A. A. Yuzhakov – Dr.Sc. (Eng.), Professor, Head of Department of Automatics and Telemechanics, Perm National Research Polytechnic University

E-mail: uz@at.pstu.ru

A. N. Kokoulin – Ph.D. (Eng.), Associate Professor, Department of Automatics and Telemechanics, Perm National Research Polytechnic University

E-mail: a.n.kokoulin@at.pstu.ru

A. I. Tur – Post-graduate Student, Assistant, Department of Automatics and Telemechanics, Perm National Research Polytechnic University

Face recognition methods are a subset of object recognition methods, and play an important role in various real-time applications, such as video surveillance and face identification systems, smart cameras, intelligent robots. The question of studying the universal method of recognition for various individuals, lighting conditions, emotions and postures of a person remains relevant.

The main disadvantage of the traditional approach, which uses a centralized recognition system scheme, is the high load on the data network and the computational load on the central processing server: the more CCTV cameras the central unit serves, the greater its load. But most of the time, the video stream does not contain images of individuals, and computing resources are wasted. The basic principle of our recognition system is a distributed hierarchical processing network, using the rough-to-detailed research paradigm. Each source video stream is processed in place by a tiny SoC computer that acts as an Edge computer and detects the presence of fragments of faces in a video frame and crops the region of interest (ROI). The resulting stream, including ROI, is transmitted to the main server if a face is detected. In our work, we propose a distributed hierarchical processing network using the coarse-to-fine research paradigm and perform three stages of recognizing a human face in a distributed system:

stage 1 (FD), the task of face detection performed on the Edge node (not including recognition), when the ROI boundaries are selected and the images of faces or human figures are aligned; stage 2 (FE), when we normalize and improve a limited ROI on Edge node and send it as a series of images or a video stream to the main server; stage 3 (FR), recognition when the main server performs face recognition using a multi-CNN cascade scheme and central database.

According to our modeling results, traffic will be decreased by 80–90%. In addition, we offer the concept of a CNN cascade system of facial recognition, which can make recognition and identification more accurate through soft decision-making.

1. Kokoulin А.N., Tur A.I., Yuzhakov A.A. Convolutional neural networks application in plastic waste recognition and sorting. Proceedings of the 2018 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering. 2018. P. 1094–1098. DOI: 10.1109/EIConRus.2018.8317281.
2. Viola P., Jones M.J., Snow D. Detecting pedestrians using patterns of motion and appearance. The 9th ICCV. Nice, France. 2003. V. 1. P. 734–741.
3. Tang J., Deng C., Huang G.B., Zhao B. Compressed-domain ship detection on spaceborne optical image using deep neural network and extreme learning machine. IEEE Transactions on Geoscience and Remote Sensing. 2015. V. 53. P. 1174–1185.
4. Su C.Y., Yang J.F. Histogram of gradient phases: A new local descriptor for face recognition. IET Computer Vision. 2014. V. 8. № 6. P. 556–567.
5. Ke Y., Sukthankar R. PCA-SIFT: A more distinctive representation for local image descriptors. IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington, DC, USA. 2004. P. 66–75.
6. Dalal N., Triggs B. Histograms of oriented gradients for human detection. IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, USA. 2005. V. 1. P. 886–893.
7. Lowe D.G. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision. 2004. V. 60. № 2. P. 91–110.
8. Zhang X., Gonnot Th., Saniie J. Real-time face detection and recognition in complex background // Journal of Signal and Information Processing. 2017. V. 8. № 2. P. 99–112.
9. Kokoulin A.N. Distributed storage approach in content delivery networks. Proceedings of 13th International Scientific-Technical Conference on Actual Problems of Electronic Instrument Engineering. 2016. P. 479–484. DOI: 10.1109/APEIE.2016.7806398.
10. Kulikov A.A., Demkin D.V., Melkov A.E. The analysis of the maximum face compression impact on face recognition result. Science Prospects. 2014. № 3 (54). P. 104–108.
11. Kokoulin A.N., Yuzhakov A.A., Kiryanov D.A. Scalable distributed storage for big scientific data. Proceedings of the 2018 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering. 2018. P. 1099–1103.
12. Kokoulin A. Methods for large image distributed processing and storage. IEEE EuroCon 2013. 2013. P. 1606–1610. DOI: 10.1109/ EUROCON.2013.6625191.
13. Fleuret F., Gemand D. Coarse-to-fine face detection. International Journal of Computer Vision. 2001. V. 41. № 1–2. P. 85–107.