HARAKATDAGI VA SHOVQINLI MUHITLARDA YUZNI ANIQLASH ANIQLIGINI OSHIRISHNING GIBRID YONDASHUVI
Keywords:
yuzni aniqlash, harakat xiralashuvi, shovqin filtrlash, optik oqim, gibrid model, RetinaFace, Wiener filtri, multi-scale detektor, real vaqt qayta ishlash.Abstract
Ushbu maqolada harakatdagi va shovqinli muhitlarda yuzni aniqlash aniqligini oshirishga qaratilgan gibrid yondashuv taqdim etiladi. Tadqiqotning dolzarbligi shundaki, real vaqtdagi video oqimlarida harakat xiralashuvi (motion blur), Gaussiy va tuz-qalampir shovqinlari, hamda ob'ektning tez harakati kabi omillar klassik yuzni aniqlash algoritmlarining samaradorligini 30–45% ga pasaytiradi. Muammoni hal etish uchun o'tib ketish kompensatsiyasi (optical flow), adaptiv shovqin filtrlash va ko'p miqyosli (multi-scale) neyron tarmoqni o'z ichiga olgan uch bosqichli gibrid konveyer taklif etiladi. LFW-Video va WIDER FACE ma'lumotlar to'plamlarida o'tkazilgan tajribalar tizimning 93.8% aniqlikka erishganini va harakatdagi ob'yektlar uchun sezgirlik 91.2% ekanligini ko'rsatdi. Ilmiy yangilik — optik oqim va adaptiv Wiener filtrlashni chuqur o'rganish arxitekturasi bilan birlashtirishdan iborat. Amaliy qo'llanish sohalari: transport xavfsizligi, ommaviy tadbirlar monitoringi va video kuzatuv tizimlari.
References
[1] Zafeiriou S., Zhang C., Zhang Z. A survey on face detection in the wild: past, present and future // Computer Vision and Image Understanding. — 2015. — Vol. 138. — P. 1–24. DOI: 10.1016/j.cviu.2015.03.015
[2] Minaee S. et al. Going deeper into face detection: A survey // arXiv:2103.14983. — 2021.
[3] Fergus R., Singh B., Hertzmann A., Roweis S., Freeman W. Removing camera shake from a single photograph // ACM Trans. Graphics. — 2006. — Vol. 25, No. 3. — P. 787–794. DOI: 10.1145/1141911.1141956
[4] Buades A., Coll B., Morel J.M. A non-local algorithm for image denoising // Proc. IEEE CVPR. — 2005. — Vol. 2. — P. 60–65. DOI: 10.1109/CVPR.2005.38
[5] Jiang H., Learned-Miller E. Face detection with the faster R-CNN // Proc. IEEE FG. — 2017. — P. 650–657. DOI: 10.1109/FG.2017.82
[6] Viola P., Jones M. Rapid object detection using a boosted cascade of simple features // Proc. IEEE CVPR. — 2001. — Vol. 1. — P. 511–518. DOI: 10.1109/CVPR.2001.990517
[7] Dalal N., Triggs B. Histograms of oriented gradients for human detection // Proc. IEEE CVPR. — 2005. — Vol. 1. — P. 886–893. DOI: 10.1109/CVPR.2005.177
[8] Krishnan D., Fergus R. Fast image deconvolution using hyper-Laplacian priors // Proc. NeurIPS. — 2009. — Vol. 22. — P. 1033–1041.
[9] Cho S., Lee S. Fast motion deblurring // ACM Trans. Graphics. — 2009. — Vol. 28, No. 5. — P. 1–8. DOI: 10.1145/1618452.1618491
[10] Horn B.K.P., Schunck B.G. Determining optical flow // Artificial Intelligence. — 1981. — Vol. 17, No. 1–3. — P. 185–203. DOI: 10.1016/0004-3702(81)90024-2
[11] Zhang K., Zhang Z., Li Z., Qiao Y. Joint face detection and alignment using multitask cascaded convolutional networks // IEEE Signal Process. Lett. — 2016. — Vol. 23, No. 10. — P. 1499–1503. DOI: 10.1109/LSP.2016.2603342
[12] Deng J., Guo J., Ververas E., Kotsia I., Zafeiriou S. RetinaFace: Single-shot multi-level face localisation in the wild // Proc. IEEE CVPR. — 2020. — P. 5203–5212. DOI: 10.1109/CVPR42600.2020.00525
[13] Nada H., Sindagi V.A., Zhang H., Patel V.M. Pushing the limits of unconstrained face detection // Proc. IEEE BTAS. — 2018. — P. 1–9. DOI: 10.1109/BTAS.2018.8698561
[14] Yang S. et al. DSFD: Dual shot face detector // Proc. IEEE CVPR. — 2019. — P. 5060–5069. DOI: 10.1109/CVPR.2019.00520
