Seminar Report

Seminar Report ’11 Sensors on 3D Digitization

Dept. of ECE Marian Engineering college 1

1. INTRODUCTION

Digital 3D imaging can benefit from advances in VLSI technology in

order to accelerate its deployment in many fields like visual communication and

industrial automation. High-resolution 3D images can be acquired using laser-based

vision systems. With this approach, the 3D information becomes relatively

insensitive to background illumination and surface texture. Complete images of

visible surfaces that are rather featureless to the human eye or a video camera can be

generated. Intelligent digitizers will be capable of measuring accurately and

simultaneously colour and 3D.



3. COLOUR 3D IMAGING TECHNOLOGY

Machine vision involves the analysis of the properties of the

luminous flux reflected or radiated by objects. To recover the geometrical structures

of these objects, either to recognize or to measure their dimension, two basic vision

strategies are available.

Passive vision, attempts to analyze the structure of the scene under

ambient light. Stereoscopic vision is a passive optical technique. The basic idea is

that two or more digital images are taken from known locations. The images are then

processed to find the correlations between them. As soon as matching points are

identified, the geometry can be computed.

Active vision attempts to reduce the ambiguity of scene analysis by

structuring the way in which images are formed. Sensors that capitalize on active

vision can resolve most of the ambiguities found with two-dimensional imaging

systems. Lidar based or triangulation based laser range cameras are examples of

active vision technique. One digital 3D imaging system based on optical triangulation

were developed and demonstrated.



4. AUTOSYNCHRONIZED SCANNER

Figure 4.1

The auto-synchronized scanner, depicted schematically on Figure 1,

can provide registered range and colour data of visible surfaces. A 3D surface map is

captured by scanning a laser spot onto a scene, collecting the reflected laser light, and

finally focusing the beam onto a linear laser spot sensor. Geometric and photometric

corrections of the raw data give two images in perfect registration: one with x, y, z co-

ordinates and a second with reflectance data. The laser beam composed of multiple

visible wavelengths is used for the purpose of measuring the colour map of a scene

(reflectance map).

Advantage Triangulation is the most precise method of 3D

Limitation Increasing the accuracy increases the triangulation distance.

The larger the triangulation distance, the more shadows appear

on the scanning object and the scanning head must be made

larger.



5. SENSORS FOR 3D IMAGING

The sensors used in the autosynchronized scanner include

5.1 SYNCHRONIZATION CIRCUIT BASED UPON DUAL

PHOTOCELLS

This sensor ensures the stability and the repeatability of range measurements in

environment with varying temperature. Discrete implementations of the so-called

synchronization circuits have posed many problems in the past. A monolithic version

of an improved circuit has been built to alleviate those problems.

5.2 LASER SPOT POSITION MEASUREMENT SENSORS

High-resolution 3D images can be acquired using laser-based vision systems.

With this approach, the 3D information becomes relatively insensitive to background

illumination and surface texture. Complete images of visible surfaces that are rather

featureless to the human eye or a video camera can be generated.



6. POSITION SENSITIVE DETECTORS

Many devices have been built or considered in the past for measuring the

position of a laser spot more efficiently. One method of position detection uses a

video camera to capture an image of an object electronically. Image processing

techniques are then used to determine the location of the object .For situations

requiring the location of a light source on a plane, a position sensitive detector (PSD)

offers the potential for better resolution at a lower system cost.

The PSD is a precision semiconductor optical sensor which produces

output currents related to the “centre of mass” of light incident on the surface of the

device.

While several design variations exist for PSDs, the basic device can be

described as a large area silicon p-i-n photodetector. The detectors can be available in

single axis and dual axis models.



7. DUAL AXIS PSD

This particular PSD is a five terminal device bounded by four collection

surfaces; one terminal is connected to each collection surface and one provides a

common return. Photocurrent is generated by light which falls on the active area of

the PSD will be collected by these four perimeter electrodes.

Figure 7.1 A typical dual axis PSD The amount of current flowing between each perimeter terminal and the common

return is related to the proximity of the centre of mass of the incident light spot to each

collection surface. The difference between the “up” current and the “down” current is

proportional to the Y-axis position of the light spot .Similarly ,the “ right “current minus the

“left” current gives the X-axis position. The designations “up”,”down”, “right”and ‘left”are

arbitrary; the device may be operated in any relative spatial orientation.



8. LASER SENSORS FOR 3D IMAGING

The state of the art in laser spot position sensing methods can be

divided into two broad classes according to the way the spot position is sensed.

Among those, one finds continuous response position sensitive detectors (CRPSD)

and discrete response position sensitive detectors (DRPSD)

CONTINOUS RESPONSE POSITION SENSITIVE DETECTORS

(CRPSD) The category CRPSD includes lateral effect photodiode.

Figure 8.1 Basic structure of a single-axis lateral effect photodiode

Figure illustrates the basic structure of a p-n type single axis lateral effect

photodiode. Carriers are produced by light impinging on the device are separated in

the depletion region and distributed to the two sensing electrodes according to the

Ohm’s law. Assuming equal impedances, The electrode that is the farthest from the

centroid of the light distribution collects the least current. The normalized position of

the centroid when the light intensity fluctuates is given by P =I2-I1/I1+I2.



The actual position on the detector is found by multiplying P by d/2 ,where

d is the distance between the two sensing electrodes.[3]. A CRPSD provides the

centroid of the light distribution with a very fast response time (in the order of 10

MHz). Theory predicts that a CRPSD provides very precise measurement of the

centroid versus a DRPSD. By precision, we mean measurement uncertainty. It

depends among other things on the signal to noise ratio and the quantization noise. In

practice, precision is important but accuracy is even more important. A CRPSD is in

fact a good estimator of the central location of a light distribution.

DISCRETE RESPONSE POSITION SENSITIVE DETECTORS

(DRPSD)

DRPSD on the other hand comprise detectors such as Charge Coupled

Devices (CCD) and arrays of photodiodes equipped with a multiplexer for sequential

reading. They are slower because all the photo-detectors have to be read sequentially

prior to the measurement of the location of the peak of the light distribution. [1]

DRPSDs are very accurate because of the knowledge of the distribution but slow.

Obviously, not all photo-sensors contribute to the computation of the peak. In fact,

what is required for the measurement of the light distribution peak is only a small

portion of the total array. Once the pertinent light distribution (after windowing

around an estimate around the peak) is available, one can compute the location of the

desired peak very accurately.



Figure 8.2 In a monochrome system, the incoming beam is split into two components

Figure 8.3 Artistic view of a smart sensor with colour capabilities,

Figure shows schematically the new smart position sensor for light spot

measurement in the context of 3D and colour measurement. In a monochrome range

camera, a portion of the reflected radiation upon entering the system is split into two

beams (Figure 7.2). One portion is directed to a CRPSD that determines the location

of the best window and sends that information to the DRPSD.

In order to measure colour information, a different optical element is used

to split the returned beam into four components, e.g., a diffractive optical element

(Figure 7.3). The white zero order component is directed to the DRPSD, while the

RGB 1storder components are directed onto three CRPSD, which are used for colour

detection. The CRPSDs are also used to find the centroid of the light distribution

impinging on them and to estimate the total light intensity The centroid is computed



on chip with the well-known current ratio method i.e. (I1-I2)/ (I1+I2) where I1 and I2

are the currents generated by that type of sensor. The weighed centroid value is fed to

a control unit that will select a sub-set (window) of contiguous photo-detectors on the

DRPSD. That sub-set is located around the estimate of the centroid supplied by the

CRPSD.Then the best algorithms for peak extraction can be applied to the portion of

interest.



9. PROPOSED SENSOR –COLORANGE

9.1 ARCHITECTURE

Figure5 shows the proposed architecture for the color range chip.

Figure9.1 The proposed architecture for the colour range chip

An object is illuminated by a collimated RGB laser spot and a portion of

the reflected radiation upon entering the system is split into four components by a

diffracting optical element as shown in figure 4b. The white zero order component is

directed to the DRPSD, while the RGB 1storder components are directed onto three

CRPSD, which are used for colour detection. The CRPSDs are also used to find the

centroid of the light distribution impinging on them and to estimate the total light

intensity The centroid is computed on chip with the well-known current ratio method

i.e. (I1-I2)/ (I1+I2) where I1 and I2 are the currents generated by that type of sensor.



The weighed centroid value is fed to a control unit that will select a sub-set (window)

of contiguous photo-detectors on the DRPSD. That sub-set is located around the

estimate of the centroid supplied by the CRPSD. Then, the best algorithms for peak

extraction can be applied to the portion of interest.

9.2 ‘32’ PIXEL PROTOTYPE CHIP

Figure6 shows the architecture and preliminary experimental results of a

first prototype chip of a DRPSD with selectable readout window. This is the first

block of a more complex chip that will include all the components illustrated in

Figure 3c.

Figure 9.2 Block diagram of the 32-pixel prototype array.



The prototype chip consists of an array of 32 pixels with related readout channels and

has been fabricated using a 0. mm commercial CMOS process. The novelties implemented

consist in a variable gain of the readout channels and a selectable readout window of 16

contiguous pixels. Both features are necessary to comply with the requirements of 3D single

laser spot sensors, i.e., a linear dynamic range of at least 12 bits and a high 3D data

throughput. In the prototype, many of the signals, which, in the final system are supposed to

be generated by the CRPSDs, are now generated by means of external circuitry. The large

dimensions of the pixel are required, on one side to cope with speckle noise and, on the other

side, to facilitate system alignment. Each pixel is provided with its own readout channel for

parallel reading. The channel contains a charge amplifier and a correlated double sampling

circuit (CDS). To span 12 bits of dynamic range, the integrating capacitor can assume five

different values. In the prototype chip, the proper integrating capacitor value is externally

selected by means of external switches C0-C4. In the final sensor, however, the proper value

will be automatically set by an on chip circuitry on the basis of the total light intensity as

calculated by the CRPSDs.

During normal operation, all 32 pixels are first reset at their bias value and

then left to integrate the light for a period of 1 ms. Within this time the CRPSDs and an

external processing unit estimate both the spot position and its total intensity and those

parameters are fed to the window selection logic. After that, 16 contiguous pixels, as

addressed by the window selection logic, are read out in 5 ms, for a total frame rate of

6 ms. Future sensors will operate at full speed, i.e. an order of magnitude faster. The

window selection logic, LOGIC_A, receives the address of the central pixel of the 16 pixels

and calculates the address of the starting and ending pixel. The analog value at the output of

the each CA within the addressed window is sequentially put on the bit line by a decoding

logic, DECODER, and read by the video amplifier .LOGIC-A generates also synchronization

and end-of-frame signals which are used from the external processing units. LOGIC_B

instead is devoted to the generation of logic signals that derive both the CA and the CDS

blocks. To add flexibility also the integration time can be changed by means of the external

switches T0-T4.The chip has been tested and its functionality proven to be in agreement with

specifications.



11. ADVANTAGES

Reduced size and cost

Better resolution at a lower system cost

High reliability that is required for high accuracy 3D vision systems

Complete images of visible surfaces that are rather featureless to the

human eye or a video camera can be generated.

.

12. DISADVANTAGE

The elimination of all stray light in an optical system requires

sophisticated techniques.



13. APPLICATIONS

Intelligent digitizers will be capable of measuring accurately and

simultaneously colour and 3D

For the development of hand –held 3D cameras

Multi resolution random access laser scanners for fast search and tracking

of 3D features



14. CONCLUSION

The results obtained so far have shown that optical sensors have reached a

high level of development and reliability those are suited for high accuracy 3D vision

systems. The availability of standard fabrication technologies and the acquired know-

how in the design techniques, allow the implementation of optical sensors that are

application specific: Opto-ASICs. The trend shows that the use of the low cost

CMOS technology leads competitive optical sensors. Furthermore post-processing

modules, as for example anti reflecting coating film deposition and RGB filter

deposition to enhance sensitivity and for colour sensing, are at the final certification

stage and will soon be available in standard fabrication technologies. The work on the

Colorange is being finalized and work has started on a new improved architecture.



REFERENCES

[1] L.GONZO, A.SIMONI, A.GOTTARDI, DAVID STAPPA, J.A

BERNALDIN,”Sensors optimized for 3D digitization”, IEEE transactions on

instrumentation and measurement, vol 52, no.3, June 2003, pp.903-908.

[2] P.SCHAFER, R.D.WILLIAMS. G.K DAVIS, ROBERT A. ROSS,”

Accuracy of position detection using a position sensitive detector, “IEEE

transactions on instrumentation and measurement, vol 47, no.4, August 1998,

pp.914-918

[3] J.A.BERALDIN,”Design of Bessel type pre-amplifiers for lateral effect

photodiodes”, International Journal Electronics, vol 67, pp 591-615, 1989.

[4] A.M.DHAKE,”Television and video engineering”, Tata Mc Graw Hill

[5] X.ARREGUIT, F.A.VAN SCHAIK, F.V.BAUDUIN, M.BIDIVILLE,

E.RAEBER,”A CMOS motion detector system for pointing devices”, IEEE

journal solid state circuits, vol 31, pp.1916-1921, Dec 1996

[6] P.AUBERT, H.J.OGUEY, R.VUILLEUNEIR, “Monolithic optical position

encoder with on-chipphotodiodes,”IEEE J.solid state circuits, vol 23, pp.465-

473, April 1988

[7] K.THYAGARAJAN, A.K.GHATAK,”Lasers-Theory and applications”,

Plenum Press

[8] N.H.E.Weste, Kamran Eshraghian,”Principles of CMOS VLSIdesign, A

systems perspective”, Low Price Edition



CONTENTS

INTRODUCTION

COLOUR 3-D IMAGING TECHNOLOGY

LASER SENSORS FOR 3-D IMAGING

POSITION SENSITIVE DETECTORS

PROPOSED SENSOR-COLORANGE

ADVANTAGES

DISADVANTAGES

APPLICATIONS

FUTURE SCOPE

CONCLUSION

REFERENCES



ABSTRACT

Machine vision involves the analysis of the properties of the luminous flux

reflected or radiated by objects. To recover the geometrical structures of these

objects, either to recognize or to measure their dimension, two basic vision

strategies are available.

The first strategy, known as passive vision, attempts to analyze the structure

of the scene under ambient light. In contrast, the second, known as active vision,

attempts to reduce the way in which images are formed. Sensors that capitalize on

active vision can resolve most of the ambiguities found with 2-D imaging systems.

Moreover, with laser –based approaches, the 3-D information becomes relatively

insensitive to background illumination and surface texture. Complete images of

visible surfaces that are rather featureless to the human eye or a video camera can be

generated. Thus the task of processing 3-D data is greatly simplified.



ACKNOWLEDGEMENT

I extend my sincere gratitude towards Prof. Sasikumar Head of Department

for giving us his invaluable knowledge and wonderful technical guidance

I express my thanks to our staff advisor Mrs. Remola Joy for their kind co-

operation and guidance for preparing and presenting this seminar.

I also thank all the other faculty members of ECE department and my

friends for their help and support.



Many devices have been built or considered in the past for measuring the

position of a light beam. Among those, one finds continuous response position sensitive

detectors (CRPSD) and discrete response position sensitive detectors (DRPSD) [9-20].

The category CRPSD includes lateral effect photodiode and geometrically shaped

photo-diodes (wedges or segmented). DRPSD on the other hand comprise detectors

such as Charge Coupled Devices (CCD) and arrays (1-D or 2-D) of photodiodes

equipped with a multiplexer for sequential reading. One cannot achieve high

measurement speed like found with continuous response position sensitive detectors

and keep the same accuracy as with discrete response position sensitive detectors. On

Figure 1, we report two basic types of devices that have been proposed in the literature.

A CRPSD provides the centroid of the light distribution with a very fast response time

(in the order of 10Mhz) [20]. On the other hand, DRPSD are slower because all the

photo-detectors have to be read sequentially prior to the measurement of the location of

the real peak of the light distribution [21]. People try to cope with the detector that they

have chosen. In other words, they pick a detector for a particular application and accept

the tradeoffs inherent in their choice of a position sensor. Consider the situation

depicted on Figure 2, a CRPSD would provide A as an answer. But a DRPSD can

provide B, the desired response. This situation occurs frequently in real applications.

The elimination of all stray light in an optical system requires sophisticated techniques

that increase the cost of a system. Also, in some applications, background illumination

cannot be completely eliminated even with optical light filters. We propose to use the

best of both worlds. Theory predicts that a CRPSD provides very precise measurement

of the centroid versus a DRPSD (because of higher spatial resolution). By precision, we

mean measurement uncertainty. It depends among other things on the signal to noise

ratio, the quantization noise, and the sampling noise. In practice, precision is important

but accuracy is even more important. A CRPSD is in fact a good estimator of the

central location of a light distribution. Accuracy is understood to be the true value of a

quantity. DRPSD are very accurate (because of the knowledge of the distribution) but

can be less precise (due to spatial sampling of

the light distribution).



Figure 9.3 A typical situations where stray light blurs the measurement of the real but

much narrower peak. A CRPSD would provide A as an answer. But a DRPSD can provide B, the desired response peak

Their slow speed is due to the fact that all the pixels of the array have to be read

but they don’t all contribute to the computation of the peak. In fact, what is required for

the measurement of the light distribution peak is only a small portion of the discrete

detector. Hence the new smart detector. Once the pertinent light distribution is

available, one can compute the location of the desired peak very accurately. Figure 3

shows an example of the embodiment of the new smart position sensor for single light

spot measurement in the context of 3D and colour measurement. An object is

illuminated by a collimated RGB laser spot and a portion of the reflected radiation upon

entering the system is split into four components by a diffracting optical element. The

white zero order component is directed onto the DRPSD, while the RGB 1st order

components are directed onto three CRPSD which are used for colour detection. The

CRPSDs are also used to find the centroid of the light distribution impinging on them

and to estimate the total light intensity. The centroid is computed on chip or offchip

with the well-known current ratio method i.e. (I1-I2)/(I1+I2) where I1 and I2 are the

currents generated by that type of sensor [20]. The centroid value is fed to a control unit

that will select a sub-set of contiguous photo-detectors on the DRPSD. That sub-set is

located around the estimate of the centroid supplied by the CRPSD. Then, the best

algorithms for peak extraction can be applied to the light distribution of interest



10. IMPLEMENTATION AND EXPERIMENTAL RESULTS

We present here the architecture and preliminary experimental results of a first

prototype chip of a DRPSD with selectable readout window. This is the first block of a

more complex chip that will include all the components illustrated in Figure 3. The

prototype chip consists of an array of 32 pixels with related readout channels and has

been fabricated using a 0.8mm commercial CMOS process [22]. The sensor

architecture is typical for linear array optical sensors [23] and is shown in Figure 4. The

novelties implemented consist in a variable gain of the readout channels and a

selectable readout window of 16 contiguous pixels. Both features are necessary to

comply with the requirements of 3D single spot vision systems a linear dynamic range

of at least 12 bits and a fast readout. Of course, in the prototype, many of the signals

which, in the final system are supposed to be generated by the CRPSDs, are now

generated by means of external circuitry. As stated above the photosensor array

contains 32 pixels with a pitch of 50mm, each pixel having a sensitive area of 48 x 500

mm2. The large dimensions of the pixel are, required, on one side to cope with speckle

noise [8] and, on the other side, to facilitate system alignment. Each pixel is provided

with its own readout channel for parallel reading. The channel contains a charge

amplifier (CA) and a correlated double sampling circuit (CDS). To span 12 bits of

dynamic range, the integrating capacitor of the CA can assume five different values

(CAP). In the prototype chip the proper integrating capacitor value is externally

selected by means of the switches C0-C4. In the final sensor, however, the proper

capacitance value will be automatically set by an on chip circuitry, on the basis of the

total light intensity as calculated by the CRPSDs. During normal operation all 32 pixels

are first reset at their bias value and then left to integrate the light for a period of 12ms.

Within this time the CRPSDs and an external processing unit are supposed to estimate

both the spot position and its total intensity and to feed those parameters to the window

selection logic, LOGIC_A, and to the gain selection logic, LOGIC_B. After that, 16

contiguous pixels, as addressed by the window selection logic, are read out in 52ms, for

a total frame rate of 64ms.



Figure 10.1 Artistic view of a smart sensor for accurate, precise and rapid light position measurement.

The window selection logic, LOGIC_A, receives the address of the central pixel

of the 16 pixel and calculates the address of the starting and ending pixel. The analog

value at the output of the each CA within the addressed window is sequentially put on

the bit line by a decoding logic, DECODER, and read by the video amplifier.

LOGIC_A generates also synchronization and end-of-frame signals which are used

from the external processing units. LOGIC_B, instead, is devoted to the generation of

logic signals that drive both the CA and the CDS blocks. To add flexibility also the

integration time can be changed by means of the external switches T0-T4. The chip has

been tested and its functionality has been proven to being agreement with

specifications. In Figures 5 and 6 are illustrated the experimental results relative to

spectral responsivity and power responsivity, respectively.



Figure10.2 Spectral responsivity of the photo elements

The spectral responsivity, obtained by measuring the response of the

photoelements to a normal impinging beam with varying wavelength is in agreement

with data reported in literature [ 23]. Its maximum value of ~0.25 A/W is found around

l~660nm. The several peaks in the curve are due to multiple reflection of light passing

through the oxide layers on top of the photosensitive area. The power responsivity has

been measured by illuminating the whole array with a white light source and measuring

the pixel response as the light power is increased. As expected the curve can be divided

into three main regions: a far left region dominated by photoelement noise, a central

region where the photoelement response is linear with the impinging power and a

saturation region. The values of the slope, linearity and dynamic range of the central

region have been calculated for three chips



Figure 10.3 Normalized power responsivity measured on two different samples.



Figure Sensor response to a moving spot laser



2. Overview of 3D imaging techniques

3D imaging sensors generally operate by projecting (in the active form) or

acquiring (in the passive form) electromagnetic energy onto/from an object followed by

recording the transmitted or reflected energy. The most important example of

transmission gauging is the industrial computer tomography (CT), which uses high-

energy x-rays and measures the radiation transmitted through the object. Reflection

sensors for shape acquisition can be subdivided into non-optical and optical sensing.

Non-optical sensing includes acoustic sensors (ultrasonic, seismic), electromagnetic

(infrared, ultraviolet, microwave radar, etc...) and others. These techniques typically

measure distances to objects by measuring the time required for a pulse of sound or

microwave energy to bounce back from an object. In reflection optical sensing, light

carries the measurement information. There is a remarkable variety of 3D optical

techniques, and their classification is not unique. In this section, we mention the more

prominent approaches, and we classify them as shown in Table 1

Table 2.1 Classification of 3D imaging techniques



The 3D techniques are based on optical triangulation, on time delay, and on the

use of monocular images. They are classified into passive and active methods. In

passive methods, the reflectance of the object and the illumination of the scene are used

to derive the shape information: no active device is necessary. In the active form,

suitable light sources are used as the internal vector of information. A distinction is also

made between direct and indirect measurements. Direct techniques result in range data,

i.e., into a set of distances between the unknown surface and the range sensor. Indirect

measurements are inferred from monocular images and from prior knowledge of the

target properties. They result either in range data or in surface orientation. Excellent

reviews of optical methods and range sensors for 3D measurement are presented in

references [1–7]. In [1] a review of 20 years of development in the field of 3D laser

imaging with emphasis on commercial systems available in 2004 is given.

Comprehensive summaries of earlier techniques and systems are provided in the other

publications. A survey of 3D imaging techniques is also given in [8] and in [9], in the

wider context of both contact and contactless 3D measurement techniques for the

reverse engineering of shapes.



4.1 Laser triangulators

Both single-point triangulators and laser stripes belong to this category. They

are all based on the active triangulation principle [17]. Figure 1 shows a typical system

configuration. Points OP and OC are the exit and the entrance pupils of a laser source

and of a camera. Their mutual distance is the baseline d. The optical axes zP and zC of

the laser and the camera form an angle of degrees

Figure 4.2 Schematics of the triangulation principle.

The laser source generates a narrow beam, impinging the object at point S (single-point

triangulators). The back scattered beam is imaged at point S’ at image plane . The

measurement of the location (iS, jS) of image point S’ defines the line of sight C S'O ,

and, by means of simple geometry, yields the position of S. The measurement of the

surface is achieved by scanning. In a conventional triangulation configuration, a

compromise is necessary between the Field of View (FOV), the measurement resolution

and uncertainty, and the shadow effects due to large values of angle . To overcome

this limitation, a method called ‘synchronized scanning’ has been proposed. Using the

approach illustrated in [18, 19], a large field of view can be achieved with a small angle

, without sacrificing range measurement precision. Laser stripes exploit the optical



triangulation principle shown in Figure 1. However, in this case, the laser is equipped

with a cylindrical lens, which expands the light beam along one direction. Hence, a

plane of light is generated, and multiple points of the object are illuminated at the same

time. In the figure, the light plane is denoted by S, and the illuminated points belong

to the intersection between the plane and the unknown object (line AB). The

measurement of the location of all the image points from A’ to B’ at plane allows the

determination of the 3D shape of the object in correspondence with the illuminated

points. For dense reconstruction, the plane of light must scan the scene [20–22]. An

interesting enhancement of the basic principle exploits the Biris principle [23]. A laser

source produces two stripes by passing through the Bi-Iris component (a mask with two

apertures). The measurement of the point location at the image plane can be carried out

on a differential basis, and shows decreased dependence on the object reflectivity and

on the laser speckles. One of the most significant advantages of laser triangulators is

their accuracy, and their relative insensitivity to illumination conditions and surface

texture effects. Single-point laser triangulators are widely used in the industrial field,

for the measurement of distances, diameters, thicknesses, as well as in surface quality

control applications. Laser stripes are becoming popular for quality control, reverse

engineering, and modeling of heritage. Examples are the Vivid 910 system (Konica

Minolta, Inc.), the sensors of the SmartRay series (SmartRay GmbH, Germany) and the

ShapeGrabber systems (ShapeGrabber Inc., CA, USA). 3D sensors are designed for

integration with robot arms, to implement “pick and place” operations in de-palletizing

stations. An example is the Ranger system (Sick, Inc.). The NextEngine system

(NextEngine, Inc., CA, USA) and the TriAngles scanner (ComTronics, Inc., USA)

device deserve particular attention, since they show how the field is progressing. These

two systems, in fact, break the price barrier of laser-based systems, which is a limiting

factor in their diffusion.

4.2 Structured light

Structured light based sensors share the active triangulation approach above

mentioned. However, instead of scanning the surface, they project bi-dimensional

patterns of non-coherent light, and elaborate them to obtain the range information for

each viewed point simultaneously. A single pattern as well as multiple patterns can be



projected. In both cases, the single light plane S shown in Figure is replaced by a

bundle of geometrical planes, schematically shown in Figure 2. The planes are indexed

along the LP coordinate at the projector plane. The depth information at object point S

is obtained as the intersection between line sight SS’ and the plane indexed by LP=LPS

[25]. The critical point is to guarantee that different object points are assigned to

different indexes along the LP coordinate. To this aim, a large number of projection

strategies have been developed. The projection of grid patterns [26, 27], of dot patterns

[28], of multiple vertical slits [29] and of multi-color projection patterns [30] have been

extensively studied. Particular research has been carried out on the projection of fringe

patterns [31, 32], that are particularly suitable to maximize the measurement resolution.

As an example, in Figure 3, three fringe patterns are shown. The first one is formed by

fringes of sinusoidal profile, the second one is obtained by using the superposition of

two patterns with sinusoidal fringes at different frequencies [33], and the last one is

formed by fringes of rectangular profile [34].

Figure 4.3 Principle of measurement in structured light systems



4.3 Photogrammetry

Photogrammetry obtains reliable 3D models by means of photographs. In digital

photogrammetry digital images are used. The elaboration pipeline consists basically of

the following steps: camera calibration and orientation, image point measurements, 3D

point cloud generation, surface generation and texture mapping. Camera calibration is

crucial in view of obtaining accurate models. Image measurement can be performed by

means of automatic or semi-automatic procedures. In the first case, very dense point

clouds are obtained, even if at the expense of inaccuracy and missing parts. In the

second case, very accurate measurements are obtained, over a significantly smaller

number of points, and at increased elaboration and operator times [40–42]. Typical

applications of photogrammetry, besides the aerial photogrammetry, are in close range

photogrammetry, for city modeling, medical imaging and heritage [48–43].

Reliable packages are now commercially available for complete scene modeling or

sensor calibration and orientation. Examples are the Australis software (Photometrics,

Australia), Photomodeler (Eos Systems, Inc., Canada), Leica Photogrammetry Suite

(Leica Geosystems GIS & Mapping), and Menci (Menci Software srl, Italy). Both

stereo vision and photogrammetry share the objective of obtaining 3D models from

images and are currently used in computer vision applications. The methods rely on the

use of camera calibration techniques. The elaboration chain is basically the same, even

if linear models for camera calibration and simplified point measurement procedures

are used in computer vision applications. In fact, in computer vision the automation

level of the overall process is more important with respect to the accuracy of the models

[44]. In recent times, the photogrammetric and computer vision communities have

proposed combined procedures, for increasing both the measurement performance and

the automation levels. Examples are in the use of laser scanning and photogrammetry

[45–47] and in the use of stereo vision and photogrammetry for facade reconstruction

and object tracking [48, 50]. On the market, combinations of structured light projection

with photogrammetry are available (Atos II scanner,Gom GmbH,

Germany) for improved measurement performance.



4.4 Time of Flight

Surface range measurement can be made directly using the radar time-of-flight

principle. The emitter unit generates a laser pulse, which impinges onto the target

surface. A receiver detects the reflected pulse, and suitable electronics measures the

round trip travel time of the returning signal and its intensity. Single point sensors

perform point-to-point distance measurement, and scanning devices are combined to the

optical head to cover large bi-dimensional scenes. Large range sensors allow

measurement ranges from 15 m to 100 m (reflective markers must be put on the target

surfaces). Medium range sensors can acquire 3D data in shorter ranges. The

measurement resolutions vary with the range. For large measuring ranges, time-of flight

sensors give excellent results. On the other side, for smaller objects, about one meter in

size, attaining 1 part per 1,000 accuracy with time-of-flight radar requires very high

speed timing circuitry, because the time differences are in the pico-second range. A few

amplitude and frequency modulated radars have shown promise for close range distance

measurement [37, 51, 52]. In many applications, the technique is range-limited by

allowable power levels of laser radiation, determined by laser safety considerations.

Additionally, time-of-flight sensors face difficulties with shiny surfaces, which reflect

little back-scattered light energy except when oriented perpendicularly to the line of

sight.

4.5 Interferometry

Interferometric methods operate by projecting a spatially or temporally varying

periodic pattern onto a surface, followed by mixing the reflected light with a reference

pattern. The reference pattern demodulates the signal to reveal the variation in surface

geometry. The measurement resolution is very high, since it is a fraction of the

wavelength of the laser radiation. For this reason, surface quality control and

microprofilometry are the most explored applications. Use of multiple wavelengths and

of double heterodyne detection lengthens the non-ambiguity range. Systems based on

this principle are successfully used to calibrate CMMs. Holographic interferometry

relies on mixing coherent illumination with different wave vectors. Holographic



methods typically yield range accuracy of a fraction of the light wavelength over

microscopic fields of view [5, 53].

4.6 Moiré fringe range contours

Moiré fringe range contours are obtained by illuminating the scene with a light

source passing through a grating, and viewing the scene with a laterally displaced

camera through an identical second grating. The resulting interference pattern shows

equidistant contour lines, but there is no information about the corresponding depth

values. In addition, it is impossible to have direct information about the sign of the

concavities. Moirè methods can have phase discrimination problems when the surface

does not exhibit smooth shape variations [54].

4.7 Shape from focusing

Depth and range can be determined by exploiting the focal properties of a lens.

A camera lens can be used as a range finder by exploiting the “depth of view”

phenomena. In fact, the target image is blurred by an amount proportional to the

distance between points on the object and the in-focus object plane. This technique has

evolved from the passive approach to an active sensing strategy. In the passive case,

surface texture is used to determine the amount of blurring. Thus, the object must have

surface texture covering the whole surface in order to extract shape. The active version

of the method operates by projecting light onto the object to avoid difficulties in

discerning surface texture. Most prior work in active depth from focus has yielded

moderate accuracy up to one part per 400 over the field of view [55]. There is a direct

trade-of between depth of view and field of view; satisfactory depth resolution is

achieved at the expense of a sub-sampling the scene, which in turn requires some

form of mechanical scanning to acquire range measurement over the whole scene.

Moreover, spatial resolution is non-uniform. Specifically, depth resolution is

substantially less than resolution perpendicular to the observation axis. Finally, objects

not aligned perpendicularly to the optical axis and having a depth dimension greater

than the depth of view will come in to focus at different ranges, complicating the scene

analysis and interpretation [56].



4.8 Shape from shadows

This technique is a variant of the structured light approach. The 3D model of the

unknown object is rebuilt by capturing the shadow of a known object projected onto the

target when the light is moving. Low cost and simple hardware are the main advantages

of this approach, even if at the expense of low accuracy [57].

4.9 Shape from texture

The idea is to find possible transformations of texture elements (texels) to

reproduce the object surface orientation. For example, given a planar surface, the image

of a circular texel is an ellipse with varying values of its axes depending on its

orientation [58]. A review of the techniques developed in this area is in reference [37].

These methods yield simple, low cost hardware. However, the measurement quality is

low.

4.10 Strengths and weaknesses of 3D imaging techniques

The main characteristics of optical range imaging techniques are summarized in

Table . The comments on the table are quite general in nature, and a number of

exceptions are known. Strengths and weaknesses of the different techniques are

strongly application-dependent. The common attribute of being non-contact is an

important consideration in those applications characterized by either fragile,

deformable objects, or on-line measurements or hostile environments. Acquisition time

is another important aspect, when the overall cost of the measurement process suggests

the reduction of the operator time even at the expense of deteriorating the quality of the

data. On the other hand, active vision systems using a laser beam that illuminates the

object inherently involve safety considerations, and possible surface interaction effects.

In addition, the speckle effect introduces disturbances that represent a serious limit to

the accuracy of the measurements [60]. Fortunately, recent advances in LEDs

technology yield a valuable solution to this problem, since they are suitable candidates

as light projectors in active triangulation systems. The dramatic evolution of CPUs and

memories has led in the last three years to elaboration performances that were by far



impossible before, at low costs. For this reasons, techniques that are computation

demanding (e.g., passive stereo vision) are now more reliable and efficient. The

selection of which sensor type should be used to solve a given depth measurement

problem is a very complex task, that must consider (i) the measurement time, (ii) the

budget, and (iii) the quality expected from the measurement. In this respect, 3D imaging

sensors may be affected by missing or bad quality data. Reasons are related to the

optical geometry of the system, the type of projector and/or acquisition optical device,

the measurement technique and the characteristics of the target objects. The sensor

performance may depend on the dimension, the shape, the texture, the temperature and

the accessibility of the object. Relevant factors that influence the choice are also the

ruggedness, the portability, and the adaptiveness of the sensor to the measurement

problem, the easiness of the data handling, and the simplicity of use of the sensor.



Experience in the application of 3D imaging sensors

A wide range of applications is possible nowadays, thanks to the availability of

engineered sensors, of laboratory prototypes, and of suitable software environments to

elaborate the measurements. This section presents a brief insight of the most important

fields where 3D imaging sensors can be fruitfully used. For each one, an application

example that has been carried out in our Laboratory is presented.

13.1 Industrial applications

Typical measurement problems in the industrial field deal with (i) the control of

machined surfaces, for the quantitative measure of roughness, waviness and form

factor, (ii) the dimensional measurement and the quality control of products, and (iii)

the reverse engineering of complex shapes.

13.2 Surface quality control

Microprofilometry techniques and sensors are the best suited to carry out the

measurement processes in the surface control field. In this application, the objective is

to gauge the 3D quantitative

FUTURE SCOPE

Antireflection coating film deposition and RGB filter deposition can be

used to enhance sensitivity and for colour sensing

Seminar Report

Documents

Transcript of Seminar Report