Conference PaperPDF Available

Authors:

## Figures

Content may be subject to copyright.
F. Villa, Member IEEE, R. Lussana, D. Bronzi, F. Zappa, Senior Member IEEE
Dipartimento di Elettronica, Informazione e Bioingegneria
Politecnico di Milano, Milano, Italy
federica.villa@polimi.it
A. Giudice
Micro Photon Devices srl
Bolzano, Italy
andrea.giudice@micro-photon-devices.com
Abstract We present a 3D-ranging camera based on the
optical indirect Time-of-Flight technique, suitable for automotive
applications. The camera is based on a 64×32-pixel chip,
integrating a Single-Photon Avalanche Diode (SPAD) in each
pixel, fabricated in an automotive-certified CMOS technology,
and on an 808 nm active illuminator, emitting 1.5 W average
optical power. Thanks to the SPAD single-photon sensitivity and
the in-pixel processing, the camera provides a precision better
than 1 m over a 40° × 20° field-of-view, at 45 m distance.
Keywords — Advanced Driver Assistance Systems; Indirect
Time-of-Flight; 3D ranging; SPAD camera; photon-counting.
I. INTRODUCTION
During the last few decades, many automotive companies
have developed more and more complete Advanced Driver
and at mitigating their effects. In order to perform tasks such
as collision avoidance and adaptive cruise control, many of
these systems employ radar-, lidar-, ultrasonic- or camera-
based depth sensors for mapping the environment surrounding
the vehicle. These sensors show complementary strengths in
measuring certain object parameters: for instance, radars have
a long detection range, but their field-of-view is much
narrower than camera-based systems [1].
Camera-based 3D vision systems can be grouped into two
main categories, namely stereo-vision (SV) and time-of-flight
(TOF) ones. SV systems employ two cameras to provide high
spatial resolution at low power consumption, but they are only
suitable for high-contrast scenes and they require high
computational efforts to solve the correspondence problem for
matching the information from the two cameras. On the other
hand, TOF vision systems make use of a single camera,
synchronized with an active light source, and require less data
processing, thus being able to achieve high frame-rates. The
direct time-of-flight (dTOF) is the most straightforward TOF
technique, relying on the measurement of the round-trip
duration of a laser or LED light pulse, backscattered by the
target. The trade-off between maximum range and power
consumption can be mitigated by employing photodetectors
with very high sensitivity, such as the Single-Photon
Avalanche Diodes (SPADs), which are sensitive to single
photons in addition to providing precise timing information
[2]. However, many cameras that implement single-photon
dTOF technique cannot be operated in full daylight conditions,
since they can only measure the arrival time of the first photon
in each frame, thus being easily saturated by ambient light [3]-
[6]. Indirect Time-of-Flight (iTOF) systems represent an
alternative solution, in which the distance information is
computed by measuring the phase-delay between a
continuous-wave (CW) excitation shone toward the target and
its back-reflected light echo. This technique does not need
precise arrival time estimation and it can be implemented
either with linear-mode (e.g. CMOS and CCD) detectors,
which provide an analog voltage proportional to the light
intensity, or with photon counting (e.g. SPAD) detectors,
which instead provide a digital pulse for every detected
photon. In short-medium range applications, the iTOF
technique could be preferable compared to dToF, since it
results in a simpler, more compact and cost-effective system,
requiring neither high-bandwidth electronics nor very-short
pulse-width lasers.
Typically, SPAD arrays feature lower fill-factor and lower
quantum efficiency at near-infrared wavelengths with respect
to linear detectors, but also inherently better timing resolution
(since the SPAD timing jitter is typically just few tens of
picoseconds) and higher accuracy (impaired only by photon
shot-noise and not by readout noise) [7].
In this paper, we present the SPAD camera we designed
for optical 3D iToF ranging. The camera is based on a 64×32-
pixel CMOS SPAD imager for simultaneous two-dimensional
(2D) imaging and three-dimensional (3D) ranging, and an eye-
safe near-infrared active illuminator at 808 nm. We validated
the system in outdoor scenarios, yielding 110 dB dynamic-
range, high-speed (100 fps, frames per second) depth
measurements in light-starved environment, with better than
1 m precision at 45 m distance.
II. INDIRECT TOF TECHNIQUES
Two different techniques can be exploited for iTOF
measurements: continuous-wave iTOF (CW-iTOF) and
pulsed-light iTOF (PL-iTOF).
In the CW-iTOF technique, the active illumination is
sinusoid modulated, with modulation period TP, and the
reflected light reaches back the detector with a phase-shift of
Δφ. The maximum unambiguous 3D range is set by the
modulation period and is computed as dMAX=TPc/2, given the
speed of light c. The object’s distance d is computed as:
22 2
P
MAX
cT
dd


(1)
To retrieve phase-shift information, the reflected wave is
synchronously sampled by four integration windows of same
duration TTAP (with TTAP=1/4·TP), thus providing C0, C1, C2
and C3 samples. Phase delay Δφ, reflected light intensity AR
and background B can be computed through Discrete Fourier
Transform, and are given by:
31
02
arctan CC
CC

(2)
22
31 0 2
()( )
sinc( )
R
TAP TAP P
CC CC
ATTT

 (3)
0123
42
R
TAP
CCCC
A
BT


(4)
In PL-iTOF systems, instead, an active illuminator emits
two light pulses with amplitude A and duration TP, which sets
the maximum distance range to dMAX=TPc/2. The back-
reflected signal, together with background light and detector
noise, is integrated within three distinct time slots with the
same duration, TP: the first slot (W0) is synchronous with the
first laser excitation, the second one (W1) starts at the end of
the second laser pulse and the third one (WB) is acquired
without any laser excitation, just for measuring the
background [8]. If C0, C1, and CB are the counts accumulated
in W0, W1, and WB, respectively, then object distance d, active-
light intensity AR, and background B are given by:
1
01
22
PB
B
cT C C
dCC C





(5)
01
2
B
R
P
CC C
AT

(6)
P
C
BT
(7)
By applying the error propagation rule, it is easy to
demonstrate that for both CW-iTOF and PL-iTOF techniques
depth-precision depends on distance range (dMAX), received
light intensity (AR) and background noise (B) [9]. Moreover, in
PL-iTOF only a fraction of the echo light signal is acquired in
the windows W0 and W1, depending on the object distance,
which therefore strongly influences the measurement
precision. In CW-iTOF, instead, the whole light echo
waveform is collected and the 3D precision depends on
distance only, because of the geometrical attenuation of the
reflected signal. Therefore, with the same average power, PL-
iTOF requires higher peak power, or longer integration time.
Furthermore, while CW-iTOF employs only one period (TP)
for a complete measurement, PL-iTOF requires three periods
to assess the distance. For these reasons, CW is preferred for
iToF-based systems.
III. INDIRECT TOF CAMERA
We developed a 3D ranging camera for medium-distance
automotive applications, aimed at minimizing both system
complexity and power consumption. For the above reasons,
we chose to implement the iTOF technique, employing a CW
illumination to optimize the precision along the entire
measurement range, as explained in Section II.
Fig. 1: Picture of the 3D SPAD system (sensor module and laser illuminator),
mounted on the roof of a car, together with a GoPro standard action camera.
A picture of the complete system (comprising the SPAD
camera and the laser illuminator), mounted on the roof of a
car, is shown in Fig. 1. We also developed simple post
processing algorithms for added information, such as profiles
recognition and distance labelling, from the acquired videos.
The proposed image sensor is based on a CMOS SPAD
imager, which is controlled by an FPGA board for settings,
data-readout, and USB data transfer. Each of the 64×32 pixels
integrates a SPAD detector, a quenching circuit, shaping
electronics, three 9-bit counters and their respective storage
memories. Each SPAD features very good performance in
terms of intrinsic noise (dark counts and afterpulsing), and
about 5% efficiency at 808 nm [11]. The three 9-bit counters
are used for in-pixel demodulation, providing respectively the
counts C3-C1, C0-C1 and C0+C1+C2+C3, which are required by
Eq.s (2)-(4). Thanks to in-pixel memories, the array works in a
fully-parallel fashion: at the end of each frame, the output
from each counter is stored into an in-pixel register, and a new
frame can be acquired concurrently with the read-out of the
previous one [10]. Thanks to this global shutter feature, the
acquired image undergoes neither deformation (jello effect)
nor motion artifacts, even in presence of fast moving objects.
In order to operate the SPAD sensor chip, we developed a
complete high-speed camera module composed by three
printed circuit boards: a board hosting the chip, a board with a
Spartan-3 FPGA and a third board in charge of generating
power supplies and providing arbitrary analog waveforms
modulation to the light source. The camera is housed in an
aluminum case, supporting a 12 mm F/1.4 C-mount imaging
lens, whose field-of-view is about 40° × 20° (H × V). The
whole camera is very rugged and compact, with 80 mm ×
70 mm × 45 mm dimensions, and consumes about 1 W,
mostly due to the FPGA.
A MATLAB interface allows to set parameters (e.g. frame
duration, number of frames to be acquired, modulation
frequency, etc.) and to post-process data (see Sect. III-C).
B. Continuous-wave illuminator
The illuminator has a modular design based on a
power-board and five laser driver cards, each one mounting 3
laser diodes with peak CW power of 200 mW at 808 nm
wavelength (thus the total peak optical power is 3 W).
Fig. 2: Emitted light spectra of the 15 laser diodes, centered around 808 nm.
The current into each channel is controlled by two signals
driven by the camera: the enable signal (EN) switches ON and
OFF the current in the laser, while the Current control Input
(CI) is an analog voltage signal controlling the LD current.
As already explained in Sect. II, the maximum non-
ambiguous range is typically limited by the modulation
frequency (the lower the modulation frequency, the longer the
maximum range). On the other hand, low modulation
frequencies have detrimental effects on the 3D-ranging
precision. Therefore, in order to achieve the required range
frequency continuous wave (DFCW) modulation, where each
frame is acquired twice, using two different modulation
frequencies, thus allowing to extend the maximum non-
ambiguous range (dMAX) without impairing measurement
precision [12]. In our case, we implemented an 8.333 MHz
(18 m range) and a 5 MHz (30 m range) modulation
frequency, for achieving a final distance range of 45 m.
The illuminator performance plays a significant role in
determining the final measurement precision and accuracy.
For instance, the spectral cleanness of the illumination light
sine wave, and in particular the presence of odd harmonics,
has a direct impact on linearity errors. Moreover, another
critical parameter is the modulation contrast (defined as the
ratio between DC value and fundamental frequency amplitude,
for a pure sinusoidal waveform), which needs to be
maximized in order to minimize the contribution of the laser
to the background DC light. Our illuminator features a very
good 35 dB ratio between first and third harmonics power, and
a nearly unitary modulation contrast. We also measured the
emission wavelength of each of the 15 laser diodes, whose
illumination spectra are plotted in Fig. 2. The dispersion of the
peak emission around the targeted 808 nm influences the
selection of the optical filters to be placed in front of the
camera optical aperture. As the best compromise to optimize
Signal-to-Background ratio, we chose the band-pass filter
FB800-40 by Thorlabs (central wavelength 800 nm, 40 nm
full-width at half maximum).
Finally, it is important to observe that the interference of
many illuminators belonging to different cameras does not
prevent 3D measurements, but only slightly degrades the
performance. In fact, different cameras’ clocks, although
running at same nominal frequencies, are not correlated;
hence, the disturbing illumination contributes as a common-
mode signal that is rejected through the in-pixel demodulation.
Fig. 3: Main steps of the post-processing algorithms.
C. Post-processing
We conceived and implemented simple post processing
algorithms in MATLAB, to recognize objects and mark their
distance. The main steps are shown in Fig. 3. The top images
show the total background B (left) and the active light Ar
(right) intensities. The middle ones show the 3D image after a
two dimensional median filtering to improve the quality of the
image (left) and after censuring pixels imaging the floor
(right). For the floor cancellation, it can be intended as an
object in the lower half part of the image, which is far from
the camera in the middle rows of the image and it is close in
the lower rows. If an object rises from the floor, it covers the
floor behind it and the distance measurement in those pixels is
less than in the case of only floor acquisition. With simple
trigonometric equations it is possible to compute the expected
distance of the floor for every pixel. If the distance evaluated
for a pixel is less than that expected for the floor there is an
object closer to the camera and the distance measurement is
kept. On the contrary, the pixel is collecting light from a
portion of the floor and the distance measurement is neglected.
The bottom images represent the segmented image (left) and
the superimposed 2D and 3D information, with pedestrian
recognition and its distance marked and labelled (right). Both
active-light image and 3D image are segmented in order to
determine clusters of pixels relating to the objects in the scene.
Every cluster holds information about the object which is
connected to (horizontal and vertical dimensions,
average/minimum/maximum distance, border shape, etc...).
All these information can be computed and displayed in
real-time. Moreover, it is possible to track a single object in
order to measure its speed with respect to the speed of the
camera and eventually some information about the expected
movements in front of the camera (e.g. by a Kalman filter).
Fig. 4: Typical automotive outdoor scenarios acquired with a standard action camera (top) and the SPAD camera (bottom), running at 100 fps.
IV. AUTOMOTIVE TESTS
illuminator source), was installed on a car, together with a
standard action camera (Hero3 by GoPro) for co-registration
(Fig. 1), and tested in real automotive scenarios.
Many measurements were performed during afternoon and
evening hours, also in adverse (foggy and light rain) weather
conditions, with less than 1,000 lux of background
illumination. Fig. 4 shows two scenes, both acquired with the
GoPro camera and with the SPAD camera running at 100 fps,
i.e. with 10 ms frame-time. In Fig. 4 left, there are two
pedestrians crossing the street at 7.2 m and 7.5 m respectively
and a tram at 22 m; in Fig. 4 right, there are a pedestrian and a
cyclist on the zebra crossings at 6.8 m and 8.0 m and a car
behind them at 11.7 m. As can be seen, even with the low
2048 pixels resolution of the SPAD camera, it is possible to
identify pedestrians and discriminate them from vehicles (e.g.
cars or trucks) and other street signs, through some post-
processing. From our measurements, we can infer that the
illumination system is able to uniformly illuminate the scene
without shadow zones and, although the camera resolution is
limited to 2048 pixels, the integration of all information
provided by the SPAD sensor enables to generate images rich
of details, thus allowing to locate and recognize objects in the
scene through real-time image processing. A video, showing
all the potentialities of the automotive 3D SPAD camera is
available at [13].
We performed other measurements in daylight conditions,
with a background illumination of about 33,000 lux. In those
measurements, the modulated signal reflected from the
vehicles (
100 counts per frame, cpf) was overwhelmed by
the background light (
10,000 cpf) and only the light received
from license plates and the tail lights (which are not
lambertian reflectors, but retro-reflectors, as well as road
signs) was high enough (
1,000 cpf) to allow distance
detection.
Thanks to superimposition of 2D and 3D information, cars
and road signs may be accurately detected even with strong
sunlight by using a proper image segmentation algorithm. To
this aim, Fig. 5 shows a frame of a movie acquired by the
SPAD camera with 100 µs frames and averaged over 100
frames, i.e. with a total equivalent 3D integration time of
10 ms, corresponding to 100 fps. As can be seen, seven
pedestrians are standing in the scene at different distances,
from 3.2 m up to 6.2 m. The picture at the top shows the raw
3D acquisition, where the false colors represent the distance
(see the scale on the right-hand side). Instead, the bottom
picture shows the overlay of the 3D depth-resolved map with
the 2D intensity map, obtained by computing the reflected
light intensity AR of Eq. (3).
Fig. 5: Frame from a 3D movie acquired over-imposing the 3D map distance
information with the 2D photon intensity, both provided concurrently by the
Fig. 6: Horizontal multishots panorama of 3D ranging images over a 130° × 20° final field-of-view (targets were between 3 m and 10 m distance).
objects (e.g. the separation among face, sweater and trousers).
The different colors represent different distances, while the
different color luminosities identify different reflected
intensities. Therefore, the pedestrian in center position is in
yellow false-color (corresponding to a 3D distance of 6 m),
but hairs are dark, and trousers are more opaque than the shirt.
All measurements shown so far were acquired with the
SPAD camera fixed on the vehicle roof, for ease of
installation. In order to cover a larger field-of-view compared
to the intrinsic 40° × 20° one, we implemented also a simple
angular scanning. Fig. 6 shows a scene acquired in four
consecutive shots over a 90° angular rotation (a final 130° ×
20° field-of-view) and the resulting panorama assembly of
four 64×32 images, resulting in a 3D image of 210×32 pixels.
V. CONCLUSIONS
We have presented a 3D vision system based on a SPAD
image sensor, manufactured in a cost-effective 0.35 µm
automotive-certified CMOS technology and an 808 nm laser
illuminator. The active illuminator outputs low-power eye-safe
light, thanks to SPAD detectors single-photons sensitivity in
the near-infrared wavelength range. Each pixel of the 64×32
SPAD imager can acquire simultaneously 2D and 3D images
from the scene under observation, through indirect Time-of-
Flight measurement, i.e. by counting photons in four time-
slots synchronized with the continuous-wave (CW-iTOF),
sinusoid modulated, active illumination. We validated the 3D
camera in real outdoor automotive scenario both under low
ambient light condition (<1,000 lux) and in daylight
(33,000 lux). In low ambient light scenarios, we acquired 3D
maps of targets located up to 45 m away, in a field-of-view of
40° × 20°, with better than 1 m precision at the farthest
distance. We also over-imposed 3D depth-resolved maps with
2D intensity data, both concurrently provided by the SPAD
camera, to augment data from the scene under observation.
Finally, we operated the camera in an angular scanning
system, to exceed the intrinsic field-of-view. For example, we
reached a 130° × 20° final field-of-view, by means of four
angular scan and a resulting resolution of 210×32 pixels.
For the future, we envision to further improve the camera
functionalities and performance. For instance, object
recognition will be implemented within the FPGA, in order to
have a standalone system able to communicate over CAN bus
with the vehicle control unit, to fulfill functional requirements
of a collisions mitigation system.
ACKNOWLEDGMENTS
This work has been partially supported by the “DEIS”
project (Dependability Engineering Innovation for automotive
CPS), funded by the European Union’s Horizon 2020 research
and innovation programme, under the grant no. 732242.
VI. REFERENCES
[1] D. Bronzi, F. Villa, S. Tisa, A. Tosi, and F. Zappa, “SPAD Figures of
Merit for photon-counting, photon-timing, and imaging applications,”
IEEE Sensors Journal, Vol. 16, No. 1, pp. 3-12, Jan. 1, 2016.
[2] R. Rasshofer and K. Gresser, “Automotive Radar and Lidar Systems for
no. 10, pp. 205–209, 2005.
[3] C. Veerappan, J. Richardson, R. Walker, D. Li, M.W. Fishburn, et al. “A
160×128 Single-Photon Image Sensor with On-Pixel 55ps 10b Time-to-
Digital Converter”, IEEE Int. Solid-State Circuits Conf., 2011.
[4] M. Gersbach, Y. Maruyama, R. Trimananda, M.W. Fishburn, et al. “A
Time-Resolved, Low-Noise Single-Photon Image Sensor Fabricated in
Deep-Submicron CMOS Technology,” Journal of Solid-State Circuits,
Vol. 47 no. 6, pp. 1394-1407, June 2012.
[5] F. Villa, R. Lussana, D. Bronzi, S. Tisa, A. Tosi, F. Zappa, A. Dalla
Mora, D. Contini, D. Durini, S. Weyers, W. Brockherde, “CMOS imager
with 1024 SPADs and TDCs for single-photon timing and 3D time-of-
flight”, IEEE Journal of Selected Topics in Quantum Electronics, Vol.
20, no. 6, Nov-Dec. 2014.
[6] R. Lussana, F. Villa, A. Dalla Mora, D. Contini, A. Tosi, and F. Zappa,
“Enhanced single-photon time-of-flight 3D ranging,” Opt. Express, Vol.
23, pp. 24962-24973, 2015.
[7] C. Niclass, C. Favi, T. Kluter, F. Monnier, and E. Charbon, “Single-
Photon Synchronous Detection,” IEEE J. Solid-State Circuits, vol. 44,
no. 7, pp. 1977–1989, Jul. 2009.
[8] S. Bellisai, D. Bronzi, F. Villa, S. Tisa, A. Tosi, and F. Zappa, “Single-
photon pulsed-light indirect time-of-flight 3D ranging,” Opt. Express,
vol. 21, no. 4, pp. 5086-5098, Feb. 2013.
[9] D. Bronzi, Y. Zou, F. Villa, S. Tisa, A. Tosi, and F. Zappa, “Automotive
Three-Dimensional vision through a Single-Photon Counting SPAD
camera,” IEEE Transaction on Intelligent Transportation Systems, vol.
17, no. 3, March 2016.
[10] D. Bronzi, F. Villa, S. Tisa, A. Tosi, F. Zappa, D. Durini, S. Weyers, and
W. Brockherde, “100,000 frames/s 64×32 single-photon detector array
for 2D imaging and 3D ranging,” IEEE J. Selected Topics in Quant.
Electronics, Vol. 20, No. 6, Nov./Dec. 2014
[11] F. Villa, D. Bronzi, Y. Zou, C. Scarcella, G. Boso, S. Tisa, A. Tosi, F.
Zappa, D. Durini, S. Weyers, W. Brockherde, U. Paschen, “CMOS
SPADs with up to 500 µm diameter and 55% detection efficiency at
420 nm,” Journal of Modern Optics, Jan. 2014.
[12] A. D. Payne, A. P. P. Jongenelen, A. A. Dorrington, M. J. Cree, and D.
A. Carnegie, “Multiple frequency range imaging to remove
measurement ambiguity,” in Proc. Optical 3-D Measurement
Techniques, pp. 139–148, Jul. 2009.
[13] D. Bronzi. (2014). SP-ADAS: High-Speed Single-Photon Camera for
Advanced Driver Assistance Systems [Online]. Available:
2017.
... Single-photon counting techniques have been used in low-light sensing applications such as LIDAR [1,2], quantum key distribution [3], medical imaging technology [4], and 3D imaging technology [5,6]. Photomultiplier tubes have been the traditional solution for photon counting applications, but in recent years single-photon avalanche diode (SPAD) has become an alternative candidate due to its lower cost, lower operating voltage, higher sensitivity, and smaller size. ...
Article
Full-text available
A compact single-photon counting module that can accurately control the bias voltage and hold-off time is developed in this work. The module is a microcontroller-based system which mainly consists of a microcontroller, a programmable negative voltage generator, a silicon-based single-photon avalanche diode, and an integrated active quench and reset circuit. The module is 3.8 cm × 3.6 cm × 2 cm in size and can communicate with the end user and be powered through a USB cable (5 V). In this module, the bias voltage of the single-photon avalanche diode (SPAD) is precisely controllable from −14 V ~ −38 V and the hold-off time (consequently the dead time) of the SPAD can be adjusted from a few nanoseconds to around 1.6 μs with a setting resolution of ∼6.5 ns. Experimental results show that the module achieves a minimum dead time of around 28.5 ns, giving a saturation counting rate of around 35 Mcounts/s. Results also show that at a controlled reverse bias voltage of 26.8 V, the dark count rate measured is about 300 counts/s and the timing jitter measured is about 158 ps. Photodetection probability measurements show that the module is suited for detection of visible light from 450 nm to 800 nm with a 40% peak photon detection efficiency achieved at around 600 nm.
... However, its operative range is short for automotive applications (10-20 meters) and has problems working under intense ambient light. Some research lines as indirect time-of-flight [21], pulsed light time-of-flight or avalanche photodiodes [22] could increase working range to 50-250 meters. ...
Article
Full-text available
After more than 20 years of research, ADAS are common in modern vehicles available in the market. Automated Driving systems, still in research phase and limited in their capabilities, are starting early commercial tests in public roads. These systems rely on the information provided by on-board sensors, which allow to describe the state of the vehicle, its environment and other actors. Selection and arrangement of sensors represent a key factor in the design of the system. This survey reviews existing, novel and upcoming sensor technologies, applied to common perception tasks for ADAS and Automated Driving. They are put in context making a historical review of the most relevant demonstrations on Automated Driving, focused on their sensing setup. Finally, the article presents a snapshot of the future challenges for sensing technologies and perception, finishing with an overview of the commercial initiatives and manufacturers alliances that will show the intention of the market in sensors technologies for Automated Vehicles.
Article
Full-text available
We present an optical 3-D ranging camera for automotive applications that is able to provide a centimeter depth resolution over a 40° x 20° field of view up to 45 m with just 1.5 W of active illumination at 808 nm. The enabling technology we developed is based on a CMOS imager chip of 64 x 32 pixels, each with a single-photon avalanche diode (SPAD) and three 9-bit digital counters, able to perform lock-in time-of-flight calculation of individual photons emitted by a laser illuminator, reflected by the objects in the scene, and eventually detected by the camera. Due to the SPAD single-photon sensitivity and the smart in-pixel processing, the camera provides state-of-the-art performance at both high frame rates and very low light levels without the need for scanning and with global shutter benefits. Furthermore, the CMOS process is automotive certified.
Article
Full-text available
SPADs (Single-Photon Avalanche Diodes) emerged as the most suitable photodetectors for both singlephoton counting and photon-timing applications. Different complementary metal-oxide semiconductor (CMOS) devices have been reported in literature, with quite different performance and some excelling in just few of them, but often at different operating conditions. In order to provide proper criteria for performance assessment, we present some figures of merit (FoMs) able to summarize the typical SPAD performance (i.e. photon detection efficiency, dark counting rate, afterpulsing probability, hold-off time, and timing jitter) and to identify a proper metric for SPAD comparisons, when used either as single pixel detectors or in imaging arrays. The ultimate goal is not to define a ranking list of best-in-class detectors, but to quantitatively help the end-user to state the overall performance of different SPADs in either photon-counting, timing, or imaging applications. We review many CMOS SPADs from different research groups and companies, we compute the proposed FoMs for all them and, eventually, we provide an insight on present CMOS SPAD technologies and future trends.
Conference Paper
Full-text available
Advanced Driver Assistance Systems (ADAS) are the most advanced technologies to fight road accidents. Within ADAS, an important role is played by radar- and lidar-based sensors, which are mostly employed for collision avoidance and adaptive cruise control. Nonetheless, they have a narrow field-of-view and a limited ability to detect and differentiate objects. Standard camera-based technologies (e.g. stereovision) could balance these weaknesses, but they are currently not able to fulfill all automotive requirements (distance range, accuracy, acquisition speed, and frame-rate). To this purpose, we developed an automotive-oriented CMOS single-photon camera for optical 3D ranging based on indirect time-of-flight (iTOF) measurements. Imagers based on Single-photon avalanche diode (SPAD) arrays offer higher sensitivity with respect to CCD/CMOS rangefinders, have inherent better time resolution, higher accuracy and better linearity. Moreover, iTOF requires neither high bandwidth electronics nor short-pulsed lasers, hence allowing the development of cost-effective systems. The CMOS SPAD sensor is based on 64 × 32 pixels, each able to process both 2D intensity-data and 3D depth-ranging information, with background suppression. Pixel-level memories allow fully parallel imaging and prevents motion artefacts (skew, wobble, motion blur) and partial exposure effects, which otherwise would hinder the detection of fast moving objects. The camera is housed in an aluminum case supporting a 12 mm F/1.4 C-mount imaging lens, with a 40°×20° field-of-view. The whole system is very rugged and compact and a perfect solution for vehicle’s cockpit, with dimensions of 80 mm × 45 mm × 70 mm, and less that 1 W consumption. To provide the required optical power (1.5 W, eye safe) and to allow fast (up to 25 MHz) modulation of the active illumination, we developed a modular laser source, based on five laser driver cards, with three 808 nm lasers each. We present the full characterization of the 3D automotive system, operated both at night and during daytime, in both indoor and outdoor, in real traffic, scenario. The achieved long-range (up to 45m), high dynamic-range (118 dB), highspeed (over 200 fps) 3D depth measurement, and high precision (better than 90 cm at 45 m), highlight the excellent performance of this CMOS SPAD camera for automotive applications.
Article
Full-text available
We present a CMOS imager consisting of 32×32 smart pixels, each one able to detect single photons in the 300-900 nm wavelength range and to perform both photon-counting and photon-timing operations on very fast optical events with faint intensities. In photon-counting mode, the imager provides photon-number (i.e, intensity) resolved movies of the scene under observation, up to 100 000 frames/s. In photon-timing, the imager provides photon arrival times with 312 ps resolution. The result are videos with either time-resolved (e.g., fluorescence) maps of a sample, or 3-D depth-resolved maps of a target scene. The imager is fabricated in a cost-effective 0.35-μm CMOS technology, automotive certified. Each pixel consists of a single-photon avalanche diode with 30 μm photoactive diameter, coupled to an in-pixel 10-bit time-to-digital converter with 320-ns full-scale range, an INL of 10% LSB and a DNL of 2% LSB. The chip operates in global shutter mode, with full frame times down to 10 μs and just 1-ns conversion time. The reconfigurable imager design enables a broad set of applications, like time-resolved spectroscopy, fluorescence lifetime imaging, diffusive optical tomography, molecular imaging, time-of-flight 3-D ranging and atmospheric layer sensing through LIDAR.
Article
Full-text available
We report on the design and characterization of a multipurpose 64x32CMOS single-photon avalanche diode (SPAD) array. The chip is fabricated in a high-voltage 0.35-μm CMOS technology and consists of 2048 pixels, each combining a very low noise (100 cps at 5-V excess bias) 30-μm SPAD, a prompt avalanche sensing circuit, and digital processing electronics. The array not only delivers two-dimensional intensity information through photon counting in either free-running (down to 10-μs integration time) or time-gated mode, but can also perform smart light demodulation with in-pixel background suppression. The latter feature enables phase-resolved imaging for extracting either three-dimensional depth-resolved images or decay lifetime maps, by measuring the phase shift between a modulated excitation light and the reflected photons. Pixel-level memories enable fully parallel processing and global-shutter readout, preventing motion artifacts (e.g., skew, wobble, motion blur) and partial exposure effects. The array is able to acquire very fast optical events at high frame-rate (up to 100 000 fps) and at single-photon level. Low-noise SPADs ensure high dynamic range (up to 110 dB at 100 fps) with peak photon detection efficiency of almost 50% at 410 nm. The SPAD imager provides different operating modes, thus, enabling both time-domain applications, like fluorescence lifetime imaging (FLIM) and fluorescence correlation spectroscopy, as well as frequency-domain FLIM and lock-in 3-D ranging for automotive vision and lidar.
Article
Full-text available
Many demanding applications require single-photon detectors with very large active area, very low noise, high detection efficiency, and precise time response. Single-photon avalanche diodes (SPADs) provide all the advantages of solid-state devices, but in many applications other single-photon detectors, like photomultiplier tubes, have been preferred so far due to their larger active area. We developed silicon SPADs with active area diameters as large as 500 μm in a fully standard CMOS process. The 500 μm SPAD exhibits 55% peak photon detection efficiency at 420 nm, 8 kcps of dark counting rate at 0°C, and high uniformity of the sensitivity in the active area. These devices can be used with on-chip integrated quenching circuitry, which reduces the afterpulsing probability, or with external circuits to achieve even better photon-timing performances, as good as 92 ps FWHM for a 100 μm diameter SPAD. Owing to the state-of-the-art performance, not only compared to CMOS SPADs but also SPADs developed in custom technologies, very high uniformity and low crosstalk probability, these CMOS SPADs can be successfully employed in detector arrays and single-chip imagers for single-photon counting and timing applications.
Article
Full-text available
"Indirect" time-of-flight is one technique to obtain depth-resolved images through active illumination that is becoming more popular in the recent years. Several methods and light timing patterns are used nowadays, aimed at improving measurement precision with smarter algorithms, while using less and less light power. Purpose of this work is to present an indirect time-of-flight imaging camera based on pulsed-light active illumination and a 32 × 32 single-photon avalanche diode array with an improved illumination timing pattern, able to increase depth resolution and to reach single-photon level sensitivity.
Article
Full-text available
Phase and intensity of light are detected simultaneously using a fully digital imaging technique: single-photon synchronous detection. This approach has been theoretically and experimentally investigated in this paper. We designed a fully integrated camera implementing the new technique that was fabricated in a 0.35 mum CMOS technology. The camera demonstrator features a modulated light source, so as to independently capture the time-of-flight of the photons reflected by a target, thereby reconstructing a depth map of the scene. The camera also enables image enhancement of 2D scenes when used in passive mode, where differential maps of the reflection patterns are the basis for advanced image processing algorithms. Extensive testing has shown the suitability of the technique and confirmed phase accuracy predictions. Experimental results showed that the proposed rangefinder method is effective. Distance measurement performance was characterized with a maximum nonlinearity error lower than 12 cm within a range of a few meters. In the same range, the maximum repeatability error was 3.8 cm.
Article
We developed a system for acquiring 3D depth-resolved maps by measuring the Time-of-Flight (TOF) of single photons. It is based on a CMOS 32 × 32 array of Single-Photon Avalanche Diodes (SPADs) and 350 ps resolution Time-to-Digital Converters (TDCs) into each pixel, able to provide photon-counting or photon-timing frames every 10 μs. We show how such a system can be used to scan large scenes in just hundreds of milliseconds. Moreover, we show how to exploit TDC unwarping and refolding for improving signal-to-noise ratio and extending the full-scale depth range. Additionally, we merged 2D and 3D information in a single image, for easing object recognition and tracking.
Article
We report on the design and characterization of a novel time-resolved image sensor fabricated in a 130 nm CMOS process. Each pixel within the 32$\times$32 pixel array contains a low-noise single-photon detector and a high-precision time-to-digital converter (TDC). The 10-bit TDC exhibits a timing resolution of 119 ps with a timing uniformity across the entire array of less than 2 LSBs. The differential non-linearity (DNL) and integral non-linearity (INL) were measured at ±0.4 and ±1.2 LSBs, respectively. The pixel array was fabricated with a pitch of 50 µm in both directions and with a total TDC area of less than 2000 µm². The target application for this sensor is time-resolved imaging, in particular fluorescence lifetime imaging microscopy and 3D imaging. The characterization shows the suitability of the proposed sensor technology for these applications.