 2001 There are two main ide ...

  2. Converting a fisheye image into a panoramic, spherical or perspective projection [转]

    Converting a fisheye image into a panoramic, spherical or perspective projection Written by Paul Bou ...

  3. 鱼眼模式(Fisheye projection)的软件实现

    简单实现 鱼眼模式(Fisheye)和普通的透视投影(Perspective projection),一个很大的区别就是鱼眼的投影算法是非线性的(non-linear),实际照相机的情况是在镜头外面包 ...

  4. 鱼眼投影方式(Fisheye projection)的软件实现

    简单实现 鱼眼模式(Fisheye)和普通的透视投影(Perspective projection),一个很大的区别就是鱼眼的投影算法是非线性的(non-linear),实际照相机的情况是在镜头外面包 ...

  5. Google Maps地图投影全解析(3):WKT形式表示

    update20090601:EPSG对该投影的编号设定为EPSG:3857,对应的WKT也发生了变化,下文不再修改,相对来说格式都是那样,可以到http://www.epsg-registry.or ...

  6. 球面墨卡托(Spherical Mercator)

    地理信息描述空间位置相关的信息,在空间位置的表达中,需要基于空间参照系来保证数据精度以及不同数据源之间的相互叠加/空间分析操作.自Google Maps与2005年发布以来,电子地图服务与普通民众的日 ...

  7. Self-organizing Maps及其改进算法Neural gas聚类在异常进程事件识别可行性初探

    catalogue . SOM简介 . SOM模型在应用中的设计细节 . SOM功能分析 . Self-Organizing Maps with TensorFlow . SOM在异常进程事件中自动分 ...

  8. MapServer Tutorial——MapServer7.2.1教程学习——第一节用例实践:Example1.6 Defining Projections and Extents

    MapServer Tutorial——MapServer7.2.1教程学习——第一节用例实践:Example1.6 Defining Projections and Extents 一.前言 当在m ...

  9. How to Tell Science Stories with Maps

    Reported Features How to Tell Science Stories with Maps August 25, 2015   Greg Miller This map, part ...

随机推荐

  1. Redis实战配置(三)

    程序配置 我们安装好了Redis的系统服务,此时Redis服务已经运行. 现在我们需要让我们的程序能正确读取到Redis服务地址等一系列的配置信息,首先,需要在Web.config文件中添加如下信息: ...

  2. Vue.js 系列教程 3:Vue

    原文:intro-to-vue-3-vue-cli-lifecycle-hooks 译者:nzbin 这是 JavaScript 框架 Vue.js 五篇教程的第三部分.在这一部分,我们将学习 Vue ...

  3. Scrapy基础------css选择器基础

    基本语法: * 选择所有节点 #container 选择id为container的节点 .container 选择所有class包含container的节点 li a 选取所有li 下所有a节点 ul ...

  4. codevs 2173 忠诚

    2173 忠诚  时间限制: 1 s  空间限制: 32000 KB  题目等级 : 钻石 Diamond   题目描述 Description 老管家是一个聪明能干的人.他为财主工作了整整10年,财 ...

  5. bzoj1093 [ZJOI2007]最大半联通子图 缩点 + 拓扑序

    最大半联通子图对应缩点后的$DAG$上的最长链 复杂度$O(n + m)$ #include <cstdio> #include <cstring> #include < ...

  6. [BZOJ4888][TJOI2017]异或和(树状数组)

    题目描述 在加里敦中学的小明最近爱上了数学竞赛,很多数学竞赛的题都是与序列的连续和相关的.所以对于一个序列,求出它们所有的连续和来说,小明觉得十分的简单.但今天小明遇到了一个序列和的难题,这个题目不仅 ...

  7. 【20181024T2】小C的序列【GCD性质+链表】

    题面 [错解] 一眼不可做啊 哎分治? 算不了啊 真的是,打暴力走人 20pts (事实上,还有20pts是随机数据,加个小小的特判就可以) [正解] 首先,从l开始往后gcd最多只有O(log)种取 ...

  8. spoj4155 OTOCI LCT

    动态树,支持加边,修改点权,查询链的点权和. #include <cstdio> #include <iostream> #define maxn 30010 using na ...

  9. Codeforces Round #348 (VK Cup 2016 Round 2, Div. 2 Edition) C. Little Artem and Matrix 模拟

    C. Little Artem and Matrix 题目连接: http://www.codeforces.com/contest/669/problem/C Description Little ...

  10. 在阅读sqlmap源码时学到的知识--检查运行环境

    最近在读sqlmap的源码,懵懵懂懂中页大约学到了一些知识(说给自己听的话:由此可见,所谓的能够解决所有遇到问题的python水平,只能说明你遇见的都是简单的需求....),老规矩,在这里写一下,一则 ...