渐变FOV相机阵列实现大视场3D成像

197 浏览量更新于2024-08-26 收藏 360KB PDF 举报

"使用一系列逐渐变化的FOV摄像机进行大视野3D成像" 这篇研究论文探讨了在计算机视觉领域中的大视野3D成像技术，特别是如何利用一个由多个具有逐渐变化视场角（Field of View, FOV）的摄像机阵列实现这一目标。传统的图像拼接技术通常只需要小范围的图像重叠区域，而多视图3D重建则需要相机具有接近相同的视场角以确保足够的重叠区域来计算目标场景的3D信息。作者提出了一种创新的相机阵列结构设计，将多个具有逐渐变化视场角的相机结合在一起。这种设计旨在同时满足大视野图像拼接和3D重建的需求。通过这样的相机阵列，首先可以计算出3D空间中各点的深度信息，然后利用这些深度信息生成具有大视场角的全景图像。论文中描述的原型成像系统基于这个新颖的相机阵列结构实现，并进行了实际场景的成像实验。通过这种方法，研究人员能够扩展单一设备的视界，实现对更广阔场景的3D捕获，这对于各种应用，如虚拟现实、遥感、自动驾驶和安全监控等具有重要意义。该研究进一步深化了我们对3D成像技术的理解，尤其是在处理大视场图像时如何克服传统方法的局限性。它强调了视场角的变化对于优化图像拼接和3D重建效果的重要性，同时证明了这种相机阵列设计的有效性和实用性。这项工作为3D成像技术开辟了新的可能性，通过相机阵列的创新设计，实现了对大范围场景的精细3D重构。这不仅可能推动计算机视觉技术的发展，也可能对相关领域的工程应用产生深远影响。

Large-field 3D Imaging Using an Array of Gradually-

changed FOV Cameras

Yan Wang, Lidong Chen

and Liang Bai

Science and Technology on Information Systems Engineering Laboratory

National University of Defense Technology

Changsha, China

nudtdong11@163.com

Abstract—Large-field image stitching and 3D reconstruction

are two hot topics in computer vision. Generally, only a small

amount of overlapped area is required in large-field image

stitching. On the other hand, cameras usually share almost the

same field of view for multi-view 3D reconstruction, which

guarantees enough overlapped areas to calculate 3D information

of the target scene. To realize the two purposes using a single

imaging device simultaneously, we propose a novel structure

design of camera array, which combines multiple cameras with

gradually-changed field of view. Based on the proposed camera

array, we can firstly calculate the depth of points in 3D space, then

a panoramic image with large field of view can be generated in

consideration of the depth information. A prototype imaging

system based on the novel camera-array structure is realized, and

then real-scene imaging experiments are implemented to prove the

feasibility and effectiveness of our method.

Keywords—camera array with gradually-changed FOV; image

stitching; multi-view 3D reconstruction; light-field imaging

I. INTRODUCTION

Since the emergence of cameras, camcorders and other

imaging equipment, higher resolution and larger field of view

have always been the two main objectives in a long time.

Unfortunately, these two goals often seem to contradict each

other. Normally, each camera has a limited field of view and

cannot captures enough imaging information in many realistic

applications [1]. Panoramic imaging is an effective solution,

which can stitch multiple images to generate a panorama with a

much larger field of view [2]. Moreover, with the development

of digital image processing technology, the innovation of

imaging structure design and the emergence of light-field

imaging, the acquisition and display of only two dimensional

information are not satisfied in various applications. Three-

dimensional (3D) information are usually expected. Compared

with conventional imaging in two-dimensional space, the 3D

structure information of the spatial scene is much more useful,

which provides the possibility of 3D object recognition, robot

visual navigation and virtual space reconstruction.

In order to break the limitation of physical resolution and

field of view of a single image sensor, combining multiple

cameras into a camera array with some specific structure is a

common solution [3]. There are two main imaging methods

using camera array according to different application

backgrounds:

1) Multi-camera stitching imaging

Multi-camera stitching imaging uses a number of cameras

oriented in different directions, among which adjacent cameras

shares only a little overlapped vision, and respectively, to obtain

scene information of different direction and get panoramic

image of FOV [4]. Typical devices are: the PARASCAN

panoramic camera launched by Honeywell of the United States

and the high-altitude aerial photography of the military ground

reconnaissance imaging system (ARGUS-IS) put forward by the

United States Defense Advanced Research Projects Agency

(DARPA) [5] and so on. However, due to the deviation of the

camera viewpoints in multi-camera mosaic imaging system, the

optical center of each camera cannot be completely coincided

[6]. When obvious depth changes exists, it cannot satisfy the

assumption that global scene are in a space plane of relatively

fixed depth. No matter how precise the camera calibration and

image registration is, the image splicing cracks is theoretically

inevitable.

2) Light-field imaging based on camera array

A common method of light field imaging is to take multiple

images of the same scene to obtain abundant light field

information of the scene [7] [8] [9]. There are two main

achievable approaches: one is that to design a compact

arrangement of cameras, all of which have almost overlapped

field of view; another is that all the cameras are located in a

large spherical distribution surrounding the target, which can

capture different perspectives of the target from different

directions. B. Wilburn studied the light field imaging based on

camera array and proposed a variety of spatial arrangements of

camera array [10] [11]. Moreover, MIT 64-camera array [12],

3D Room [13] in the Carnegie Mellon Virtual Reality Project,

and multi-view dynamic scene video capture platform

developed by Li and Dai [14] are also some successful

prototypes. However, all the above imaging systems have very

limited 3D field of view, which cannot capture light-field

information of large scale scenes.

A novel structure of camera array is proposed in this paper,

which can realize large-field 3D reconstruction and high-

resolution panoramic imaging simultaneously. Multiple cameras

with gradually-changed field of view are combined in the array,

which are regularly fixed in both horizontal and vertical

directions. Overlapped field of view between the adjacent

cameras in two directions causes that a same spatial point may

2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC)

Banff Center, Banff, Canada, October 5-8, 2017

下载后可阅读完整内容，剩余4页未读，立即下载

weixin_38682518

粉丝: 3
资源: 935

渐变FOV相机阵列实现大视场3D成像

监控摄像机相关资料

摄像机焦距计算器

机器视觉中使用工业镜头的计算方式

远心镜头详解：消除畸变，提升成像质量

工业镜头选择关键要素：视野、景深与接口详解

工业相机镜头：放大倍率详解与参数选择

图像拼接技术在近距离大视场应用分析

接收机视场角 fov 计算

地级市GDP及产业结构数据-最新.zip

2006-2023年上市公司资产误定价Misp数据集（4.9万样本，含原始数据、代码及结果，最新）.zip

最新资源