a survey on algorithms of shadow removal in vehicle detection · a survey on algorithms of shadow...
TRANSCRIPT
A Survey on Algorithms of Shadow Removal in
Vehicle Detection Padmavathi.S1, Abylash Kumar.KT2
1,2Department of Computer Science, Amrita School of Engineering, Coimbatore.
Abstract
A shadow appears on an area when the light from a source
cannot reach the area due to obstruction by an object. Shadow
detection and removal in real time images is always a challenging
task. In contrast with the rapidly expanding and continuous
interests on this area, Shadow detection is critical for robust and
reliable vision-based systems for traffic flow analysis. In our case
Vehicle Detection, Shadow Removal is used to prevent moving
shadows being misclassified as moving objects (or parts of them),
thus avoiding the merging of two or more objects into one. The
environment considered is Local Highway with multiple lanes
and a fixed camera. This paper is aimed at giving comprehensive
algorithms to detect and remove vehicle shadows in Real-time
Videos.
Terms used – Shadow Detection, cast shadow, Shadow
Detection, Shadow Removal
I. INTRODUCTION
Many Computer applications involving video segments
require detecting and tracking moving objects. For Vehicle
detection and speed estimation, Vehicles are to be detected for
which various techniques are explained in the paper [19]. We
identify shadow removal as a critical step for improving object
detection and tracking. Shadows can be classified into self-
shadows and cast shadows. Neither motion segmentation nor
change detection methods can distinguish between moving
objects and moving shadows. Our job is to detect the cast
shadow and not the self-shadow because removing the later
will result in parts of the object removed. A frame may contain
a vehicle with or without shadow or multiple vehicles
connected by shadows. The key success for the shadow
removal is achieving the robustness and efficiency of the
images or sequences that are been used for progress. Proper
segmentation is required for the correct isolation of objects and
shadows in the foreground regions. The previous works on
shadow detection are reviewed in [1]. Many of these methods
involve using background subtraction technique. Results vary
depending on the algorithms used. Tracking of dynamic objects
can be taken by image frames extraction from the video
sequences and shadow detection can work at pixel level. Fact is
that illumination changes based on Sunlight and Street lights at
night. First step computed for shadow detection is the
difference between the current frame and a reference image.
Second, the assumption is used to distinguish between shadow
and object in changed image regions. Third, shadows smooth
the background they cover. Many of the works published
locally exploit pixel appearance change due to cast shadows. A
possible approach is to compute the ratio between the
luminance of pixels after and before shadow appears. E.
Salvador et al proposed an approach to detect and classify
shadows for still images [2]. They exploit invariant color
features to classify cast and self- shadows.
A. Prati et al [3] conducted a survey on detecting moving
shadows; algorithms dealing with shadows are classified in two
divisions as Deterministic approach and Statistical approach.
The first classification considers whether the decision process
introduces and exploits uncertainty. Deterministic approaches
use an on/off decision process, whereas statistical approaches
use probabilistic functions. As the parameter selection is a
crucial problem for statistical methods, the authors further
divided statistical methods into parametric and non-parametric
methods. For deterministic approaches, algorithms are
classified by whether or not the decision can be supported by
model-based knowledge. The authors reviewed four
representative methods for the categories of his taxonomy and
argued that Deterministic Model-based methods [4] rely so
much on models of the scene that they inevitably become too
complex and time-consuming. The survey of A. Prati et al
mainly focuses on the moving shadow detection, typically they
concentrate the attention on umbra, considering the penumbra
as a particular case of umbra. It is because the distance between
the objects and the background is negligible compared to the
distance of illumination sources to the objects in a highway
scene and most or all of the shadows are umbra or strong
shadow.
T. Horprasert et al’s method [5] is a statistical
nonparametric approach and the authors denote it with symbol
SNP. This approach uses color information and a trained
classify to distinguish between object and shadows. I. Mikic et
al [6] proposed a statistical parametric approach (SP) and
utilized both spatial and local features, which improved the
detection performance by imposing spatial constraints. R.
Cucchiara et al’s method (DNM1) [7] and J. Stauder et al’s
work (DNM2) [8] were representatives of deterministic non-
model based method. DNM1 is based on an assumption that
shadows in image do not change the hue of surfaces. The
reason why the author reviewed DNM2 is that it is the only
work that handles the penumbra regions in image.
In section II deals with the previous works on shadow
detection algorithms. By section III a survey on effective
algorithms used in shadow detection and concludes with
section IV with the conclusion of the Paper along with the
References...
Abylash Kumar KT et al, Int.J.Computer Technology & Applications,Vol 5 (2),518-521
IJCTA | March-April 2014 Available [email protected]
518
ISSN:2229-6093
II. SHADOW DETECTION
Most of the shadow removal algorithms use the background
subtraction and temporal differences in case of using stationary
cameras. Yang et al. [9] described the ratio of intensities
between a shaded pixel and its neighboring shaded pixel in the
current image, and this intensity ratio was found to be close to
that in the background image. They also made use of the slight
change of intensities on the normalized R and G channels
between the current and background image. Hsieh et al. [10]
used the histograms of vehicles and the calculated center of
lane to detect the lane markings, and also developed a
horizontal and vertical line-based method to remove shadows
by characteristics of those lane markings. This method might
become ineffective in case of no lane markings. Song et al.
[11] applied Gaussian model to representing the constant RGB-
color ratios, and determined whether a moving pixel belonged
to the shadow or foreground-object by setting ±1.5 standard
deviation as a threshold. Mohammed et al. [12] proposed their
method by using division image analysis and projection
histogram analysis. Image division operation was performed on
the current and reference frames to highlight the homogeneous
property of shadows. They afterwards eliminated the left pixels
on the boundaries of shadows by using both column- and row-
projection histogram analyses. Geometric model attempted to
remove shadowed regions or the shadowing effect by
observing the geometric information of objects.
III. EFFECTIVE ALGORITHMS
An algorithm is said to be effective if it uses least resources
such as Processing time, power consumption with better
results.
A. Shadow removal from Single image:
1) Vague Shadow estimation
Vague shadow refers to the shadows or shadings in images
which have no clear boundaries and gradually changed
intensity. In real scenes, self-shadows and shadings usually
have smooth changes and can be classified as vague shadow.
However, even in the cast shadow regions, we can separate the
gradually changed components. For instance, the penumbra
regions of the cast shadows are more likely to be vague shadow
components. Unlike the Land’s work on illumination removal
[13], our approach works in the gradient space rather than
intensity space. It is mainly because that it is easier to combine
the vague and hard shadow removal in the gradient domain.
Therefore, we first transfer the raw logarithmic image into
gradient domain:
( ) || ( )||)) Eq. (1)
Where (x,y) represents the logarithmic image, k denotes
the color channel: k = 1,2,3 and is the gradient
response. Since vague shadows are expected to be spatially
smooth, we clip out the high derivative peaks corresponding to
the object boundaries and obtain the vague shadow mask. To
determine the clipping threshold, the following criteria is
applied:
T= * * ( )++ Eq.(2)
Where is a predetermined value and T represents a
threshold. After deriving threshold T, we could do the
following clipping operation to obtain a binary vague shadow
mask for further process:
( ) { ( ( ))
Eq.(3)
The VS(x,y) is the binary mask. It could either be applied
directly to the derivatives to remove the vague shadows in
image or be combined with the hard shadow mask to perform
the full shadow removal task.
B. SAKBOT and ATON:
a) SAKBOT
SAKBOT (Statistical And Knowledge-Based Object Tracker)
is a deterministic non-model based approach and it aims to be
a general purpose object detector and tracker. It performs
object detection by means of Background Suppression.
Sakbot first converts pixel information from the RGB color space to the HSV color space. HSV color space corresponds closely to the human perception of color [14] And it has revealed more accuracy to distinguish shadows. Then, it tries to estimate how the occlusion due to shadow changes the value of H, S and V. The rationale is that cast shadow's occlusion darkens the background pixel and saturate its color. The first assumption is well evident. The second assumption helps distinguish the object points from the shadow points. Another interesting point is that shadows often lower the saturation of the points. Sakbot exploits these assumptions to define the shadow point mask S as follows,
{
( )
( )
Eq. (4)
b) ATON
The moving shadow detection algorithm developed
for the ATON project is described in detail in [15]. It is based
on the detection of umbra of moving cast shadow in outdoor
traffic video scene. It deals with non-modeled environments
and it takes into account most of the assumptions described in
Section II. Its key novelty is the use of three sources of
information to help in detecting shadows and objects:
Local – based on appearance of the individual
pixel.
Spatial – information based on the assumption that
objects and shadows are compact regions in the
scene.
Temporal – information that states that object and
shadow position can be predicted from previous
frames.
The local information is exploited assuming that if , - is the value of the pixel not shadowed, a shadow
Abylash Kumar KT et al, Int.J.Computer Technology & Applications,Vol 5 (2),518-521
IJCTA | March-April 2014 Available [email protected]
519
ISSN:2229-6093
changes the color appearance by means of an approximated linear transformation , where
( ) ( ) is a diagonal
matrix obtained by experimental evaluation. The
corresponding values under shadows are
and
, with i R,G,B. A pixel is classified maximizing
the probability of the class membership.
(
. ⁄ / ( )
∑ . ⁄ / ( ) Eq.(5)
ATON improves the detection performance using spatial
constraints. Experimental results are shown in Table 1
TABLE 1
FN FP SHADOW
ACCURAC
Y
OVERALL
ACCURAC
Y
SAKBOT
W/O
COLOUR
150
8
319
0
60.59
91.65
SAKBOT
WITH
COLOUR
180
6
177
8
68.56 93.10
ATON W/O
POST
PROCESSIN
G
252
9
152
7
57.37 92.09
ATON WITH
POST
PROCESSIN
G
265
4
115
9
60.67 92.65
C. Analysis of Results
From Table 1 I is clear that SAKBOT and ATON have
almost same accuracy. If SAKBOT is applied for colour video
then the discussion favors it.The overall accuracy is more than
90%. In conclusion there is a trade-off between Effectiveness
and accuracy of the system and this must be taken into account
during parameter selection.
D. Illumination Assessment
The objective of Illumination assessment is to decide
whether there are shadows present in an image. Wixon et al
[16][17] have proposed approaches for evaluating the
distribution of illumination over an Image. Here two values,
Brightness and energy values, are calculated for each image
pixel. Based on these the distribution of illumination over the
images are asserted. The brightness value of a pixel is defined
as the average of intensity values of neighbors of the pixel. The
energy value is defined as the average of intensity differences
between image pixels around the pixel. Rather than apply to
the entire input image, try to apply illumination assessment to
a selected foreground figure. Two criteria are employed for the
selection: size and value, where
Eq.(6)
Here and represent the number of bright and Dark
Pixels. If a foreground pixel has its R,G,B values all smaller
than those of the corresponding background pixel, the
foreground pixel is regarded as 'dark' pixel; otherwise regarded
as a 'bright' pixel. Energy value is calculated by the formula
∑
, j {b, d} Eq.(7)
E. Finlayson et. Al[18]
Two independent steps are involved in the shadow
removal method of Finlayson et al. [18]. The first step consists
in performing operations to get an illumination invariant
image, which is a gray scale representation proportional to the
color reflectance of the surfaces. To work properly, this step
requires some color calibration. The second step consists in
finding shadow edges and removing them from the color
image. Shadow edges are defined as edges that are part of the
intensity image, but not of the illumination invariant image.
With these shadow edges, it is possible to stamp these edges
off from the gradient of each color channel, and to recover a
shadow less color image by solving a linear system
approximating the Poisson’s equation to integrate the now
shadow less image from the modified gradient information.
∫ ( ) ( ) ( ) where k=R,G,B Eq.(8)
and E( ) is the spectral power distribution of the
illumination, S( ) is the surface reflectance distribution, and
( ) is the sensor sensitivity distribution.The Shadow edges
algorithm is as ( ) { ( )
( ) Eq.(9)
IV. CONCLUSION
This paper presented a critical review on the algorithms and
techniques used in detecting shadows in dynamic videos and
methods to remove them. Even if shadow removal is one of the
steps in achieving vehicle tracking; we now understand the
importance of Shadow removal in Vehicle Tracking and Speed
Estimation. Even though the topic is bit complex in nature
many great minds have dedicated their work in this field. If the
shadows are removed with right algorithms then the accuracy
of Vehicle tracking will be increased dramatically.
Abylash Kumar KT et al, Int.J.Computer Technology & Applications,Vol 5 (2),518-521
IJCTA | March-April 2014 Available [email protected]
520
ISSN:2229-6093
REFERENCES
[1] Huerta, I.; Holte, M.; Moeslund, T.; Gonzàlez, J.; "Detection and removal of chromatic moving shadows in surveillance scenarios," Computer Vision, 2009 IEEE 12th International Conference on, vol., no., pp.1499-1506, Sept. 29 2009-Oct. 2 2009..
[2] E. Salvador, P. Green, and T. Ebrahimi. Shadow identification
and classification using invariant color models. In Proceedings
of ICASSP 01, volume 3, pages 1545--1548. IEEE, 2001.
[3] A. Prati, I. Mikic, M. Trivedi, and R. Cucchiara, Detecting
moving shadows: Algorithms and evaluation, IEEE Transactions
on Pattern Analysis and Machine Intelligence, 25, pp. 918--923,
2003..
[4] D. Koller, K. Danilidis, H.-H. Nagel, Model-based object
tracking in monocular image sequences of road traffic scenes,
Int. J. Comput. Vis. 10 (3) (1993) 257–281.
[5] T. Horprasert, D. Harwood, and L.S. Davis, ―A Statistical
Approach for Real-Time Robust Background Subtraction and
Shadow Detection,‖ Proc. IEEE Int’l Conf. Computer Vision
’99 FRAME-RATE Workshop, 1999.
[6] I. Mikic, P. Cosman, G. Kogut, and M.M. Trivedi, ―Moving
Shadow and Object Detection in Traffic Scenes,‖ Proc. Int’l
Conf. Pattern Recognition, vol. 1, pp. 321-324, Sept. 2000.
[7] R. Cucchiara, C. Grana, G. Neri, M. Piccardi, and A. Prati, ―The
Sakbot System for Moving Object Detection and Tracking,‖
Video-Based Surveillance Systems—Computer Vision and
Distributed Processing, pp. 145-157, 2001.
[8] J. Stauder, R. Mech, and J. Ostermann, ―Detection of Moving
Cast Shadows for Object Segmentation,‖ IEEE Trans.
Multimedia, vol. 1, no. 1, pp. 65-76, Mar. 1999.
[9] Yang, M.-T.; Lo, K.-H.; Chiang, C.-C.; Tai, W.-K., "Moving
cast shadow detection by exploiting multiple cues," Image
Processing, IET , vol.2, no.2, pp.95-104, April 2008.
[10] Kai-Tai S.; Jen-Chao T., "Image-Based Traffic Monitoring With
Shadow Suppression," Proceedings of the IEEE , vol.95, no.2,
pp.413-426, Feb. 2007.
[11] Jun-Wei H.; Shih-Hao Y.; Yung-Sheng C.; Wen-Fong H.,
"Automatic traffic surveillance system for vehicle tracking and
classification," IEEE Transactions on Intelligent Transportation
Systems, vol.7, no.2, pp. 175-187, June 2006.
[12] Mohammed Ibrahim M; Anupama R., ―Scene Adaptive Shadow
Detection Algorithm‖, Proceedings Of World Academy Of
Science, Engineering and Technology, vol. 2, pp. 1307-6884,
Jan. 2005.
[13] E.H. Land, ―The Retinex Theory of Color Vision‖, Scientific
American, vol. 237, pp. 108-128, 1977.
[14] N. Herodotou, K.N. Plataniotis, and A.N. Venetsanopoulos, \A
color segmentation scheme for object-based video coding," in
Proceedings of the IEEE Symposium on Advances in Digital
Filtering and Signal Processing, 1998, pp. 25{29.
[15] I. Mikic, P. Cosman, G. Kogut, and M.M. Trivedi, \Moving
shadow and object detection in tra scenes," in Proceedings of
Int'l Conference on Pattern Recognition, Sept. 2000.
[16] L. Wixson, "Illumination Assessment for Vision-Based Traffic
Monitoring", proc. of 13th Int'l Conf. on Pattern Recognition,
Vol.3, pp. 56-62, 1996.
[17] L. Wixson, K. Hanna and D. Mishra, "Improved Illumination
Assessment for Vision-Based Traffic Monitoring", proc. of
IEEE Workshop on Visual Surveillance, pp. 34-41, 1998
[18] Finlayson, G. D., S. D. Hordley, M. S. Drew. Removing
Shadows from Images. - In: Proceedings of 7th European
Conference on Computer Vision - Part IV, ECCV’02, London,
UK, Springer-Verlag, 2002, 823-836
[19] Z. Sun, G.Bebis, and R.Miller, ―On-road vehicle detection: A
review,‖ IEEE Transactions Pattern Analysis and Machine
Intelligence, vol. 28, no. 5, pp. 694 –711, May 2006.
Abylash Kumar KT et al, Int.J.Computer Technology & Applications,Vol 5 (2),518-521
IJCTA | March-April 2014 Available [email protected]
521
ISSN:2229-6093