IRJET- Identification of Missing Person in the Crowd using Pretrained Neural Network
0 likes13 views
The document describes a proposed system to identify missing persons in crowded areas using pretrained convolutional neural networks. The system would involve collecting images of missing persons from different angles to create a dataset for training. An AlexNet pretrained neural network would then be used to detect faces in live video captured by a drone camera of crowded areas. Detected faces would be cropped, stored in a database, and used to further train the network. During testing, the system could identify missing persons by displaying their images when detected in the crowd. The goal of the system is to help police efficiently locate missing people in crowded public settings like festivals or meetings.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 390
IDENTIFICATION OF MISSING PERSON IN THE CROWD USING
PRETRAINED NEURAL NETWORK
Mr. A. David rajkumar1, Mr. R. Karthick Raja2, Mr. S. Sankar Ganesh3, Dr. V. R. S. Mani4
1,2,3Department of Electronics and Communication Engineering , National Engineering College, Kovilpatti
4Associate Professor, Department of Electronics and Communication Engineering,
National Engineering College, Kovilpatti
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - The proposed work deals with the identification of a missing person inthecrowdedareasuchasfestivals,temples,public
places, meetings, etc. Nowadays identification of a particular person in the crowded area is a complex task. For this, a solutionis
provided on this with the helpof a deep learning concept. Convolutional Neural Network(CNN)isemployedfortheidentificationof
a person. The missing person is identified using various facial features. Face Detection plays an important role in this project.
AlexNet which competed in the ImageNet Large Scale VisualRecognitionChallengeisused.Theprimaryresultwasthatthedepthof
the model was essential for its high performance, which was computationally expensive but made feasible due to the utilization of
graphics processing units (GPUs) during training. By providing training with various imagesitispossibletofindthespecifiedobject
in the target area. The flow of the project is getting the live-streamed video. The faces in the video are cropped and stored in the
database. The people who need to identify has a dataset with several images. The AlexNet is trained by configuring several layers
with our own dataset. The database images are classified by using the pretrained network to identifythepresenceofapersoninthe
crowd. Further, the position of the person should be obtained and provide live tracking with the KLT algorithm.
1. INTRODUCTION
Identification of a missing person in the crowded scene such as festivals,temples,publicplaces,andmeetingsisa complextask.
A solution with the help of deep learning concept is to be proposed. Pretrained Convolutional Neural Network (CNN) is
employed for the identification of a person. The Convolutional Neural Network indicates that the network employs a
mathematical operation called convolution. Convolution is a specialized kind of linear operation. Convolutional networks are
simply neural networks that use convolution in place of general matrix multiplication in at least one of their layers. AlexNet
contained eight layers, the first five were convolutional layers, some of them followed bymax-poolinglayers,andthelastthree
were fully connected layers. Itusedthe non-saturatingReLUactivationfunction,whichshowedimprovedtraining performance
over tanh and sigmoid. The layers are trained based on our requirement to identify the particular object. The last convolution
layer and fully connected layers are configured to our requirement.
2. LITERATURE SURVEY
The idea forthe project was erected from the papers thatwere glimpsedislistedbelow.However,anoriginalplanindesignuses
some differential way from the papers mentioned below. A. Agarwal and B. Triggs[1] proposed a new technique to extract the
features of the image. By the new technique, morefeaturesareextractedcomparedtotheoldtechnique.Thenewtechniqueused
multilevel image coding for extracting the features of images. X. Cao, Y. Wei, F. Wen, and J. Sun [2] developed an approach that
says replacing the old technique for aligning the shape of the face with aligning the face by explicit shape regression. Using this
aligning the shape of the face becomes very easy. D. Chen, S. Ren, Y. Wei, X. Cao, and J. Sun [3] proposed a new technique for the
detection and alignment of face. In this technique detection and alignment of the face done in one processbecausejointcascade
is used for the detection and alignment of face.
D. Eigen and R. Fergus [4] proposed that Predicting depth, surface normals and semantic labels with a common multi-scale
convolutional architecture. P. Felzenszwalb, R. Girshick, D. McAllester, and D.Ramanan [5] proposed anewmodelthatcanable
to detect the object i.eObjectdetection with discriminatively trained part-basedmodels.R.Girshick,J.Donahue,T.Darrell,andJ.
Malik [6] Extracting the features is very important so that they proposed to extract more features from the image sothatthey
highly focused the rich feature hierarchies for accurate object detection and semantic segmentation.N. Hu, W. Huang, and S.
Ranganath [7] proposed that by the position of the head in the image it is very easy to embed and map the image so they
proposed an estimation for the position of head by non-linear embedding and mapping. Y. Jia, E. Shelhamer, J. Donahue, S.
Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell [8] proposed a new architecture for the embedding process because
the old architecture is not working properly to embed because of fast feature embedding so that they proposed a new
Convolutional architecture for fast feature embedding.
A. Kumar, R. Ranjan, V. Patel, and R. Chellappa [9] proposed a method which aligns the face with the help of regression. L. Liang,
R. Xiao, F. Wen, and J. Sun [10] developed an approach that says replacing the old technique for aligning the shape of the face](https://image.slidesharecdn.com/irjet-v7i285-201107101618/85/IRJET-Identification-of-Missing-Person-in-the-Crowd-using-Pretrained-Neural-Network-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 391
with a component-based discriminative search. J. Long, E. Shelhamer, and T. Darrell [11] proposed that Fully convolutional
networks forsemantic segmentation. Q. Zhao, S. S. Ge, M. Ye, S. Liu,and W. He, [12] proposed that Learning saliency features for
face detection and recognition using the multi-task network. M. D. Zeiler and R. Fergus[13] In this proposed system it says how
to understand the convolutional network and how to visualizeit. Y. Gong, L.Wang, R. Guo, and S. Lazebnik, [14] proposed that
Multi- scale orderless pooling of deep convolutional activation features. A. Babenko, A. Slesarev, A. Chigorin, and V. Lempitsky
[15] proposed a new method to retrieve the image so that the original image can be retrieved so that some new codes are
proposed for the image retrieval i.e. Neural codes for image retrieval.A. Eitel, J. T. Springenberg, L. Spinello, M. Riedmiller, and
W. Burgard [16] proposed that Multi modal deep learning for robust RGB-D object recognition.
3. PROPOSED WORK
The aim of this project is to identify the person who is missing in the crowded area using the drone which is having the camera
and by this the person who ismissed in the crowd will be easily identified within a minute. This will be very usefulforthepolice
to find the missing person. Nowadays many children are missing in the festival crowd so that by this project the police can
monitor from one place and then they can able to find the person and also they can live track the person where he is moving so
that they can save the time and can able to identify the missing person veryeasily
Fig – 1 Block Diagram of Proposed Work
First, the images of missing persons arecollected.Eachandeveryimageshouldbefromdifferentanglessothatidentificationwill
be so simple. After collecting the images a dataset for the images should be created. Then it will be under the training process.
For this process, AlexNet is used, which is a pretrained neural network. Then the camera which is attached on the drone will
capture the video of the crowded area and it will be communicating with the router . In this project IP camera is used , it will be
communicating with the help of the IP Address so that router is used to communicate. Then it will be given to the controller so
that the captured image will undergo the cropping process. After cropping the image a database for the cropped image will be
created. After the database is created the database will be used to train the network. After completing the training process
testing is done in a classifier having only a softmax layer and if the image is detected means in the display the image will be
shown.](https://image.slidesharecdn.com/irjet-v7i285-201107101618/85/IRJET-Identification-of-Missing-Person-in-the-Crowd-using-Pretrained-Neural-Network-2-320.jpg)
IRJET- Identification of Missing Person in the Crowd using Pretrained Neural Network
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072 © 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 390 IDENTIFICATION OF MISSING PERSON IN THE CROWD USING PRETRAINED NEURAL NETWORK Mr. A. David rajkumar1, Mr. R. Karthick Raja2, Mr. S. Sankar Ganesh3, Dr. V. R. S. Mani4 1,2,3Department of Electronics and Communication Engineering , National Engineering College, Kovilpatti 4Associate Professor, Department of Electronics and Communication Engineering, National Engineering College, Kovilpatti ---------------------------------------------------------------------***---------------------------------------------------------------------- Abstract - The proposed work deals with the identification of a missing person inthecrowdedareasuchasfestivals,temples,public places, meetings, etc. Nowadays identification of a particular person in the crowded area is a complex task. For this, a solutionis provided on this with the helpof a deep learning concept. Convolutional Neural Network(CNN)isemployedfortheidentificationof a person. The missing person is identified using various facial features. Face Detection plays an important role in this project. AlexNet which competed in the ImageNet Large Scale VisualRecognitionChallengeisused.Theprimaryresultwasthatthedepthof the model was essential for its high performance, which was computationally expensive but made feasible due to the utilization of graphics processing units (GPUs) during training. By providing training with various imagesitispossibletofindthespecifiedobject in the target area. The flow of the project is getting the live-streamed video. The faces in the video are cropped and stored in the database. The people who need to identify has a dataset with several images. The AlexNet is trained by configuring several layers with our own dataset. The database images are classified by using the pretrained network to identifythepresenceofapersoninthe crowd. Further, the position of the person should be obtained and provide live tracking with the KLT algorithm. 1. INTRODUCTION Identification of a missing person in the crowded scene such as festivals,temples,publicplaces,andmeetingsisa complextask. A solution with the help of deep learning concept is to be proposed. Pretrained Convolutional Neural Network (CNN) is employed for the identification of a person. The Convolutional Neural Network indicates that the network employs a mathematical operation called convolution. Convolution is a specialized kind of linear operation. Convolutional networks are simply neural networks that use convolution in place of general matrix multiplication in at least one of their layers. AlexNet contained eight layers, the first five were convolutional layers, some of them followed bymax-poolinglayers,andthelastthree were fully connected layers. Itusedthe non-saturatingReLUactivationfunction,whichshowedimprovedtraining performance over tanh and sigmoid. The layers are trained based on our requirement to identify the particular object. The last convolution layer and fully connected layers are configured to our requirement. 2. LITERATURE SURVEY The idea forthe project was erected from the papers thatwere glimpsedislistedbelow.However,anoriginalplanindesignuses some differential way from the papers mentioned below. A. Agarwal and B. Triggs[1] proposed a new technique to extract the features of the image. By the new technique, morefeaturesareextractedcomparedtotheoldtechnique.Thenewtechniqueused multilevel image coding for extracting the features of images. X. Cao, Y. Wei, F. Wen, and J. Sun [2] developed an approach that says replacing the old technique for aligning the shape of the face with aligning the face by explicit shape regression. Using this aligning the shape of the face becomes very easy. D. Chen, S. Ren, Y. Wei, X. Cao, and J. Sun [3] proposed a new technique for the detection and alignment of face. In this technique detection and alignment of the face done in one processbecausejointcascade is used for the detection and alignment of face. D. Eigen and R. Fergus [4] proposed that Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture. P. Felzenszwalb, R. Girshick, D. McAllester, and D.Ramanan [5] proposed anewmodelthatcanable to detect the object i.eObjectdetection with discriminatively trained part-basedmodels.R.Girshick,J.Donahue,T.Darrell,andJ. Malik [6] Extracting the features is very important so that they proposed to extract more features from the image sothatthey highly focused the rich feature hierarchies for accurate object detection and semantic segmentation.N. Hu, W. Huang, and S. Ranganath [7] proposed that by the position of the head in the image it is very easy to embed and map the image so they proposed an estimation for the position of head by non-linear embedding and mapping. Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell [8] proposed a new architecture for the embedding process because the old architecture is not working properly to embed because of fast feature embedding so that they proposed a new Convolutional architecture for fast feature embedding. A. Kumar, R. Ranjan, V. Patel, and R. Chellappa [9] proposed a method which aligns the face with the help of regression. L. Liang, R. Xiao, F. Wen, and J. Sun [10] developed an approach that says replacing the old technique for aligning the shape of the face
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072 © 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 391 with a component-based discriminative search. J. Long, E. Shelhamer, and T. Darrell [11] proposed that Fully convolutional networks forsemantic segmentation. Q. Zhao, S. S. Ge, M. Ye, S. Liu,and W. He, [12] proposed that Learning saliency features for face detection and recognition using the multi-task network. M. D. Zeiler and R. Fergus[13] In this proposed system it says how to understand the convolutional network and how to visualizeit. Y. Gong, L.Wang, R. Guo, and S. Lazebnik, [14] proposed that Multi- scale orderless pooling of deep convolutional activation features. A. Babenko, A. Slesarev, A. Chigorin, and V. Lempitsky [15] proposed a new method to retrieve the image so that the original image can be retrieved so that some new codes are proposed for the image retrieval i.e. Neural codes for image retrieval.A. Eitel, J. T. Springenberg, L. Spinello, M. Riedmiller, and W. Burgard [16] proposed that Multi modal deep learning for robust RGB-D object recognition. 3. PROPOSED WORK The aim of this project is to identify the person who is missing in the crowded area using the drone which is having the camera and by this the person who ismissed in the crowd will be easily identified within a minute. This will be very usefulforthepolice to find the missing person. Nowadays many children are missing in the festival crowd so that by this project the police can monitor from one place and then they can able to find the person and also they can live track the person where he is moving so that they can save the time and can able to identify the missing person veryeasily Fig – 1 Block Diagram of Proposed Work First, the images of missing persons arecollected.Eachandeveryimageshouldbefromdifferentanglessothatidentificationwill be so simple. After collecting the images a dataset for the images should be created. Then it will be under the training process. For this process, AlexNet is used, which is a pretrained neural network. Then the camera which is attached on the drone will capture the video of the crowded area and it will be communicating with the router . In this project IP camera is used , it will be communicating with the help of the IP Address so that router is used to communicate. Then it will be given to the controller so that the captured image will undergo the cropping process. After cropping the image a database for the cropped image will be created. After the database is created the database will be used to train the network. After completing the training process testing is done in a classifier having only a softmax layer and if the image is detected means in the display the image will be shown.
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072 © 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 392 4. RESULTS AND DISCUSSION Fig - 2 Detection of the target with bounding box The test image was given to the proposed system .Thepretrained systemtrainedwithtargetsregionofinterestwoulddetectthe presence of target in given test image and mapping the target region with bounding box based on the highest score obtained from detection score mechanism of system .The detection score improves the confidence of the target to be detected and displayed in monitor. For further reference the system detects all the faces in the test images and crop their faces and store in database for manual checking process to improve the detection probability. 5. CONCLUSION The Identification of Missing Person in the Crowd scene is one of the most important needstoensuresurveillanceandsafetyof the people all over places. But unfortunately, the existing technology cannot provide sufficient information to find the target person in the crowd. In order to rectify the deficiency in existing system and to achieve the goal the proposed model is employed and ensure the higher probability of detection by including various automatic as well as manual method. REFERENCES 1. X. Cao, Y. Wei, F. Wen, and J. Sun, “Face alignment by explicit shape regression,” Int. J. Comput. Vis., vol. 107no.2,pp. 177–190, 2. Agarwal and B. Triggs, “Multilevel image coding with hyperfeatures,” Int. J. Comput. Vis., vol. 78, pp. 15–27, 2008. 3. D. Chen, S. Ren, Y. Wei, X. Cao, and J. Sun, “Joint cascade face detection and alignment,” in Proc. Eur. Conf. Comput. Vis., 2014, vol. 8694, pp. 109–122. 4. D. Eigen and R. Fergus, “Predicting depth, surface normal and semantic labels with a common multi-scale convolutional architecture,” in Proc. IEEE Int. Conf. Comput. Vis., 2015, pp. 2650–2658. 5. P. Felzenszwalb, R. Girshick, D. McAllester, and D. Ramanan, “Object detection with discriminatively trained part- based models,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 32 no. 9, pp. 1627–1645, Sep.2010. 6. R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2014,pp. 580–587. 7. N. Hu, W. Huang, and S. Ranganath, “Head pose estimation by non-linear embeddingandmapping,”inProc.IEEEInt. Conf.Image Process., Sep. 2005, vol. 2, pp. II–342–5. 8. Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell, “Caffe: Convolutional architecture for fast feature embedding,” in Proc. 22nd ACM Int. Conf. Multimedia,2014, pp. 675–678. 9. Kumar, R. Ranjan, V. Patel, and R. Chellappa, “Face alignment by local deep descriptor regression,” arXiv preprint arXiv: 1601.07950, 2016.
- 4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 07 Issue: 02 | Feb 2020 www.irjet.net p-ISSN: 2395-0072 © 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 393 10. L. Liang, R. Xiao, F. Wen, and J. Sun, “Face alignment via component-baseddiscriminativesearch,”inProc.Eur.Conf. Comput. Vis., 2008, vol. 5303. 11. J. Long, E. Shelhamer, and T. Darrell, ``Fully convolutional networks for semanticsegmentation,''in Proc.IEEEConf. Comput. Vis. Pattern Recognit., 2015, pp. 3431_3440. 12. Q. Zhao, S. S. Ge, M. Ye, S. Liu, and W. He, ``Learning saliency features for face detection and recognition using the multi-task network,'' Int. J. Social Robot., vol. 8, no. 5, pp. 709_720,2016. 13. M. D. Zeiler and R. Fergus, ``Visualizing and understanding convolutional networks,''inProc.Eur.Conf.Comput.Vis., Springer, 2014, pp. 818_833. 14. Y. Gong, L.Wang, R. Guo, and S. Lazebnik, ``Multi-scale orderless pooling of deep convolutional activationfeatures,'' in Proc. Eur. Conf. Comput. Vis., 2014, pp. 392_407. 15. Babenko and V. Lempitsky, ``Aggregating local deep features for image retrieval,'' in Proc. IEEE Int. Conf. Comput. Vis., Dec. 2015, pp. 1269_1277. 16. Eitel, J. T. Springenberg, L. Spinello, M. Riedmiller, and W. Burgard, ``Multimodal deep learning for robust RGB- D object recognition,'' in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst., Sep./Oct. 2015, pp. 681_687. 17. https://www.mathworks.com/help/matlab 18. https://www.mathworks.com/help/vision/examples/object-detection-using-faster-r-cnn-deep-learning.html 19. https://www.mathworks.com/help/deeplearning/ref/alexnet.html 20. https://www.mathworks.com/help/vision/object-detection-using-deep- learning.html?s_tid=CRUX_lftnav