A Survey on Shadow Detection and Removal in Images and

Video Sequences
Arti Tiwari, Pradeep Kumar Singh, Sobia Amin
AmitySchool of Engineering and Technology
AmityUniversity Uttar Pradesh, Noida, India.
arti.tiwari94@gmail.com, pradeep_84cs@yahoo.com , sobiya.amin60@gmail.com
Abstract: Shadow Detection and Removal is the in a systematic manner. Finally in section 4, we have
process to enhance the performance, reliability and concluded our finding with this survey along with the
accuracy of the computer vision applications future scope.
including image segmentation and object recognition,
object tracking ,surveillance etc. Detection and 2. Related Work
Removal of shadow from the images and videos can
reduce the undesirable outcomes in the computer In order to perform this systematic review we have
vision applications and algorithms. The prime formulated some research questions, which are
objective of this survey paper is to analyze the mentioned as follows:
performance of various currently used shadow
detection techniques. In this paper we have discussed RQ.1 To Identify the most popular technique for
the techniques for detecting and removing shadow shadow detection and removal ?
from the still images and video sequences. The scope RQ.2 To identify the most common issues and
of discussed shadow detection and removal challenges in existing techniques for shadow
techniques is limited to different scenarios: (i) detection and removal in Images and Video?
Shadow detection for Indoor and Outdoor scenes, (ii)
Shadow detection using fixed or moving camera, (iii) RQ.3 To illustrate the limitations and future scope of
Shadow detection of umbra and penumbra shadows the existing techniques?
etc. To conduct the analytical and illustrative
review we have downloaded nearly 160 papers from
Keywords: Shadow detection, Shadow removal, various digital libraries i.e. IEEE Xplore, Science
Edge detection, Shadow mask, Surveillance, Hue Direct, ACM, Google Scholar, after reading the
Saturation Value (HSV). paper’s title, abstract and conclusion part manually
we have identifies 20 papers for the review which are
1. Introduction most suitable and related to our objective. In order to
do this analytical review we have followed the
Shadow is formed when an opaque object barricades similar approach as adopted by Singh et al. [29].
the light from its source. A shadow in the computer In [1], Prati et al. proposed, shadow
vision applications has become a serious problem as detection approach to detect the moving shadow. In
it causes various issues as distortion of object shape, their approach the environment is considered as
object merging, and actual object loss; accordingly outdoor scenes having several lanes observed by
shadow produces a negative impact on the accuracy single fixed camera. In their work two models are
and efficiency of the result. Therefore, effective and discussed and comparison is done on the basis of two
robust shadow detection techniques are essential for criticality; Accuracy and Computation Complexity.
the computer vision applications such as surveillance, The first model discussed is SAKBOT (Statistical
object recognition, traffic analysis, tracking systems and Knowledge Based Object Tracker) and ATON
etc. In this paper we have shown our related work in (Autonomous Transportation Agents and Onscreen
systematic and well-organized manner and finally, networked Incident Management). As the conclusion
concluded our survey with the pros and cons with ATON system is better discriminate moving shadow
other essential conditions of every discussed and moving object, and SAKBOT can better detect
technique. This paper is divided into four sections. the shadow which shares the same movement of the
Introduction is followed by Section 2, in which the corresponding object. In future, for single system the
literature review of shadow detection and removal purpose will be to better adapt the parameters and
technique are reported and briefly discussed. In will try to explain the saturation behavior of
section 3, the related work is analyzed and presented shadowed point of SAKBOT.

978-1-4673-8203-8/16/$31.00 2016
c IEEE 518
In [2], Shiting et al. proposed an approach to detect may be extended for the non-stationary background
shadow edge using clustering. The main purpose of images.
their approach is to improve the Finlayson’s method In [5], Ishida et al. proposed a method
[21] which is based on Canny, process for edge constructs three shadow model. The method uses
detection [22]. In proposed method, it takes three features robust to illumination changes as: (i)
minimally processed (linier) image as input, and Color Information, (ii) Peripheral Increment Sign
takes the derivative difference of both light-invariant Correlation (PISC), and (iii) Edge Information, to
image and brightness image followed by K-mean construct the model each of these feature removes
clustering algorithm; further result is improved by shadow effect from the moving foreground and
some corrections and finally produce a shadow edge further result will be improved by region based
mask. The algorithm improves the limitation of fixed analysis. For the probability model of distribution,
parameter in Finlayson’s method and comparatively method uses DPEM algorithm [25], results are
gives robust and stable results and performs corroborated by experiments using real video of
efficiently in both of the scenes i.e. indoor and outdoor scene by using single camera. In comparison
outdoor, their proposed method can not give accurate with the method [8], their proposed method is better
result for the Vague shadow edge detection same as in two scenarios: (a) Proposed method uses HSV
the Finlayson’s method. color model rather than using YUV color model (b)
In [3], Khare et al. proposed a method for To improve the accuracy of the result, method
shadow detection and removal using Discrete improves the process. In their work, future method
Wavelet Transform (DWT). DWT has the multi- should be focused on to determine the parameters of
resolution property which spits the images into four initial shadow model and automatically detection of
different bands. In proposed method, DWT is used shadow. It is stated that, to construct the shadow
with the HSV (Hue, Saturation, Value) color model. model, without any explicit background image and
As a new threshold value the Relative standard detailed knowledge of result.
deviation is used rather than the standard deviation In [6], Sasi et al. proposed, a method for
because it is more informative and consistent. shadow detection form a single image based on
Method does not require any supervised learning and Fuzzy split and merge approach that uses fuzzy
human interaction to determine the threshold, it can predicate. Fuzzy predicate is trained by the ANFIS
be determine automatically. Method is dependent on (Adaptive Neuro-Fuzzy Inference System). Method
only one parameter i.e. Wavelet Coefficient. The first performs the top-down approach that recursively
proposed method is analyzed and assessed for the splits the image into four homogeneous quad tree
video sequences. And comparatively method gives block and the image is sparsely represented in the
good qualitative and quantitative result for both tree form then, follow the bottom-up approach to
outdoor and indoor conditions and also best suited in merge the adjacent homogeneous region using the
the condition of poor lighting. fuzzy predicate. The comparative study states that
In [4], Rufenacht et al. proposed, automatic proposed method gives better performance in term of
shadow detection method using Near-Infrared (NIR) accuracy and efficiency.
information. The method uses the inherent sensitivity In [7], Huerta et al. proposed an approach to
of camera sensors to Near-infrared spectrum. The successfully detect both umbra and penumbra
method requires the visible image and NIR images shadow regions. First, for detection and removal of
input and gives high quality binary shadow mask as the chromatic moving shadow in surveillance
output. Their algorithm works best for minimally scenario a bottom-up approach based on gradient and
processed images (Raw images). Their proposed color models is presented. Secondly, to refine the
method gives computationally fast and accurate result performance of bottom-up chromatic shadow
in comparison to a segmentation based method; Guo detection approach, a tracking based top-down
et al. [23], and a pixel based method Tian et al. [24], approach is used where motion filters are used for
and can also perform well in complex illumination tracking. The shadow detection rate is improved
conditions. In spite of good performance there are approximately by 13% by using top-down shadow
some failure points; if an object has lower reflectance tracking. Unlike other methods, this approach does
in the NIR than in the visible image then it can be not need any priori assumption about camera
wrongly detected as shadow and if an object has location, surface geometrics, surface textures, shapes
higher reflectance in NIR than in visible image then and types of shadows, objects and background
shadow can be neglected. Their discussed method is properties and their proposed method is more
more flexible, robust and gives good performance for advantageous that other method by having both
both indoor and outdoor scenes. As in the future it indoor and outdoor sequences.

2016 6th International Conference - Cloud System and Big Data Engineering (Confluence) 519
 In [8], Shim et al. proposed, a method for intensity model has been developed using intensity
shadow removal for vision based robot localization change rate, intensity variation of the penumbra
using the surveillance camera installed in the room shadow. By adjusting the intensity model the initial
ceiling. HSV (Hue, saturation, value) color space is umbra and penumbra shadow edge segment is
used to detect the shadow and then shadow mask is identified then the continuous complete boundary is
applied to remove the shadow form a single image. optimized globally with the level set method. The
The method mainly comprises three steps: (i). Firstly proposed method, gives good accuracy for the even
the method detects the binary background subtraction surface, outdoor based single images but inconsistent,
image. (ii) Using HSV color space it detects the asymmetric casting background is still the problem.
object area, and (iii) Removal of shadow and detects As the future work other parameters as color and
the actual object area. The result can be shown by the gradient should be considered for the performance
experiments of mobile robot localization. The method improvement.
gives fast and efficient result in detection and In [12], Wang et al. proposed a method by
removal of shadow. In future the false detection rate combining two methods of shadow detection color
should be minimize while increasing the speed of based and model based. First moving region of the
shadow removal. shadow is detected using property-based method and
In [9], Dai et al. proposed, a statistical then coarse region is obtained by model-based
discriminant model for video surveillance. The model method, then this coarse region is used in the shadow
first, extracts various features for classification then detection approach based on HSV color-space.
adopts statistical learning as Partial Least Square Proposed method performed well in moving shadow
(PLS) and Logistic Discriminant (LD) to distinguish detection, especially for the shadow which lies in the
moving shadow and finally post processing is applied boundary of the vehicle. However, this method is not
to improve the accuracy of result. found suitable for the shadow which lies under the
Their model allow to detect moving shadow without vehicle or the shadow which is covered by the
any prior knowledge about scenes and illumination vehicle.
and it adapts various scenarios in indoor and outdoor In [13], Yang et al. proposed an approach to
automatically and only needs some set of labeled remove the shadow in work piece. The method first
samples. In future, additionally the model should be converts the RGB image to HSV color model,
developed, with semi-supervised learning so, it will according to the association between the variation of
reduce the requirements for labeled samples. values (V)-elements and the shadow of the image, the
In [10], Zhu et al. presented a method based elements of the corresponding shadow is obtained.
on some assumptions about the complexity and Then a homomorphic filter is designed, which is used
categorization of shadow images. The prime to filter the elements of shadow in the image and a
objective of their method is; (a) It gives a survey on new image is incorporated without having the
the shadow removal problems and challenges in shadow. The quality is further improved by
single image (b) The idea presented, will overcome subsequent image processing. Their method
some shadow removal problems such as illumination, effectively, completely removes the shadow in the
color constancy and type of reflection. In their outer region of the workpiece but shadow in the inner
survey, all the methods give good result only when region is not effectively removed it is reduced to
shadow edge of the image is correctly detected. some extent.
Unlike these methods a new and effective method is In [14], Asaidi et al. proposed an approach
proposed for shadow removal without shadow to automatically detect and extract shadow based on
detection. In proposed approach the original images various properties acquired by the spectral, temporal
is transformed to YCbCr color space and the Y and geometric analysis of shadow, which detects the
channel is split into a base layer and a detail layer. shadow region using time and direction of shadow.
Method is having some initial assumptions and Using the spatial analysis, method analyzed the
limitation such as: (a). Image pixel values are linearly candidate shadow region using the direction of the
related to the intensity of the incident light (b) Images shadow according to the time of the day. And then
are unaffected by specularity and inter-reflections (c) colorimetric is used to verify the hypothesis of the
The imaging device is of narrow-band sensors (d)The detected region. Proposed method is evaluated
field of view is lit by Planckian illuminants. through various real environment videos.
In [11], Dong et al. proposed method to In [15], Russell et al. proposed a method for
detect the penumbra shadow edge form a single detection of moving cast shadow based on spatial and
image of outdoor scenes. In the proposed method, temporal color constancy among the pixels. The
based on three parameters center position, method uses the texture information of the different
orientation, and width of the penumbra shadow an sub regions, then spatial constancy is obtained to

520 2016 6th International Conference - Cloud System and Big Data Engineering (Confluence)
 determine the similar texture regions and temporal extended to object edge detection and smoke region
constancy is obtained for the classification of detection.
shadow. The proposed method is efficient for both In [19], Russell et al. proposed a method for
indoor and outdoor sequences and for both chromatic shadow detection for real time moving vehicle and
and achromatic shadows. Proposed approach also traffic sequences. The method is works on two
resolves the problem of foreground and background aspects; (a) it performs image-line analysis. (b)
camouflage. illumination direction of the light source, to detect the
In [16], Pan et al. proposed a method for intervals with a decreasing function in the scanned
shadow detection of remote sensing images using the images of the method extracts two feature, namely,
actual edge features rather than the arbitrary pixels of Intensity and Spatial relationship among the points.
the images region based on WGER (Weighted Method gives good performance when there is
Gradient Edge Region). To show the edge intensity foreground-background and background –foreground
of a region a feature edge gradient is proposed, camouflages and for real time outdoor vehicles and
further for improvement of detection of shadow edge traffic sequences.
texton texture analysis is used to determine the In [20], Qi et al. presented an efficient
weight (number of pixels) of each region [26]. cascade method for the cast shadow detection.
Method gives good accuracy and better result in Method works in very systematic way firstly, it
comparison to the methods of Liu et al. [27] and extracts the initial moving frames from the Gaussian
Chung et al. [28]. Mixture model (GMM) then to separate a portion of
In [17], Khan et al. proposed a method for moving object from the initial moving pixels Local
automatically detection and removal of shadow from Binary pattern(LBP) is applied. Subsequently for the
a single image. Method automatically learns the improvement intensity ratio is acquired and then by
features by using Convolutional Deep Neural post processing techniques the spurious pixels will be
Networks (ConvNets) and then using detected corrected. Their proposed method gives good
shadow mask. Bayesian formulation is proposed for performance comparatively but for some of the
the automatically removal of shadow. The proposed outdoor scenes misclassified the pixels.
method gives good performance for both umbra and
penumbra shadows. 3. Analysis of Shadow Detection
In [18], Shen et al. proposed a efficient
learning based framework for shadow detection from and Removal
single image using structured Convolutional Neural After the analytical review of Shadow
Network(CNN). Method uses CNN to extract local Detection and Removal techniques, we can classify
structure of shadow edges to improve the local all the techniques into two parts: images and videos.
consistency throughout pixel labels. Shadow and The popular techniques for shadow detection and
bright measure are computed by the detected shadow removal in images and video are based on soft
edge and subsequently least square optimization computing and color based models. The detailed
problem will be solved efficiently for the shadow classification is shown in Table 1.
recovery. The proposed method can be further

Table 1. Analysis on Shadow Detection and Removal Techniques in Images and Video Sequences

SHADOW DETECTION AND REMOVAL TECHNIQUES

Category S.No. Techniques and Methods Environment/Scenes Reference
Images 1. Clustering Algorithm(K-means) Indoor and Outdoor [2]
2. Model is based on Near-Infrared (NIR) Information Indoor and Outdoor [4]
3. Fuzzy System(ANFIS to implement sugeno fuzzy Indoor and Outdoor [6]
model)
4. Algorithm uses Background Subtraction and HSV Indoor [8]
Color Space.
5. Mathematical model YCbCr Color space Indoor and Outdoor [10]
6. Algorithmic model using watershed algorithm, the Outdoor [11]
Canny operator and level set optimization
7. Algorithmic model using HSV color space and Indoor and Outdoor [13]
Homomorphic Filter
8. Algorithm uses WGER (Weighted Gradient Edge Outdoor [16]
Region).
9. Convolutional Deep Neural Networks (ConvNets) Indoor and Outdoor [17]

2016 6th International Conference - Cloud System and Big Data Engineering (Confluence) 521
 10. Model is based on Convolutional Neural Indoor and Outdoor [18]
Network(CNN).
Videos 11. Comparative study of two models(ATON and Outdoor [1]
SAKBOT).
12. Discrete Wavelet Transformation(DWT) with HSV Indoor and Outdoor [3]
Color Model.
13. HSV Color Space and DPEM Algorithm. Outdoor [5]
14. Mathematical Model using Top-down and Bottom-up Indoor and Outdoor [7]
approach
15. Model is based on Partial Least Square (PLS) and Indoor and outdoor [9]
Logistic Discriminant (LD)
16. Algorithmic model using HSV Color Space. Outdoor [12]
17. Algorithm uses spectral, temporal and geometric Outdoor [14]
properties of shadow.
18. Uses spatial and temporal color constancy among the Indoor and Outdoor [15]
pixels and texture Information.
19. Based on two aspects (a)image-line analysis. (b) Outdoor [19]
illumination direction of the light source.
20. Model is based on the Gaussian Mixture model Outdoor [20]
(GMM) and Local Binary pattern(LBP)

4. Conclusions and Future work non-stationary background and vague region, inner
region and smoke region shadow detection and
The overall outcome of this survey on Shadow removal. Based on this comprehensive survey, the
Detection and Removal in Images and Videos can be techniques may be improved further to effectively
demonstrated by illustrating the answers of our detect and remove the shadow from images and
research questions RQ1, RQ2 and RQ3. As we videos. There is lot of scope for further improvement
mentioned in our research question RQ1, we have with the help of the AI techniques. In some cases,
found the various techniques being used in last 15 researchers may also apply soft computing hybrid
years, which are demonstrated and explained in the approach for the effective, efficient and accurate
Table 1. By reviewing the considered research results for shadow detection techniques.
papers, we found that the techniques being used gives
efficient and effective performance covering all the References
aspects as indoor and outdoor scenes. We have [1] A. Prati, I. Mikib, C. Grana and M.M. Trivedi, Shadow
collected the papers for penumbra and umbra Detection Algorithms for Traffic Flow Analysis: a
shadows, static and moving cameras. These Comparative Study, published in IEEE Conference on
techniques have resolved the problem of background Intelligent Transportation Systems- Oakland (CA), USA,
and foreground camouflage. We have also focused on pp. 340-345, 2001.
the efficiency and identified various issues and [2] W. Shiting, Z. Hong, Clustering-based Shadow Edge
challenges with the existing techniques such as; (a) Detection in a Single Color Image, published in
International Conference on Mechatronic Sciences, Electric
There are techniques which have failed to detect the
Engineering and Computer (MEC), pp. 1038-1041. 2013.
vague shadow edges accurately (b) There are [3] M. Khare, R.K. Srivastava, A. Khare, Moving shadow
techniques which wrongly detect the shadow and can detection and removal a wavelet transform based approach,
neglect the shadow when the object is having lower published in Computer Vision, IET, Vol. 8, Issue 6, pp.
and higher reflectance respectively than visible 701–717, 2014.
images in NIR (c) Few techniques cannot efficiently [4] D. Rüfenacht, C. Fredembach, S. Süsstrunk, Automatic
detect and remove shadow with the smoke region and and Accurate Shadow Detection Using Near-Infrared
low lighting area and future scopes in this particular Information, published in IEEE Transactions on Pattern
area of Shadow Detection and Removal. Analysis and Machine Intelligence, Vol. 36, No. 8,
pp.1672-1678,2014.
In future, some improvement can be made in
[5] S. Ishida, S. Fukui, Y. Wahori, M.K. Bhuyan, R.J.
terms of efficiency and accuracy by considering the Woodham, Shadow detection by three shadow models with
following aspects such as; (a) By minimising the features robust to illumination changes, published in 18th
false detection rate while increasing the speed of International Conference on Knowledge Based and
shadow detection and removal (b) To implement the Intelligent Information & Engineering Systems, pp. 1219-
techniques of shadow detection and removal 1228, 2014.
automatically without detailed knowledge of result [6] R.K. Sasi, V.K. Govindan, Fuzzy split and merge for
i.e. unsupervised (c) To implement the techniques shadow detection, published in Egyptian Informatics
which efficiently detects the shadow edges with the Journal, Vol. 16, pp.29-35, 2015.

522 2016 6th International Conference - Cloud System and Big Data Engineering (Confluence)
[7] I. Huerta, M.B. Holte, T.B. Moeslund, J. Gonzàlez, Conference on Computer Vision and Pattern Recognition
Chromatic shadow detection and tracking for moving (CVPR),IEEE, pp. 2067 – 2074, 2015.
foreground segmentation, published in Image and Vision [19] M. Russell, J.J. Zou and G. Fang, Real-time vehicle
Computing, Vol. 41,pp. 42-53, 2015. shadow detection, published in Electronics Letters, Vol.51,
[8] J.H. Shim , Y.I. Cho, A Shadow Removal Method for a Issue 16, pp.1253-1255, 2015.
Mobile Robot Localization using External Surveillance [20] M. Qi, J. Dai, Q. Zhang, J. Kong, Cascaded Cast
Cameras, published in The 12th International Conference on Shadow Detection Method in Surveillance Scenes,
Mobile Systems and Pervasive Computing, pp.150-155, published in Optik - International Journal for Light and
2015. Electron Optics, Vol.125, Issue 3, pp. 1396-1400, 2014.
[9] J. Dai, D. Han, X. Zhao, Effective moving shadow [21] G. D. Finlayson, S. D. Hordley and C. Lu, On the
detection using statistical discriminant model, published in removal of shadows from images, IEEE Transactions on
International Journal for Light and Electron Optics, pp. 1- Pattern Analysis and Machine Intelligence, Vol. 28, 2006,
29, 2015. pp. 59-68, doi: 10.1109/TPAMI.2006.18.
[10] X. Zhu, R. Chen, H. Xia, P. Zhang, Shadow removal [22] L. Xu, F. Qi and R. Jiang, Shadow Removal from a
based on YCbCr color space, published in Single Image. 6th International Conference on Intelligent
Neurocomputing, Vol.151, Part 1, pp. 252-258, 2015. Systems Design and Applications, Vol. 2, 2006, pp. 1049-
[11] Q. Dong, Y. Liu, Q. Zhao, H. Yang, Detecting soft 1054, doi: 10.1109/ISDA.2006.253756.
shadows in a single outdoor image:From local edge-based [23] R. Guo, Q. Dai, and D. Hoiem, Single-Image Shadow
models to global constraints, published in Computer and Detection and Removal using Paired Regions, in
Graphics, Vol. 38, pp. 310-319, 2014. International Conference on Computer Vision Pattern
[12] S.K. Wang, B. Qin, Z.H. Fang, Z.S. Ma, Fast Shadow Recognition, IEEE, pp. 2033-2040, 2011.
detection According to the Moving Region, published in 6th [24] J. Tian, L. Zhu, and Y. Tang, Outdoor Shadow
International Conference on Machine Learning and Detection by Combining Tricolor Attenuation and
Cybernetics, Hong Kong, pp 19-22 ,2007. Intensity, EURASIP Journal on Advance Signal
[13] W.D. Yang, W. Gu, K. Peng, L.B. Liu, Research on Processing, Vol. 2012, p. 116, May 2012.
Removing Shadow in Workpiece Image Based on [25] T. Kimura, Y. Nakada, A. Doucet and T. Matsumoto
Homomorphic Filtering, published in International Semi-supervised learning scheme using Dirichlet Process
Workshop on Information and Electronics Engineering Em-Algorithm, IEICE Tech. Rep, pp. 77–82, 2009.
(IWIEE), pp. 2360 – 2364 ,2012. [26] M. Varma, A Statistical Approach to Texture
[14] H. Asaidi, A. Aarab, M. Bellouki, Shadow Detection Classification from Single Images, International Journal of
Approach Combining Spectral and Geometrical Properties, Computer Vision, Vol. 62, Issue. 1-2, pp. 61-81, 2005.
published in International Conference Multimedia [27] J. Liu, T. Fang, and D. Li, Shadow Detection in
Computing and System (ICMS), pp. 389-393, 2012. Remotely Sensed Images Based on Self-Adaptive Feature
[15] A. Russell, J.J. Zou, Moving Shadow Detection Based Selection, IEEE Transaction on Geoscience Remote
on Spatial-Temporal Constancy, published in 7th Sensing, Vol. 49, No. 12, pp. 5092 -5103, Dec. 2011.
International Conference on Signal Processing and [28] L. Chung, Y. R. Lin, and Y. H. Huang, Efficient
Communication System, pp. 1-6, 2013. Shadow Detection of Color Aerial Images Based on
[16] B. Pan, J. Wu, Z. Jiang, X. Luo, Shadow Detection in Successive Thresholding Scheme, IEEE Transaction on
Remote Sensing Images based on Weighted Edge Geoscience Remote Sensing, Vol. 47, No. 2, pp. 671–682,
Gradient Ratio, published in Geoscience Feb. 2009.
and Remote Sensing Symposium (IGARSS) IEEE, pp.505- [29] P.K.Singh, O.P. Sangwan, A.Sharma, A Systematic
508, 2014. Review on Fault Based Mutation Testing Techniques and
[17] S.H. Khan, M. Bennamoun, Automatic Shadow Tools for Aspacts- J Programs, published in IEEE, 3rd
detection and Removal from a Single Image, published in International Conference on Advance Computing
Transactions on Pattern Analysis and Machine Intelligence, Conference(IACC), pp. 1455-1461, 2013.
IEEE, Issue 99, pp. 1-17, 2015.
[18] L. Shen, T.W. Chua, K. Leman, Shadow Optimization
from Structured Deep Edge Detection, published in

2016 6th International Conference - Cloud System and Big Data Engineering (Confluence) 523