FACE RECOGNITION USING CLOUD COMPUTING.pptx

Published on May 2016 | Categories: Documents | Downloads: 45 | Comments: 0 | Views: 558
of 22
Download PDF   Embed   Report

Comments

Content

FACE RECOGNITION USING MOBILE CLOUDLET

ARCHITECTURE

INTRODUCTION
In our daily lives, face recognition applications that automatically identify an individual from captured images or videos.  Face recognition algorithms analyze images, extract information such as the shape, size and position of the facial features.  Then use these extracted features to search a facial database to locate matching images.  The algorithms used are of highest accuracy (e.g., over 90%).


AMBER ALERTS-APPLICATION OF FACE RECOGNITION
 face

recognition applications that will benefit one in particular is an extension of Amber Alerts to mobile phones.  Amber alerts are mainly used in the case of child missing. It is used by FBI.
A

central authority would extend their Amber alerts such that all available cell phones in the area where a missing child was last seen.  The alert would actively capture images and perform face recognition.

MOCHA ARCHITECTURE
To perform face recognition, we need a large database of images with which to compare the captured images taken by the cell phones.  This application is simply not possible using the mobile devices compute power alone, requiring access to cloud computing.  face recognition is performed on the MOCHA Architecture.  MOCHA means (Mobile Cloud Hybrid Architecture).


THE MOCHA ARCHITECTURE

MOCHA ARCHITECTURE COMPONENTS
a)Mobile devices b)Cloudlet c)Cloud A)mobile devices:  In our work, we assume mobile devices, such as smartphones and iPads,laptops.  The main task of the mobile device is to acquire the image and send it to the cloudlet for pre-processing.  After face recognition is complete, the mobile device receives the results back from the cloudlet or directly from the cloud

MOCHA ARCHITECTURE COMPONENT

B) Cloudlet:  The cloudlet is a special-purpose inexpensive compute-box with the capability of massively parallel processing (MPP).  cloudlet has 150 single Gflop of compute capability.  2 GB memory, 40 W total power consumption .  A cost under $100.  The cloudlet, equipped with a GPU and a lightweight CPU.  require high MPP power.

MOCHA ARCHITECTURE COMPONENT

C)cloud  Cloud computing provides computing and storage resources remotely in a pay-as-you-go manner.  a client program running on the cloudlet sends a request to the servers on Amazon AWS where the actual program runs and the results are sent back to the requester.

ALGORITHMIC OPTIMIZATION FOR MOCHA
Processing time and communication latency directly influence the speed of a cloud server’s response to requests for computation on data.  Two approaches for partitioning the computation  1)fixed algorithm  2)greedy algorithm  The task of these two are: (1) Fixed: the tasks are equally distributed among the available cloud servers . The total response time is the time that it takes for the last response to be returned. (2) Greedy: we first order the servers by their known response times, and give the first task to the server (cloudlet) that can complete this task in the minimum amount of time.


SIMULATIONS

In the first set of simulations, the processing time of each cloud server is a fixed value.  Chosen from a uniform distribution between 10 and 100 ms to complete each task.  Processing time of cloudlet is set as 100ms.  The latency for the communication from the cloudlet to each cloud server is also a fixed value.  Among the two algorithms the response time of greedy approach is the lowest with or without cloudlet 45% and 41% improvement in response time.


SIMULATED RESPONSE TIME USING VARIED
PROCESSING TIMES AND COMMUNICATION LATENCIES

CONTD…

Number of cloud server is 10.  Processing time of cloud server and cloudlet is set to 1ms for per task.  The communication latencies with the cloudlet varied from no difference to an increasingly more heterogeneous environment.  The latencies chosen are uniformly distributed between 10ms and upper limit gets varied up to 1s.


SIMULATED RESPONSE TIME USING A FIXED PROCESSING TIME AT EACH CLOUD SERVER AND VARYING COMMUNICATION LATENCIES.

USES OF MOCHA ARCHITECTURE
A)face detection.  The face detection phase of the overall CloudVision process determines the potential locations of the human faces within an image.  For face detection we use Haar features and Haar classifiers to perform face detection.  These simple classifiers in this initial stage have low computational complexity.  Operate on a large amount of data, and they produce a large number of face candidates.  The detection is operation is performed using pipelines here we use 32 stage pipeline

B)FACE RECOGNITION: The face recognition phase of the overall CloudVision process determines the match-likelihood of each face to a template element from a database.  The face recognition is performed using the recognition algorithm.


C)CLOUD VISION:

Assume that the mobile device, which has very limited compute power.  It simply captures the image and sends this off for face recognition.  While the mobile device could directly send the image to the cloud, this would require the mobile to coordinate the computation partitioning, as well as with the various cloud servers.  The cloudlet extracts the detected faces from the original image and sends each face to a different cloud server to perform face recognition


EXPERIMENTAL SETUP




 



Our experimental hardware platform is a distributed heterogeneous cluster of 13 servers, workstations and a laptop. These computers are distributed over three separate geographic locations, connected through a high-speed broadband link . First place is the university and the other two locations(offset1,offset2). The main location, our university, contains the laptop, cloudlet emulator, mobile emulator, and a majority of the cloud servers. The other locations were accessed using Microsoft Remote desktop to run the software we have developed.

EXPERIMENTAL RESULTS

CONCLUSIONS OBTAINED FROM THE GRAPH:
As the number of cloudlet increases the response time also increases.  In case of the high server count, it becomes a daunting task for the mobile device to capture images from the camera, dispatch them to multiple cloud servers.There by causing the performance inflection point.


CONCLUSION A mobile-cloudlet-cloud architecture called MOCHA is proposed as a platform for our target face recognition application.  This architecture is designed to minimize the overall response time of the face detection and face recognition.  MOCHA indeed reduces the overall response time for face recognition.


REFERENCES










Cloud-Vision: Real-time face recognition using a mobilecloudlet-cloud acceleration architecture Soyata, T., Muraleedharan, R. ; Funai, C, Heinzelman, W. Computers and Communications (ISCC), 2012 IEEE . Ming-Hsuan Yang, David Kriegman, and Narendra Ahuja, “Detecting faces in images : A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence. T. Soyata, R. Muraleedharan, J. H. Langdon, C. Funai, M. Kwon and W. B.Heinzelman, “Mobile cloud-based compute/communications infrastructure for battlefield applications,” in SPIE Defense, Security,and Sensing 2009. Modeling and Simulation for Defense Systems and Applications . [Mahadev Satyanarayanan, Paramvir Bahl, Ram´on C´aceres, and NigelDavies, “The case for VM-based cloudlets in mobile computing,” IEEEPervasive Computing. David B. Kirk and W.-M. Hwu, Programming Massively Parallel Processors, Morgan-Kaufmann, 2010

THANK YOU

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close