Air pollution MONITOR with MOBILE PHONE
An Android app called Visibility, developed by researchers at University of Southern California, lets users take a photo of the sky and get data on the air quality.
The free app is currently available for phones running Android 2.1 version of the operating system.
Visibility Monitoring using Mobile Phones
Sameera Poduri, Anoop Nimkar and Gaurav S. Sukhatme
Department of Computer Science,
University of Southern California
{sameera, nimkar, gaurav}@usc.edu
ABSTRACT and wilderness regions across the country. In 1999, the Re-
Airborne particulate matter is a serious threat to both our gional Haze Regulation was promulgated which mandates
health and the environment. It is also the primary cause improvement of atmospheric visibility.
for visibility degradation in urban metropolitan areas. We
present the design, implementation, and evaluation of an op-
tical technique to measure visibility using commodity cam-
eras and other sensors commonly found on mobile phones.
The user takes a picture of the sky which is tagged with lo-
cation, orientation, and time data and transfered to a back-
end server. Visibility is estimated by first calibrating the im-
age radiometrically and then comparing the intensity with
a physics-based model of sky luminance. We describe the
challenges for development of the system on the HTC G1
phone running the Android OS. We study the sensitivity of
the technique to error in the accelerometers and magnetome-
ters. Results from images gathered in Phoenix, Arizona and Figure 1: Los Angeles is ranked as one of the
the Los Angeles basin compare favorably to air quality data most polluted cities in the country in terms of
published by the US Environmental Protection Agency. year-round particle pollution
1. INTRODUCTION
While monitoring air visibility is important for our health
Atmospheric visibility refers to the clarity with which dis- as well as the environment, current monitoring stations are
tant objects are perceived. It is important as a measure of air very sparsely deployed (figure 2). Visibility is typically mea-
quality, driving safety, and for tourism. Without the effects sured using human observers, optical instruments such as
of manmade air pollution, the natural visual range would be photometers and transmissometers or chemical sensors such
nearly 140 miles in western USA and 90 miles in the eastern as integrating nephelometers. While the human observer
areas [1]. Today the visibility has decreased to 35-90 miles method suffers due to subjectivity, optical and chemical mea-
in the west and 15-25 miles in the east. The atmospheric pol- surement is very precise but expensive and requires mainte-
lutants that most often affect visibility exist as haze aerosols nance. In several developing countries around the world,
which are tiny particles (10µm and smaller) dispersed in air there is little or no monitoring infrastructure available.
that scatter sunlight, imparting a distinctive gray hue to the Our goal is to develop an air visibility sensing system that
sky. The suspended particles may originate as emissions uses off-the-shelf sensors and can be easily deployed to be
from natural sources (e.g., sea salt entrainment and wind- used by a large number of people. This will enable large-
blown dust) or from manmade sources (e.g., automobile ex- scale sensing of visibility and augment existing instrumen-
haust and mining activities). This particulate matter or PM is tation that is precise but expensive and sparse. We propose
cited as a key reason for heart and lung problems, especially to use phones for two reasons - 1) phones have proliferated
in metropolitan areas such as Los Angeles [4]. Recent stud- all over the world and can potentially allow massive sensing
ies also show that particulate matter enhances global warm- coverage 2) most high end phones are equipped with cam-
ing [13]. Atmospheric visibility is a measure of particulate eras and other sophisticated sensors that can be used to mea-
matter concentration [11]. The United States' Environment sure visibility and 3) having a human in the loop can help
Protection Agency (EPA) initiated the Interagency Monitor- intelligent data collection and also to gather data where it
ing of Protected Environments (IMPROVE) in 1985 with matters.
to monitor air quality and visibility in 157 national parks Our application works as follows. The user starts an ap-
1
----------------------- Page 2-----------------------
(a) (b)
Figure 2: (a) Average particulate matter concentration in California. The orange regions are above
the air quality standard stipulated by EPA. (b) Monitoring stations are sparsely deployed. The
counties in white have no monitoring stations.
plication on his phone, points the phone to the sky and takes describing related research in section 6 and conclude with a
a picture. The application tags the image with accelerom- discussion in section 7.
eter, magnetometer, date and time information and stores it
on the phone. It uses the GPS and time information to com- 2. VISIBILITY
pute current solar position, appends this to the tag file, and Visibility varies because light gets scattered and absorbed
sends it along with the image to a backend server. The solar by particles and gases in the atmosphere. According to the
and camera orientation data is used to compute an analytic EPA, particulate matter pollution is the major cause of re-
model of the sky as a function of the atmospheric visibility. duced visibility (haze) in parts of the United States. Because
By comparing this with the intensity profile of the image, particles typically scatter more uniformly than molecules for
we estimate visibility. If the user prefers, the application can all wavelengths, haze causes a whitening of the sky. The par-
also transfer his GPS coordinates to the backend server so ticles can come from many sources - industrial and vehicular
that the visibility information is displayed on a map to be emissions, volcanic eruptions, forest fires, cosmic bombard-
shared with other users. ment, the oceans, etc. A commonly used measure of atmo-
The main contributions of this work are as follows. spheric visibility is the meteorological range Rm which is
the distance under daylight conditions at which the apparent
• Design of a visibility estimation algorithm that takes
contrast between a black target and its background (horizon
as input an image of the sky, orientation of the camera
sky) becomes equal to a threshold constant of an observer,
and solar orientation
and it roughly corresponds to the distance to the most distant
• Design and implementation of the system to on HTC discernible geographic feature [11]. Koschmieder derived a
G1 phones taking into account privacy and efficiency formula that relates the meteorological range to the aerosol
factors extinction coefficient ß.
3912
• Evaluation of the system using images from 3 different Rm =
ß (Mm-1)
sources and and analysis of effect of sensor noise
The formula shows that visibility closely correlates with
The paper is organized as follows. The next section de- aerosol load and it is therefore a good indicator for the air
fines metrics for atmospheric visibility and gives an overview quality.
of the common methods of measuring it and its relation to In atmospheric sciences literature, another metric called
air quality. Section 3 presents the architecture and design of turbidity is used. Turbidity is a measure of the fraction of
the system and its implementation on HTC G1 smartphone. light scattering due to haze as opposed to molecules [11].
The sky luminance model, radiometric calibration technique Formally, turbidity T is defined as the ratio of the optical
and visibility estimation algorithm are described in section 4. thickness of a path in the haze atmosphere (haze particles
This is followed by experimental results including sensitiv- and molecules) to the optical thickness of the path in atmo-
ity analysis in section 5.1. We place the work in context by sphere with the molecules alone:
2
----------------------- Page 3-----------------------
and images. To ensure user privacy, we process the
data such that the user is anonymous with respect to the
information that leaves the phone. User's GPS coordi-
nates are used to compute solar position on the phone
which is communicated to the backend server with-
out the GPS data. Similarly, the image is cropped and
only the segment that contains the sky pixels is shared.
However, the user has the option of sharing complete
information which can be useful for large-scale moni-
toring and analysis.
• Communication cost: In the default usage mode,
the data transfer is initiated immediately after data log-
ging assuming that the phone is connected to the inter-
net. However, we provide an option to delay the trans-
Figure 3: Relation between turbidity and mete- fer until a preferred internet connection is available.
orological range in km
• No blocking: After data is transfered, the response
from the server may take a few minutes. During this
time, the visibility application does not block the phone.
tm + th Instead, it switches into a background mode where it
T =
tm waits for a response and frees the phone for other ap-
plications. On receiving a response it displays it as a
where tm is the vertical optical thickness of the molecular notification.
atmosphere, and th is the vertical optical thickness of the
haze atmosphere. Optical thickness t for a path of length x • Aiding data collection: Sensors on the phone can
is defined in terms of the extinction coefficient ß as follows guide data collection so that the data is well-suited for
the estimation algorithm. We use the phone's orien-
t = x ß (x)dx tation sensors to help the user hold the phone parallel
to the ground (without roll) as that yields best results.
0
Similarly, we deactivate the camera button if the zenith
Strictly speaking, visibility is only defined for a path and angle is more that 100? as we are only interested in im-
not for a region. But if the region is homogenous, we can de- ages of the sky.
fine its visibility as that of a random path. Generally horizon-
tal paths are considered homogenous and vertical paths are • Human in the loop: Several computer vision prob-
least homogenous. In fact, the aerosol concentration rapidly lems that are extremely challenging to automate are
decreases in the vertical direction. Most of the aerosol par- trivially solved by a human. In our system, segmenting
ticles exist in the region 10 - 20 km about the surface of sky pixels in an arbitrary image is one such problem.
earth. Turbidity and meteorological range are closely related When the use captures an image, we ask him to select
as shown in figure 3 a part of the image that is sky. By exploiting the partic-
In our work, we estimate turbidity directly using models ipatory nature of our paradigm, we can build a system
of sky appearance. that is robust and efficient.
• Energy e?ciency: The sensors that consume most
3. SYSTEM DESIGN
energy on our system are the GPS and camera. Of
In this section, we present the design and implementation these, the GPS is not essential because very coarse,
of the visibility sensing system. Figure 4 shows the high city-scale, localization is sufficient to for visibility es-
level architecture. The user gathers a picture that gets auto- timation. The user can turn off GPS in the application
matically tagged with relevant sensor data and transmitted to making it use either cell tower ID based locations or
a backend server which processes the data to estimate visi- the last recorded GPS locations.
bility and returns a value of turbidity to the user. Based on
the privacy preference of the user, the image, its location and We will now describe the details of our implementation
turbidity value are displayed on a publicly accessible map on the HTC G1 smartphone.
and stored onto a database for future analysis. 3.1 Hardware
Our design is based on the following considerations.
Current high-end mobile phones (such as the iPhone, HTC
• Privacy: The visibility estimation algorithm is based G1, Nokia N97, BlackBerry, etc) are embedded with cam-
on potentially sensitive information such as location eras, accelerometers, GPS sensors and in some cases, even
3
----------------------- Page 4-----------------------
Figure 4: Overview of the system
magnetometers. We chose the HTC G1 phone because it can Such information will be useful for further study of visibil-
sense 3D orientation and is easily programmable. ity conditions. As mentioned earlier, the file transfer option
HTC G1 runs Android OS that provides an open SDK and controls whether sensor data will be automatically transfered
APIs to program sensors. Android is a software stack for to the server or stored on the phone and transfered later when
different mobile devices that includes the operating system, the user clicks the 'Transfer files' button. This facility is use-
middleware and application level. Android runs linux kernel ful if either the phone does not have 3G capability or if the
and applications are developed in Java to support standard user prefers to wait for a cheaper WiFi connection. Turning
Java APIs as well as libraries for Android specific APIs. on the privacy filter will prevent GPS data being transfered to
The phone has significant computational capability with a the server. If it is off, the GPS data is used to archive the sen-
528Mhz processor, 64 MB internal RAM and 128 MB in- sor data and display it on a publicly accessible map. When
ternal ROM for OS and applications. With this processing the user clicks the start button, the image capture screen with
power, our on-phone computation of solar position is al- the camera preview appears. On this screen, the azimuth and
most instantaneous. The phone has a 1GB MicroSD card ?
zenith angles are shown in green if the roll is less than 5 ,
where the images taken from camera along with their tags i.e., the phone is parallel to the ground. If not, the angles
are stored. appear in red. If the angles are in green, the user can cap-
The phone is embedded with a 3.1 megapixel camera with ture an image by pressing the camera button. The image is
a dedicated camera button. It supports JPG, BMP, PNG, and saved and displayed on the screen and the user is prompted
GIF formats. We use RGB format to store color information. to choose two points on the image such that the rectangle
The camera does not allow optical zooming which means formed with those points as a diagonal contains only sky pix-
that all images from a phone have a fixed focal length, thus els. The image is cropped to this box and sent to the server
allowing a one-time calibration. The Android orientation along with a separate tag file containing orientation, date,
API combines information from a 3-axis magnetic sensor time, and solar position which is computed on the phone
and a 3-axis accelerometer to report the 3D orientation of (as described in the next subsection). Files are transfered
the phone. It uses a dynamic offset estimation algorithm to over standard FTP protocol in which client program runs on
compensate for the local magnetic field. We found that in phone and FTP server process runs on the backend server.
spite of this, there is a significant error in the azimuth val- While the orientation data is stored at the time of image cap-
ues (figure 16). The GPS data is highly accurate (-160 dBm ture, the GPS sensor is invoked as soon as the application
tracking sensitivity) for our purpose. starts since it can take a few seconds to locate satellites. Af-
The backend system consists of FTP and HTTP servers ter segmenting the sky portion of the image, the user can
running on a standard desktop that runs MATLAB and com- also provide additional information about cloud cover and
municates with the phone through internet. the apparent visibility. This information will be useful in
studying the performance of the system. After this step, the
3.2 Software Design application switches to a background mode where it listens
for a message from the backend server. The resulting turbid-
Figure 5 shows screenshots of the phone application built ity estimate is displayed as a notification along with time of
on Android 1.5SDK. It begins with a splash screen with op- image capture.
tions to edit settings or capture an image. On the settings Figure 6 shows the flow of information and the compu-
screen the user can choose his internet tagging, privacy and tations performed on the phone and the backend server. On
file transfer settings. Turning on the internet tagging op- the phone end, we use location and time information to com-
tion causes the application to tag the image with additional pute the solar azimuth and elevation angles using the algo-
data such as weather information obtained from the internet.
4
----------------------- Page 5-----------------------
(a) (b) (c)
(d) (e) (f)
Figure 5: The Visibility application for Android OS. (a) startup screen with options to view/edit
settings and start taking the picture. (b) settings for communication and privacy. If internet tagging
is turned on, the application will gather weather data. File transfer option allows the user to choose
between transferring the image immediately or at a later more convenient time. Privacy filter controls
whether the user's GPS data is communicated. (c) Camera preview. The azimuth and zenith angles
are displayed in green when the roll is < 5? and red (d) otherwise. (e) The user chooses a portion of
sky for processing. Clicking the camera button at this point stores the image along with orientation
data. (f) The computed turbidity value is returned as a notification
Figure 6: System Architecture
5
----------------------- Page 6-----------------------
rithm described in [16]. The 3D orientation computation is the sun. We call f the scaled luminance as it captures the
performed by Android's orientation API. ratio of true luminance to zenith luminance. It is defined as
The backend server runs Perl scripts that are triggered by follows.
the phone through HTTP requests. These scripts initiate im-
age processing which includes radiometric correction and f (? , ? , t) = (1+a·e(b/ cos(?p )))(1+c·e(d·cos(?p )+e·cos2 (?p ))
computation of image luminance. This is implemented us- p p
(2)
ing MATLAB. The camera orientation and solar orientation
where a, b, c, d, and e are adjustable coefficients. Each of
data are used to compute the analytical model of sky lumi-
the parameters has a specific physical effect on the sky dis-
nance. This is followed by an optimization step to estimate
tribution and with different values they can capture a wide
the visibility. The analytical model also uses the focal length
variety of sky conditions. An empirical model of the param-
information which can usually be obtained from the phone.
eters in terms of turbidity t has been proposed [14]
The resulting visibility value is sent back to the phone as a
response to the HTTP request. If the privacy filter is off, then
the image along with visibility value is sent to a web server ? a ? ? 0.1787 1.4630 ?
? ? ? ?
b -0.3554 0.4275
? ? ? ?
which displays them on a map (figure 7). t
? c ?= ? -0.0227 5.3251 ?
? ? ? ?
1
? ? ? ?
d 0.1206 -2.5771
e -0.0670 0.3703
Figure 8 shows the variation in the scaled luminance ratio
(f) as the value of turbidity changes. There is a significant
change in the shape of the surface and this can be used to
estimate turbidity.
Equation 2 can be expressed in terms of the pixel coordi-
nates as shown in [9]. We reproduce the equations here for
clarity.
v sin(? ) + f cos(? )
p c c c
? = arccos
p
f2 + u2 + v2
c p p
Figure 7: Images and the visibility estimates are fp = arctan fc sin fc sin ?c - up cos fc - vp sin fc cos ?c
displayed on a publicly accessible map. fc cos fc sin ?c + up sin fc - vp cos fc cos ?c
? = arccos cos(? ) cos(? )+sin(? ) sin(? ) cos(f -f )
p s p s p p s
4. VISIBILITY ESTIMATION
up and vp are the pixel coordinates of a point p and fc is
In this section, we describe the details of the analytical the focal length of the camera. ?p and fp are the correspond-
sky model, radiometric calibration and visibility estimation ing zenith and azimuth angles and ?p is the relative orienta-
algorithm. The method is based on [9] where the sky lumi- tion to solar position. By substituting for these and the pa-
nance model is used to calibrate the geometric parameters of rameters a, b, c, d, e in equation 2 we rewrite f in terms of
the camera assuming clear sky images. In our case, the focal the variables of interest to our problem, as g. We have,
length and orientation of camera are known and we seek to
estimate the turbidity of images.
g(? , f , u , v , t)
L(? , ? ) = L c c p p (3)
4.1 Sky Luminance p p 0 g(0, f ,0,0, t)
s
Several physical as well as empirical models for the sky 4.2 Radiometric Calibration
luminance have been proposed [8]. Of these, the model pro-
Digital cameras are not designed to capture the entire radi-
posed by Perez et al [12] is shown to work well for different
ance range in the real world. The camera applies a non-linear
weather conditions. It is a generalization of the CIE standard
mapping called the response function to the scene radiance
clear sky formula. The luminance of a sky element is given
to generate a smaller, fixed, range of image intensity values.
by
In order to measure scene radiance from image intensities,
it is necessary to learn the inverse response function. This
f (? , ? , t)
L(? , ? ) = L p p (1) is called radiometric calibration. A common approach is to
p p 0 f (0, ? )
s take a series of images with varying camera exposures set-
where ?p is the zenith of the sky element, ?p is the angle tings and estimate the inverse response function. This is not
between the sky element and the sun, and ?s is the zenith of feasible for our application because we cannot control the
6
----------------------- Page 7-----------------------
fc = 15? fc = 60? fc = 180?
t = 2
t = 10
t = 20
t=30
Figure 8: Illustration of the Perez model for sky luminance. The luminance ratio (f) changes signif-
?
icantly with turbidity. In the above graphs fc = 2031 (based on HTC G1 phone camera), ?c = 45 ,
?s = 30? and fs = 0. Note that the scale of the z axis is di?erent for di?erent graphs. In general, the
luminance ratio increases as the turbidity increases.
7
----------------------- Page 8-----------------------
(a) (b) (c)
Figure 9: (a) image (b) corners used for estimation (c) resulting response functions for red, blue,
green channels
exposure settings on most phones. We use the technique pro- The above optimization can be solved using standard tech-
posed by Lin et al [10] that uses a single image as suggested niques such as Levenberg-Marquardt. We use initial values
in [9]. of 2 for t and 0.5 for k and bounds of [0 40] for t and [0 1]
The method works by first extracting color edges in an im- for k.
age and finding an inverse response function that results in In our experiments, we observed that the compass on the
an almost linear color blending in RGB space. This is done phone gives a significant error. Therefore, we modify the
using maximum a posteriori estimation where the proba- optimization to estimate compass offset ?f .
c
bility of an inverse response curve is proportional to the dis-
tance in RGB space between edge colors and prior model is 2
t = argmin I - kg(? , f + ?f , u , v , t)
obtained from [6]. Both the image intensity and irradiance t,k,?fc p c c c p p
values are normalized to [0 1]. Figure 9 shows the result- p
ing curves for red, green and blue channels for the HTC G1 (4)
phone obtained by calibrating over 10 different images and
3 different phones. 5. EXPERIMENTS
This section describes a series of experiments conducted
4.3 Visibility Estimation to validate our system and study its sensitivity to error in
Visibility is estimated by matching the scaled luminance sensor data. We conducted two sets of experiments. The
ratio (f) with the observed image intensity values at sky pix- first on static camera setups used to monitor weather and
els after the radiometric calibration. Intensity I is computed visibility conditions. These images do not have significant
from RGB values using the CIE standard formula orientation error and allow us to study the system on a series
I = 0.2126R + 0.7152G + 0.0722B. We find the value of t of images of the same scene. The second set of images are
that minimizes the sum of squared error between measured taken from the HTC G1 mobile phone using the visibility
intensity and the analytic luminance value over the set P of application described above. In all these experiments, the
sky pixels. complete image was logged and the sky part was segmented
interactively at the backend during processing.
2
(t, k) = argmin I - kg(? , f , u , v , t) 5.1 Static camera image sources
t,k p c c p p
p?P
South Mountain Web Camera
The scaled luminance ratio g is computed using the solar
position, camera orientation, and focal length data reported The Arizona Department of Environmental Quality (ADEQ)
by the phone. Note that we cannot recover true irradiance maintains several cameras near Phoenix. We used images
values from the image but only scaled values. The Perez from a 2.4 megapixel camera located in the North Mountains
model for g also captures the luminance ratio with respect to looking south. The pictures from this camera are published
the zenith. Therefore there is a constant factor k between the every 15 minutes. We used the method in [9] to calibrate
image intensity I and f at each pixel which can be estimated. the focal length and the azimuth and zenith angles of this
8
----------------------- Page 9-----------------------
camera. ity in the atmosphere caused by uneven clouds of smoke.
The images shown here are facing the Los Angeles National
Altadena Weather Station Forest where the fires took place. Because the surfaces do
This 5 megapixel camera located in Altadena in California not match properly, increasing t from 5 to around 17 also
looks north-northeast towards the San Gabriel valley. It takes has very little impact on error.
a picture every 5 minutes that is published online. Again, we
5.3 Sensitivity analysis of camera orientation
use [9] to calibrate the camera.
error
USC Rooftop Camera Station We conducted a series of experiments to study the error in
Recently, we set up an image station on the roof of a build- the HTC G1 phone's orientation sensing and its impact on
ing at the University of Southern California (figure 11). It estimated luminance values. To compute the error, we cap-
has an android phone placed inside a weather-proof box. It tured images of the sun using our visibility application and
looks northeast at the Los Angeles downtown and logs im- calculated the solar position in the image using the time of
ages every 15 minutes. The focal length of the camera was day, GPS, and camera orientation reported by the phone. We
estimated using the MATLAB calibration toolbox [3]. then compared these pixel coordinates with visually detected
solar position in each image (figure 16(a)). The resulting er-
Results ror over 100 images taken using 8 different phones is shown
in figure 16(b) and (c) for azimuth and zenith angles. While
Figure 10 shows three examples of images from the South ?
the zenith error is mostly within ±5 , the azimuth error is
Mountain camera that are reported to have good, fair and
significantly larger.
poor visibility by the ADEQ. The corresponding image in-
We analyzed the impact of error on the azimuth angle on
tensity surfaces and the scaled luminance profiles for the val-
the intensity ratio. Figure 15 shows the derivative of lumi-
ues of turbidity (2,3 and 6) estimated by our algorithm are
nance ratio f (? , ? ) with respect to camera azimuth f for
shown. The turbidity values increase as the visibility de- p p c
different values of solar zenith. The solar azimuth, f , is
creases and the luminance surfaces match well with the im- s
fixed at 0? and the camera zenith, ? , is fixed at 45?. Repeat-
age intensity profiles. c
ing the computation for different values of fs and ?c pro-
For the camera in Altadena, there is no ground truth vis-
duces similar graphs. The graph shows that for fc between
ibility data available. Therefore, instead of comparing visi- ? ?
100 and 160 , the luminance ratio varies only slightly with
bility values, we look at the average trend in visibility during
f .
a day (figure 12) averaged over 50 days during March 2009 c
to July 2009 (we eliminated days that had cloudy skies). The
turbidity values are highest around 9 am and then steadily
decrease to reach a minimum around 4 pm. This is in fact
a well known trend in particulate matter concentration. As
an example, we plot the PM2.5 concentration in central Los
Angeles over the same 50 days
5.2 HTC G1 Phone
The HTC G1 phone was used to gather data using our ap-
plication in the Los Angeles basin. The data was gathered
over a period of 3 months. We focus on 3 significant visibil-
ity events during this time when the Air Quality Index (AQI)
published by the South Coast Air Quality Management Dis-
trict (AQMD) had extreme values. 1) fire day: the Los
Angeles wildfires around August 26th 2009 when the AQI
was 150 and labeled 'unhealthy' 2) hazy day: an extreme
low air quality day on November 8th 2009 when the AQI
was 120 and labeled 'unhealthy for sensitive groups' and 3) Figure 15: Sensitivity of luminance to compass
clear day: an extreme clear air quality day following rain error
and snow on December 8th 2009 when the AQI was 23 and
labeled 'good'. Figure 13 shows two representative images
5.4 Localization
for each of the clear day and fire day and the luminance pro-
files for turbidity values of 2, 10, and 11. We also show the The GPS data is used to calculate the positions of sun.
variation in error for different values of t. For the fire day, However, the position of sun changes very slowly with lo-
the intensity surface has a different shape compared to the ?
cation. For instance, in Los Angeles (latitude 34 N and
analytic model. We believe this is because of inhomogene- ?
longitude 118 W the solar azimuth changes by less than a
9
----------------------- Page 10-----------------------
Good visibility, t = 2
Fair visibility, t = 3
Poor visibility, t = 6
(a) Image (b) Intensity (c) model luminance
Figure 10: Results from the South Mountain Camera
(a) (b) (c)
Figure 11: USC Rooftop Camera Station
10
----------------------- Page 11-----------------------
(a) (b)
Figure 12: (a) Scatter plot of pollution data during a single day averaged over 50 days in central Los
Angeles. (b) Average visibility data estimated using images over 50 days from a camera in Altadena
degree for over 60 miles. Therefore, a city-level localiza- Sky model based camera calibration: Models of
tion can compute the solar position accurately enough. The sky appearance have been studied for several years and the
sensitivity of luminance to solar azimuth is exactly the same computer graphics community has used these models for re-
as camera azimuth because the model only uses relative az- alistic rendering for scenes. In the past few years, these mod-
imuth angle. els have been used to analyze images to study photometric
properties of scenes [18]. Recently, Lalonde et al, have cali-
6. RELATED WORK brated the focal lengths and orientations or a large number of
In this section we review current research in three areas web cameras using images available on the internet [9]. We
related to our work. use the same approach to estimate visibility. The key differ-
ence in our case is that unlike web cameras, the camera is
Image processing based visibility mapping: There
our case can be controlled and its geometric parameters are
is a growing interest in monitoring environmental conditions
known.
using commodity cameras. Several approaches have been
proposed to compute visibility from images. In [17], visibil-
7. CONCLUSION
ity is computed as the ratio of pixel contrast just below and
just above the horizon. The method relies on being able to Air quality is a serious global concern that affects our
detect the horizon accurately which is challenging especially health as well as the environment. While several efforts are
in poor visibility conditions. It is well suited for static web underway to monitor and disseminate air quality informa-
cameras where the horizon does not change and therefore tion, the sensing locations are still extremely sparse because
can be computed on a clear day. Another approach is to use the monitoring stations are expensive and require careful in-
Fourier analysis of the image since clear images are likely stallation and maintenance. Our vision is to augment these
to have more high frequency components ??. This approach precise but sparse air quality measurements with coarse, large-
applies to cases where the objects in the image are at a simi- scale sensing using commodity sensors. To this end, we
lar distance away as the visibility range. Other methods such propose a system that uses cameras and other sensors com-
as [2] and [15] require detection of a large number of vi- monly available on phones to estimate air visibility. Using
sual targets either manually or automatically. All the above accelerometer and magnetometer data along with coarse lo-
methods are based on image processing alone. In contrast, cation information, we generate an analytic model for sky
the approach we propose is based on a analytic model of sky appearance as a function of visibility. By comparing this
appearance that takes into account the camera's orientation model with an image taken using the phone, visibility is es-
and solar position. timated. We present the design, implementation and evalu-
Air quality mapping using phones: Owing to the ation of the system on the HTC G1 phone running Android
ubiquity of mobile phones, researchers have proposed inte- OS with a backend server. To ensure user privacy, GPS data
grating chemical sensors with mobile phones to measure air is processed on the phone and the image is cropped to con-
quality [5, 7] thus allowing large-scale sensing. Our work tain only sky pixels before sharing it with the server. Our re-
is in the same spirit of participatory sensing but we seek sults show that the system can reliably distinguish between
to measure air quality using sensors commonly available on clear and hazy days.
phones. While our initial results are promising, several challenges
11
----------------------- Page 12-----------------------
Clear day, t = 2, AQI 23
Clear day, t = 2, AQI 23
LA wildfires day, t = 11, AQI 150
LA wildfires day, t = 10, AQI 150
Figure 13: Comparison of images taken during wildfires in Los Angeles on 30th August 2009 with
those taken on a very clear day on 8th December 2009
12
----------------------- Page 13-----------------------
Clear day, t = 2, AQI 23
Hazy day, t = 9, AQI 120
Figure 14: Comparison of images taken during a hazy day on 8th November 2009 with those taken
on a very clear day on 8th December 2009. In this case, the pictures were taken at the same time of
the day 3pm
(a) (b)
Figure 16: Compass error in HTC G1 phones
13
----------------------- Page 14-----------------------
exist. First, the model assumes that the atmosphere is ho- In submission, International Journal on
mogenous. This is generally true in the horizontal direction Computer Vision, 2009.
but when looking at an angle the haze tends to be layered. [10] S. Lin, J. Gu, S. Yamazaki, and H. Shum. Radiometric
The impact of this is clear in the images we took during wild- calibration from a single image. In Computer Vision
fires in Los Angeles. To address this we plan to investigate and Pattern Recognition, 2004. CVPR 2004.
further sky models and also use the sensors on the phone to Proceedings of the 2004 IEEE Computer Society
guide the user to gather data at favorable angles. Second, we Conference on, volume 2, 2004.
currently assume that the after the user crops the image, it [11] E. McCartney. Optics of the Atmosphere:
does not have any clouds. While cloudy skies are a funda- Scattering by molecules and particles. John Wiley,
mental limitation of this approach, we plan to develop seg- New York, 1976.
mentation techniques that will allow us to use disconnected [12] R. Perez, R. Seals, and J. Michalsky. All-weather
cloud-free sky segments. model for sky luminance distribution. Preliminary
configuration and validation. Solar Energy,
Acknowledgements 50(3):235–245, 1993.
We thank Bill Westphal and the Arizona Department of En- [13] K. Prather. Our Current Understanding of the Impact
vironmental Quality for sharing the Altadena web camera of Aerosols on Climate Change. ChemSusChem,
data and the Phoenix South Mountain camera data respec- 2(5), 2009.
tively. [14] A. Preetham, P. Shirley, and B. Smits. A practical
analytic model for daylight. In Proceedings of the
8. REFERENCES 26th annual conference on Computer graphics and
interactive techniques, pages 91–100. ACM
[1] U. E. P. Agency. Visibility in mandatory federal class I Press/Addison-Wesley Publishing Co. New York, NY,
areas, 1994-1998 a report to congress, 2001. USA, 1999.
[2] D. Baumer, S. Versick, and B. Vogel. Determination of [15] D. Raina, N. Parks, W. Li, R. Gray, and S. Dattner. An
the visibility using a digital panorama camera. Innovative Methodology for Analyzing Digital
Atmospheric Environment, 42(11):2593–2602,
Visibility Images in an Urban Environment. Journal
2008.
of the Air & Waste Management Association,
[3] J.-Y. Bouguet. Camera calibration toolbox for matlab, 55(11):1733–1742, 2005.
2008.
[16] I. Reda and A. Andreas. Solar position algorithm for
[4] D. Dockery, C. Pope, X. Xu, J. Spengler, J. Ware, solar radiation applications. Solar Energy,
M. Fay, B. Ferris, and F. Speizer. An association 76(5):577–589, 2004.
between air pollution and mortality in six US cities. [17] L. Xie, A. Chiu, and S. Newsam. Estimating
The New England journal of medicine,
Atmospheric Visibility Using General-Purpose
329(24):1753, 1993.
Cameras. In Proceedings of the 4th International
[5] P. Dutta, P. Aoki, N. Kumar, A. Mainwaring, Symposium on Advances in Visual Computing,
C. Myers, W. Willett, and A. Woodruff. Common Part II, page 367. Springer, 2008.
Sense: participatory urban sensing using a network of [18] Y. Yu and J. Malik. Recovering photometric properties
handheld air quality monitors. In Proceedings of the of architectural scenes from photographs. In
7th ACM Conference on Embedded Networked SIGGRAPH '98: Proceedings of the 25th annual
Sensor Systems, pages 349–350. ACM, 2009. conference on Computer graphics and interactive
[6] M. Grossberg and S. Nayar. What is the space of techniques, pages 207–217, New York, NY, USA,
camera response functions? In IEEE Computer 1998. ACM.
Society conference on Computer Vision and
Pattern Recognition (CVPR), volume 2. Citeseer,
2003.
[7] R. Honicky, E. Brewer, E. Paulos, and R. White.
N-smarts: networked suite of mobile atmospheric
real-time sensors. In Proceedings of the second
ACM SIGCOMM workshop on Networked systems
for developing regions, pages 25–30. ACM, 2008.
[8] P. Ineichen, B. Molineaux, and R. Perez. Sky
luminance data validation: comparison of seven
models with four data banks. Solar Energy,
52(4):337–346, 1994.
[9] J.-F. Lalonde, S. G. Narasimhan, and A. A. Efros.
What do the sun and the sky tell us about the camera?
14
Earth's magnetic field:
2 Comments:
That's actually amazing and I am glad that this is a nice and interesting feature added.There is a need to get conscious about the environment and health facts and act upon them.
With bagless vacuums, it’s good practice to empty the dirt bin after every use. Many machines have a max line that serves as a helpful reminder, if you can’t make emptying the bin part of your vacuuming routine.
Post a Comment
<< Home