Académique Documents
Professionnel Documents
Culture Documents
www.elsevier.com/locate/micpro
Biorobotics Laboratory, Movement and Perception Institute, CNRS & University of the Mediterranean, 163 Avenue Luminy,
CP 938, F-13288 Marseille cedex 09, France
Available online 27 February 2007
Abstract
Tomorrows Micro-Air-Vehicles (MAVs) could be used as scouts in many civil and military missions without any risk to human life.
MAVs have to be equipped with sensors of several kinds for stabilization and guidance purposes. Many recent ndings have shown, for
example, that complex tasks such as 3-D navigation can be performed by insects using optic ow (OF) sensors although insects eyes have
a rather poor spatial resolution. At our Laboratory, we have been performing electrophysiological, micro-optical, neuroanatomical and
behavioral studies for several decades on the houseys visual system, with a view to understanding the neural principles underlying OF
detection and establishing how OF sensors might contribute to performing basic navigational tasks. Based on these studies, we developed
a functional model for an Elementary Motion Detector (EMD), which we rst transcribed into electronic terms in 1986 and subsequently
used onboard several terrestrial and aerial robots. Here we present a Field Programmable Gate Array (FPGA) implementation of an
EMD array, which was designed for estimating the OF in various parts of the visual eld of a MAV. FPGA technology is particularly
suitable for applications of this kind, where a single Integrated Circuit (IC) can receive inputs from several photoreceptors of similar (or
dierent) shapes and sizes located in various parts of the visual eld. In addition, the remarkable characteristics of present-day FPGA
applications (their high clock frequency, large number of system gates, embedded RAM blocks and Intellectual Property (IP) functions,
small size, light weight, low cost, etc.) make for the exible design of a multi-EMD visual system and its installation onboard MAVs with
extremely low permissible avionic payloads.
2007 Elsevier B.V. All rights reserved.
Keywords: Optic ow sensor; Elementary Motion Detector; Field Programmable Gate Array; Micro-Air-Vehicle; Biorobotics
1. Introduction
One recent trend in the eld of Unmanned Air Vehicle
(UAV) and robotic aircraft design has been the development of Micro-Air-Vehicles (MAVs) in the 150 cm size
409
Fig. 1. Several visual sensors covering various Fields of View for the
guidance of a Micro-Air-Vehicle.
410
K Du K 0
Dt
Dt
XEMD e s () K X
Dt
Du
X
1. A rst-order high-pass temporal lter (fc = 20 Hz) produces a transient response whenever a contrasting border crosses the photoreceptors visual eld. This lter
enhances the contrast information while eliminating
the DC components of the photoreceptor signals. In
addition, it makes a distinction between ON and
OFF contrasting edges (i.e., dark-to-light and lightto-dark edges, respectively).
2. A higher order low-pass temporal lter (fc = 30 Hz)
attenuates any high frequency noise, as well as any interferences brought about by the articial indoor lighting
(100 Hz) used.
3. A hysteresis thresholding device/step separates ON and
OFF transitions and normalizes the signals in each
channel.
4. A time delay circuit is triggered by one channel and
stopped by the neighbouring channel. This function
measures the delay time Dt elapsing between similar
(ON or OFF) transitions occurring in two adjacent
photoreceptors.
5. A converter translates the delay Dt measured into a
monotonic function that will approximate the angular
speed XEMD. A simple inverse exponential function
makes for a relatively large dynamic range (Eq. (2)).
Fig. 2. Functional scheme of the Elementary Motion Detector (EMD) principle (adapted from [20,28,29]).
411
Fig. 4. The Angular Sensitivity Function (ASF) of two adjacent photoreceptors. The inter-receptor angle is Du = 1.05, and the acceptance angle
(Angular Width at Half Height: AWHH) is 1.65.
412
time required by the algorithm is analyzed. Also, the architecture can be optimized as far as the hardware parameters
(the number of basic logic cells) are concerned. Finally, a
suitable compromise must be made between the processing
speed and the hardware functions.
The integration stage is that involving the physical
implementation in the FPGA. Two levels of description
levels are applied here: that of the logical model (RTL
model) and that of the electrical model. The logical model
describes the architecture as a netlist of interconnected
basic logic cells after a logic synthesis. The electrical model
is a low level hardware description obtained after the placing and routing of cells in the FPGA. In the framework of
this approach, a digital simulation deals with the electrical
and timing problems caused by the physical
implementation.
At the end, a le is set up for the hardware conguration
in the FPGA, which will be used to perform tests in a physical environment. The Matlab software program was used
to study the functional approach and operating approach
models presented above. However, this software is not suitable for use in the integration stage. We therefore chose the
ISE platform of CAD Xilinx tools for this purpose.
The logic approach stage was validated using stimuli
obtained from the Matlab environment and the graphical
facilities provided by this software were used to plot the
output signals. The low level hardware descriptions were
simulated only with digital stimuli.
4. EMD implementation
4.1. Photoreceptor conguration
The system model was easy to develop because the various functional blocks of an EMD were dened twenty
years ago [28,29]. With the high level behavioural model,
the sampling frequency has to be carefully chosen because
several parameters in the EMD design, such as the digital
lter coecients and the number of possible EMD channels that can be integrated into the FPGA, will depend
on the sampling time.
In aerial robotic applications, the sampling time must
comply with the requirements imposed on the EMD so that
the Micro-Air-Vehicle (MAV) can be controlled throughout its ight envelope. The maximum sampling time
TSMAX will depend on the minimum delay Dtmin encountered by the MAVs EMDs during the fastest maneuvers
in the most critical applications. One example of a fast
maneuver is automatic terrain-following, which is performed by measuring the optic ow in the downward direction [10,11,1517]. When an eye-bearing MAV is ying in
pure translation at speed vx and height h over an unknown
terrain, the image of the terrain underneath slips at an
angular speed X that depends on both vx and h:
vx
h
n
X
i1
bi xi
n1
X
ai yi
i1
A Direct-Form II structure was used because this structure reduces the number of delay-cells and decreases the
quantization errors. Ripples on the low-pass lter temporal response were prevented by using a 4th-order Butterworth Filter, the phase of which was linearized over the
frequency range of interest. The lters require 17 coecients in all (4 coecients for the 1st-order high-pass section, 12 for the 4th-order low-pass section, and 1 for the
adjustment between the two lters). Three Direct-Form II
lters suce in fact to perform all the ltering, including
that carried out by the two cascaded 2nd order low-pass
lters.
413
A specic binary format was developed and used to prevent oset and stabilization problems. A two-complement
xed-point binary format, denoted [s, mI, mD], was dened.
The bit number of integer parts, mI, and the decimal part,
mD, were dened so as to ensure maximum accuracy and
to eliminate overow from the lter calculations. Based
on the results of a study carried out with Filter Design and
Analysis and Fixed-point Blockset of the Mathworks tools,
6 bits were selected for the integer part mI and 29 bits for
the decimal part mD. The large mD bit number is due to
the low value required to make the coecients in the lowpass lter section comply with a Bode template characterized by a low cut-o frequency at high sampling frequencies.
Other digital specications were dened as regards (i)
the bit number of the counter output giving the delay time
Dt, and (ii) the inverse exponential function giving the
angular speed XEMD. The delay time Dt is measured in
terms of a count number at a given clock period. The minimum delay to be measured determines the minimum clock
period (200 ls for fS = 5 kHz, or 400 ls for fS = 2.5 kHz).
The maximum delay to be measured is taken to be
100 ms, which is compatible with the wide range of angular speed values encountered by the MAV (Eq. (4)):
X 10/s to 5000/s, for Du = 1.05. Using a 9 bits counter at fS = 5 kHz or an 8 bits counter at fS = 2.5 kHz gives
an delay of 102.4 ms.
The measured angular speed X measured is a hyperbolic
function of Dt (Eq. (1)), but we used a function that
decreases more slowly: an inverse exponential function
with a time constant s = 30 ms. A Look-Up Table (LUT)
was used to convert the delay Dt into an output that
decreases monotonically (exponentially) with the delay
and therefore approximately reects the angular speed
XEMD (Eq. (2)). The Look-Up Table features an 8-bit input
resolution (at fS = 2.5 kHz), or a 9-bit input resolution (at
fS = 5 kHz), and a 12-bit output resolution for memorizing
the results of the conversion.
An algorithm implementation model using the Matlab
language was used to check the digital specications.
Fig. 6 presents the results of simulations carried out with
this model when the EMD, with its eld of view (FOV as
dened in Fig. 4) oriented vertically downwards, was travelling horizontally at a constant speed of 2 m/s above a
gently rising terrain (Fig. 6a) covered with a randomly contrasting texture (Fig. 6b). The nal curve (Fig. 6i) is a plot
of the estimated angular speed XEMD , which is reminiscent
of the hilly relief shown in Fig. 6a.
Fig. 7 gives a magnication of the signals from two adjacent photoreceptors and their lter outputs between distances 0.8 and 2 m. The ON dark-to-light edges and
OFF light-to-dark edges are highlighted.
4.3. Architecture
Fig. 8 shows the system architecture described by the
architecture model and designed for the processing of each
EMD. This architecture has several important features,
414
Fig. 6. Simulation results of the EMD algorithm implementation (fS = 2.5 kHz, vx = 2 m/s, h = 1 m, 36-bit xed-point binary format). (a) Shallow relief
(from 0 to 0.6 m), (b) one-dimensional ground texture consisting of randomly distributed, variously contrasting rectangles, (c) and (d) output signals from
two adjacent photoreceptors, (e) and (f) band-pass ltered outputs, (g) and (h) outputs from the hysteresis comparators, (i) relative angular speed XEMD
estimated by the EMD facing vertically downwards over the terrain shown in (a) while translating at a constant speed.
Fig. 7. Zoom on the signals from two adjacent photoreceptors (see Fig. 6c and d) and their band-pass ltered versions (see Fig. 6e and f).
415
416
Fig. 10. Photoreceptor linear array board (left) and electronic board
(right) with the 12 12 mm FPGA.
60%
32%
50%
35%
16%
37%
6%
I2 I1
I1 I2
417
Fig. 12. (a and b) Real signals from the 4th and 5th photoreceptors in the array, corresponding to the down-counter (Fig. 3), (c and d) band-pass ltered
outputs, (e, f, g and h) ON and OFF comparator outputs (e and f correspond to the ON and OFF transitions in the 4th photoreceptor; g and h
correspond to the ON and OFF transitions in the 5th photoreceptor). The interval e measured was Dt = 18.04 ms, which corresponds to the angular
speed of X = 58.2 of the moving stripe presented.
418
Fig. 13. Normalized EMD output versus delay Dt (circles). This delay is equal to the time taken by an edge to cross the visual axes of two neighbouring
photoreceptors. Dt was inferred from the speed of the analog plotter arm moving the dark pattern (Fig. 11). The continuous curve shows that there was a
good match with the inverse exponential curve.
Table 2
Minimum angular speeds measured, depending on the contrast m and the
sampling frequency fS (Illuminance = 420 Lux)
m = 18%
m = 50%
m = 80%
fS = 2.5 kHz
fS = 5 kHz
40.8/s
15.2/s
8.4/s
No detection
32.5/s
24/s
frequency, it would be possible to implement up to 245 Elementary Motion Detectors on a less-than-one-gram piece
of integrated digital electronics that requires only a few
external components.
The maximum sampling frequency of 5 kHz makes it
possible to operate in a relatively large illuminance range.
The most suitable sampling frequency was found to be
2.5 kHz when the Centronic LD12A-5T photoreceptor linear array was used with a lens with a focal length of 30 mm.
These high sampling frequencies are compatible with the
fast dynamics of Micro-Air-Vehicles.
The FPGA solution is highly versatile, as it can accommodate photoreceptor arrays of various sizes and shapes
associated with lenses of various focal lengths covering various elds of view, in much the same way as the eyes of
many arthropods are able to do.
References
[1] M. Williams, D.I. Jones, G.K. Earp, Obstacle avoidance during aerial
inspection of power lines, Aircraft Engineering and Aerospace
Technology 73 (5) (2001) 472479.
[2] V.H.L. Cheng, B. Sridhar, Technologies for automating rotorcraft
Nap-of-the-earth ight, Journal of American Helicopter Society
(1993) 7887.
[3] W. Reichardt, R. Poggio, Visual control of orientation behavior in
the y, Quarterly Reviews of Biophysics 9 (3) (1976) 311375.
[4] K. Hausen, The lobula complex of the y: structure, function and
signicance in visual behaviour, in: M.A. Ali (Ed.), Photoreception
and Vision in Invertebrates, New York, 1984, pp. 523559.
[5] N. Franceschini, Early processing of color and motion in a mosaic
visual system, Neuroscience Research (Suppl. 2) (1985) 1749.
[6] N. Franceschini, Combined optical, neuroanatomical, electrophysiological and behavioural studies on signal processing in the y
compound eye, in: C. Taddei-Ferretti (Ed.), Biocybernetics of Vision:
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
419