HOPS: A Hybrid Dual Camera Vision System

HOPS: A Hybrid Dual Camera Vision System

Stefano Cagnoni (Università degli Studi di Perugia, Italy), Monica Mordonini (Università degli Studi di Perugia, Italy), Luca Mussi (Università degli Studi di Perugia, Italy) and Giovanni Adorni (Università degli Studi di Genova, Italy)
Copyright: © 2009 |Pages: 8
DOI: 10.4018/978-1-59904-849-9.ch124
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Biological vision processes are usually characterized by the following different phases: • Awareness: natural or artificial agents operating in dynamic environments can benefit from a, possibly rough, global description of the surroundings. In human this is referred to as peripheral vision, since it derives from stimuli coming from the edge of the retina. • Attention: once an interesting object/event has been detected, higher resolution is required to set focus on it and plan an appropriate reaction. In human this corresponds to the so-called foveal vision, since it originates from the center of the retina (fovea). • Analysis: extraction of detailed information about objects of interest, their three-dimensional structure and their spatial relationships completes the vision process. Achievement of these goals requires at least two views of the surrounding scene with known geometrical relations. In humans, this function is performed exploiting binocular (stereo) vision. Computer Vision has often tried to emulate natural systems or, at least, to take inspiration from them. In fact, different levels of resolution are useful also in machine vision. In the last decade a number of studies dealing with multiple cameras at different resolutions have appeared in literature. Furthermore, the ever-growing computer performances and the ever-decreasing cost of video equipment make it possible to develop systems which rely mostly, or even exclusively, on vision for navigating and reacting to environmental changes in real time. Moreover, using vision as the unique sensory input makes artificial perception closer to human perception, unlike systems relying on other kinds of sensors and allows for the development of more direct biologically-inspired approaches to interaction with the external environment (Trullier 1997). This article presents HOPS (Hybrid Omnidirectional Pin-hole Sensor), a class of dual camera vision sensors that try to exalt the connection between machine vision and biological vision.
Chapter Preview
Top

Introduction

Biological vision processes are usually characterized by the following different phases:

  • Awareness: natural or artificial agents operating in dynamic environments can benefit from a, possibly rough, global description of the surroundings. In human this is referred to as peripheral vision, since it derives from stimuli coming from the edge of the retina.

  • Attention: once an interesting object/event has been detected, higher resolution is required to set focus on it and plan an appropriate reaction. In human this corresponds to the so-called foveal vision, since it originates from the center of the retina (fovea).

  • Analysis: extraction of detailed information about objects of interest, their three-dimensional structure and their spatial relationships completes the vision process. Achievement of these goals requires at least two views of the surrounding scene with known geometrical relations. In humans, this function is performed exploiting binocular (stereo) vision.

Computer Vision has often tried to emulate natural systems or, at least, to take inspiration from them. In fact, different levels of resolution are useful also in machine vision. In the last decade a number of studies dealing with multiple cameras at different resolutions have appeared in literature. Furthermore, the ever-growing computer performances and the ever-decreasing cost of video equipment make it possible to develop systems which rely mostly, or even exclusively, on vision for navigating and reacting to environmental changes in real time. Moreover, using vision as the unique sensory input makes artificial perception closer to human perception, unlike systems relying on other kinds of sensors and allows for the development of more direct biologically-inspired approaches to interaction with the external environment (Trullier 1997).

This article presents HOPS (Hybrid Omnidirectional Pin-hole Sensor), a class of dual camera vision sensors that try to exalt the connection between machine vision and biological vision.

Top

Background

In the last decade some investigations on hybrid dual camera systems have been performed (Nayar 1997; Cui 1998; Adorni 2001; Adorni 2002; Adorni 2003; Scotti 2005; Yao 2006). The joint use of a moving standard camera and of a catadioptric sensor provides these sensors with their different and complementary features: while the traditional camera can be used to acquire detailed information about a limited region of interest (“foveal vision”), the omnidirectional sensor provides wide-range, but less detailed, information about the surroundings (“peripheral vision”). Possible employments for this class of vision systems are video surveillance applications as well as mobile robot navigation tasks. Moreover, their particular configuration makes it possible to realize different strategies to control the orientation of the standard camera; for example, scattered focus on different objects permits to perform recognition/classification tasks while continuous movements allow to track any interesting moving object. Three-dimensional reconstruction based on stereo vision is also possible.

Key Terms in this Chapter

Visual Servoing: An approach to robot control based on visual perception: a vision system extracts information from the surrounding environment to localize the robot and consequently servoing its position.

Single Viewpoint Constraint: When all incoming principal light rays of a lens intersect at a single point, an image with a non-distorted metric content is obtained. In this case all information contained in this image is seen from this view-point

Lens Distortion: Optical errors in camera lenses, usually due to mechanical misalignment of its parts, can cause straight lines in the observed scene to appear curved in the captured image. The deviation between the theoretical image and the actual one is mostly to be attributed to lens distortion

Camera Calibration: A procedure used to obtain geometrical information about image formation in a specific camera essential to relate metric distances on the image to distances in the real word. Anyway, some a priori information is needed to reconstruct the third dimension from only one image

Inverse Perspective Mapping (IPM): A procedure which allows for perspective effect to be removed from an image by homogeneously redistributing the information content of the image plane into a new two-dimensional domain.

Pin-Hole Camera: A camera that uses a tiny hole (the pin-hole) to convey all rays from the observed scene to the image plane. The smaller the pin-hole, the sharper the picture. Pin-hole cameras achieve a potentially infinite depth of field. Because of its geometric simplicity, the “pin-hole model” is used to describe most traditional cameras

Holonomous Robot: A robot with an unconstrained freedom of movement with no preferential direction. This means that, from a standing position, it can move as easily in any direction

Complete Chapter List

Search this Book:
Reset