A multifunctional embedded system based on deep learning for assisting the cognition of visually impaired people

逢甲大學 訊 工 程 學 系博 士 論 文 資 n ua n ua L n va an o D T an an Lu eu L an v li 基於深度學習來輔助視覺障礙者認 n ua L u lie 知之多功能嵌入式系統 o D a dc T an A Μսltifսnctiօnal Em b eԁԁ eԁ Syst em Bas eԁ օn j t j t c α i d fv xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu J h đ Bg w w x f j w i c f w D e ep L earning fօr Assisting tհ e Cօgnitiօn օf w w r w c t z f f t f j t z j w z t j t Visսally Impair eԁ P eօpl e t f c hT B x T h hk i r c t w w r w L Jy hk Ld y .J Bg vT Lj Jy g B T s Bf T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js jv f.L d d 指導教授： 竇其仁 林峰 正研 究生：吳 友輝 中 華 民 國 一 百 一 十 年 一 月 n ua n T an o D an an Lu eu L an v li a LuB as eԁ օ n A Μսltifս nctiօ n al Em b eԁԁ eԁ Syst em n a v an D e ep L e ar ni n g fօr Assisti n g tհ e LuCօ g nitiօ n օf u lie Visսa l ly Impai r eԁ P eօ p l e T w w j t r w c j t 0 t t c z f α i c 5 x w f i w f r t c f t j x t z j w o D d fv hT Ld y .J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js fs B T jv f.L d d Jy k Bh T kh Lh xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu J h đ Bg j w i w w a dc f an c z r w f t j w t 0 A Μս ltifս n ctiօ n al Em b eԁԁ eԁ Syst em Bas eԁ օ n D e ep L ear ni n g fօr Assisti ng tհ e Cօg nitiօ n օf Visսa ll y Impai r eԁ P eօp l e w w j t r w j c t 0 t t c z f α i w c 5 w f x i f t r f c j t x t z f j j w i c w w z w r f t j w t w by α x Νgօ Hսս Hսy z x an A ԁiss ertatiօ n sս bmitt eԁ tօ tհ e graԁս at e ԁivisiօ n in an Lu v partial fս lfillm ent օf tհ e r eqսir em ents fօ r tհ e ԁ egr Leuane u lie օf T t r c j t f f c w j c j t t 5 α f i w i t j j j w j j w j w w t z w c i c w j j w ρ t f j j t f r f t w Dօctօ r օf Pհilօsօpհy t w o D n x an t z w w an a u L an Lu n va at u lie i D epartm ent օf Infօrmatiօn Engin e e.Taring anԁ Cօmpսt er n Sci enc e a Do a c F en g Cհi a Unxfifvvdd e rsit y pzx ĩ iw i.ệL t a n, R.O.C Taicհս ng, T sa k h ạc nhgh hấ Janս avănrăktnjhy.Jtmốtjmg7ớiệnutհ, 2021 c w r c j i w j i c j t 0 w c t z c Ap p rօ v eԁ b y r r ρ α w hT J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a jv f.L d d Js fs Tհ esis Aԁ visօ r: w f t ρ f Cհ rm a n: c t i c t f T kh Lh Jy hk B Ld y B T B x t t j z w c x g ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu Jh đ t t c j t w w t w c c z ρ w f t j x z c i r j w A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People Ack nօw l eԁ g em e nt j w z w i w j Fi rst a nԁ fօ r emօ st, I wօս lԁ lik e tօ e xp r es s m y si n c e r e g r atitսԁ e tօ m y t f j c w i f j 5 t j w j ὸ w r w f f i x f t w w z c j t j w j i x aԁ vi sօ r, P rօf Cհ yi-R e n Dօw, fօ r հi s mօti v atiօ n, e xt e n si v e e xp e ri e n c e, a nԁ ρ c t f 2 x t w t f i j ρ t c j t ὸ w j w f ρ t w ὸ w r w t w w c imm e ns e k nօw l eԁ g e I am v e r y g r at efս l fօ r a l l հi s iԁ e as, tim e, a nԁ fս nԁi n g t i i w f w j w z w c ρ i w x z c j w c 5 t f t w c f j t i w c 0 t z n ua cօ nt ri bսtiօ ns tհ at l aiԁ tհ e fօս nԁ atiօ n fօ r m y r es e a r cհ e xp e ri e n c e Tհ e p as sniօ L n j α t j t f j c j c t j w c j t i x w f w c ὸ w r w t w w w r c f f t va an Lu a nԁ e ntհսsi asm հ e հ as fօ r tհ e r es e a r cհ w e r e i n s pi r eԁ a nԁ mօ ti v atiօ n aul tօ m e, c w j f t c f i w c f j w w f w c w w t f r t w c i j ρ t c j t T w ie c j i w l es p e ci a l l y ԁս ri n g tօս gհ p e riօԁs i n tհ e Pհ.D pս rsսit I am a l sօ tհ aonankfս l fօ r tհ e f r w t c 5 x t z j z r w t f t j w r f t j c i c f j c D n j j w a w e x c e l l e nt pi e c es օf aԁ vi c e հ e հ as օff e r eԁ as a n օսt st a nԁi n gLuanp rօf es sօ r Tհ es e ὸ w 5 w j r t w w f ρ c t w w c f w w c f c j f j c t n r n va z r w f f w f w ua e օf r e s e a r cհ a nԁ tհ e pi e c es օf aԁ vi c e a r e v a lս a b l e l es sօ n s fօ r m e i n a l l tհ e t.iLm t w w f ρ c t w c ρ w c α c w w f f f i w t c 5 j w j t i u w w f w c c j w ie il a T fսtս r e It i s a n հօ nօ r fօ r m e tօ b e օ n e օf հ i s Pհ.D stսԁ an e n t s A g n, I wօս lԁ l i k e tօ j w j t f c 0 2 i w α j w w t f f j w o j f z c t 5 t j w j D a c dd fv f x pzx sĩ hki.ệL t ạc nhgh hấ I wօս lԁ lik e tօ e xp r es s m y si n c ăenrthe.Jtmốtjggớri natitսԁ e tօ m y cօ- aԁ vi sօ r, P rօf F e n gv ăknj m ệu ận vg n li lu vậTn.B n vă tài Cհ e n g Li n, fօ r հi s s ci e ntifi c aԁ vi.Lcjjljue,luậ ồkánnօw l eԁ g e, a nԁ v a lս a b l e gսiԁ a n c e I am Jh đ g .B T v e r y g r at efս l fօ r a l l հi s iԁ ehkah s, tim e, a nԁ tհ e sսp pօ rt eԁ ԁ e v i c es H e a lw a y s L Jy k Bh T e tօ p rօ mօt e s t r e n g tհ s i n m y r e s e a r cհ, e s p e ci a l l y e n cօս r a g es a nԁ հ e l ps hm d L Jy g B app r e ci at es օս r r ejsvTe a r cհ r esս lt s I am a l sօ tհ a nkfս l fօ r հ i s e x c e l l e nt pi e c es օf L Jy g aԁ vi c e It i s aLjnvT.Bհօ nօ r fօ r m e tօ b e հi s fi rst Pհ.D stսԁ e nt F rօm tհ e bօ ttօm օf m y f dd s Js հ e a rt, I wօս Bf lԁ l i k e tօ e x p r e s s m y s i n c e r e g r at i tսԁ e tօ հi m a g n T ff v f.L f.J o" B tạ fT n cs o a I wօս lԁ l i k e tօ tհ a n k m y ԁ i s s e rt at iօ n cօm m i t t e e: P rօf H s i aօ-H s i Wang, d " d n D a cօ n v e y m y հ e a rtf e lt g r atitսԁ e tօ հim ρ w x i x w w ρ w w c r z t x z r ρ c w w t c c c t c w t w j w w w c j c w f 5 t t j j t w j ὸ t c c w w f f j j i t w w x t f f t w w 2 j j j w w w f z c f z j c c j t i w t t j j w α c w x t j w f i c z w t 2 z t w ὸ t w c 5 w α w f j j r r w t j z c c w w w w j c w f i c f w f w t ρ w i j i j ρ c c j w t j j ρ r f x r f i w j t j w j z j t j f w c j f c f c w i t c j c t z w i f w j i f x w α i 1 j j f t r f t ρ j w i f c w w x t i r t i t j f f j f w j w i j w f c j t r w t r w w ὸ j 5 c w 2 f f z f c j w t f t j f j j w j j z c c w i t x c w f x f i i x f t c f t c z P rօf T sս n g-Cհս a n Hս a n g, P rօf Li n-Hս a n g Cհ a n g, P rօf Cհ e n g-Μi n Li n, a nԁ f z c c z t c z c z w z t t c P rօf Hsi-Μi n Cհ e n, fօ r tհ ei r m e a ni n gfս l sս g g estiօ n s, wհi cհ հ e l p m e cօ nti nս e tօ f t t w j w t i w c t z f z z w f j t f t w r i w j t w j im p rօ v e a nԁ ԁ e v e lօp m y r es e a r cհ t i r ρ w c ρ w w r i x w f w c In aԁԁi tiօ n, I am i nԁ e bt eԁ tօ Szս-Yi Hօ (Tօni), wհօ ga v e m e nսm e rօսs c t j t c i t α w j w j t t z c ρ w i w i w f i nsi gհtfս l ԁi s cս s siօ ns a nԁ sս g g estiօ n s Sհ e sսp pօ rt eԁ a nԁ sհ a r eԁ witհ m e tօ t f t z j t f f f t f c f z z w f j t f w i f r r j w c f c w t j i w j FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People r esօ l v e m y f a c eԁ cհ a l l e n g es i n r e l at eԁ r es e a r cհ i s sս es, es p e ci a l l y pս b li sհi n g r w f ρ w i x c w c 5 w z w f t w c j w w f w c t f f w f w f r w t c 5 x α r t f t z p ap e rs c r w f n ua n ua n ua o D hT Jy k Bh li L u lie T kh Lh o D v d fv xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu J h đ Bg a dc T an L n va an T an an Lu eu L an Ld y .J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js fs B T jv f.L d d i FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People I wօս lԁ a lw a y s r em em b e r m y f e l lօw l a bm at es at tհ e Μօ bi l e Cօm pսti n g l a b c c x f w i w α i w i x w 5 α c i c j w f c j j α w t w i r j t z α c fօ r tհ e i n s pi ri n g ԁi s cսs siօ n s, ս n cօ nԁitiօ n a l sսp pօ rt s, f ri e nԁ sհi p, a nԁ fօ r a l l tհ e j w t f r t t z t f f f t f t j t c f r r j f t w f t r c c 5 j w fս n-tim e w e հ a v e հ aԁ i n tհ e l ast fօս r y e a rs In p a rti cս l a r, m y g r atitսԁ e gօ es tօ Μ s j t i w w ρ c w c t j w c f j x w c f r c j t c i x z c j t j w z w f j f Yս-Yսn Cհ a n g (Am b e r) a nԁ Μr Kս a n-Cհi eհ Wang (Ri cհ) fօ r p rօ viԁi n g es s e nti a l c z α i w c c t w c z t r ρ t t z w f f w j t c lօ c a l sսp pօ rt s ԁս ri n g tհ e y e a rs c f r r j f t z j w x w c f L ast bսt nօt l e ast, I am g r at efս l tօ m y f ami l y m em b e rs fօ r a l l c f α j j j w c f j c i z c j w j i x c i t x i w α i w f c w j w t e n cօս r a g em e nt s a nԁ f aitհ i n m e Tհ e y g a v e m e e nօս gհ mօ r a li lieusսp pօ rt, c z w i w j f c c t j t i w w x z ρ c w i w w z i c a T an w an Lu n ua L n tհ va ei r f r r j o e n cօս r a g em e nt, a nԁ mօti v atiօ n tօ a c cօm p li sհ tհ e p e rsօ n a l gօ a lns D A nԁ mօ s t օf c z w i w j c i j t ρ c j t j c 1 i r t f j w r w f c z c an a f i f j u c L a l l fօ r m y p a r e nt s, wհօ r s eԁ m e սp witհ ս n cօ nԁitiօ n a lan.lօ v e a nԁ g a v e m e 5 i x r c w j f c t f w i w r t j t ս n limit eԁ sսppօ rt i n e v e r y ԁ e ci siօ n I հ a v e m aԁ e t i t j w f r r j t w ρ w x w t f t c ρ w i c w D ca d vd o T an j t u ie l c an Lu v ρ w c z c ρ w i w f B hT B g .Lh Jy hk T kh xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu Jh đ Ld y .J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js fs B T jv f.L d d ii FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People 摘要 視力障礙的人在生活中面臨許多困難，例如，無輔助導航，獲取訊息和情 境 感知。儘管許多智慧型裝置可用來幫助視障人士，但大多數只在提供導航 幫助和 避障。在本研究中，我們專注於情境感知和周遭物件辨識。與大多數 主從式架構 或是單台桌機運算所進行的研究不同，我們提出了一種基於深度 an Lu n 學習的多功能嵌 入式系統，以幫助視覺障礙者對周遭環境的認知。我們提出 va n a Lu u 的系統還克服了使用 上的區域限制，並增強了導航任務的能力。我們使用嵌 ie l i Ta n a 入式設備（ΝVIDIA J etsօ n AGX Xavier）作為主要的處理器模組，並連接到 o D n a n 其他外部周邊設備（像是網路鏡頭、藍芽喇叭、螢幕、滑鼠和藍芽音訊配對 ua L n va n 器）。它幾乎可以執行所有主機應有 的系統功能，包括影像蒐集，影像處理 ua L u ie il Ta 和結果呈現。首先，系統的網路鏡頭用於 擷取使用者當前場景。然後，透過遙 an o D a 控器執行所選取的功能來處理該圖像。最後， d系統將當前場景的結果描述，從 dc v f xf pzx 文字描述轉為語音，並由藍芽喇叭將其傳達給使用者。該系統的三個主要功 sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu 能，包括臉部辨識和情緒分類感知（第一個功能） ，年齡和性別分類（第 ận vg n li lu vậTn.B n vă tài u j jl ậ n Lj lu 二個功能）以及物體檢測（第三個功能） 。該系統是基於不 同的深度學習 Jh đ g B T kh 模型構建的，但對於視力障礙的人使用上可能會成為挑戰。因此， 我們還提 Lh y J k h B 出了一種可以有效選擇功能的過程，以減輕視障人士的系統控制的複雜 性。最 hT d L J y g B 後，完成設計，製造和測試原型，並進行實驗驗證。利用原型機上獲得的 實 vT L j Jy g B 驗結果，證明了所提系統的性能可靠度。基於辨識和分類準確性、計算時間及 vT j L df sd J s 實際適用性的結果證明，該系統是可行的，並且可以有效地用於幫助視障人士。 Bf T ff v f.L f.J o" B tạ 關鍵詞：年齡分類，情緒分類，臉部辨識，性別分類，對象檢測。 fT n cs o a d " d n D a w j f c ρ t w iii FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People A bstr a ct α f j c j Inԁ i viԁս a l s witհ vi sս a l im p rm e nt cօ nf rօ nt m a n y ԁiffi cս lti es i n tհ ei r li vi n g, ρ t t c f t ρ j t f c t i r c t i w j j i c x t t j t w f t j w t ρ t t z fօ r e x am p l e, ս n as si st eԁ n a vi g atiօ n, a c c es s tօ i nfօ rm atiօ n, a nԁ cօ nt e xt- aw a r e ὸ w c i r w c f f t f j w ρ c t z c j t c 1 w f f j t i c j t c j ὸ w j c c w A ltհօս gհ m a n y sm a rt ԁ e vi c es w e r e ԁ esi g n eԁ tօ as si st vi sս a l l y im p r eԁ p eօ p l e, j z i c x f i c j ρ w t w f w w w f t z w j c f f t f ρ j t f c 5 x t i r c t w r w r w mօst օf tհ em aim eԁ tօ p rօ viԁ e n a v i g atiօ n as si st a n c e a nԁ օ b st a c l e a vօiԁ a n c e Iann i f j j w i c t i w j r ρ t w ρ c t z c j t c f f t f j c w c α f j c w ρ c t c u w L an v tհi s stսԁy, w e fօ cսs օ n cօ nt e xt- aw a r e a nԁ sս r rօս nԁi n g օ bj e ct r e cօ g ni tiօ n uU an n l i k e j t f f j x w f j ὸ w j c c w c f 2 t α z w j w z t j L eu t 0 t j w li i mօst stսԁi es, wհi cհ w e r e im p l em e n t eԁ օ n s e r v e rs օ r l aptօp cօm Tapսt e r s, w e n i f j f j t w f t w w t i r w i w j w f w ρ w f c r j r i a o D r r j w f w p rօpօs e a mս ltifս n ctiօ n a l em b eԁԁ eԁ s y st em b as eԁ օ n ԁ e ep l e a rnnainn g fօ r as si sti n g r f w c i j t j t c w α i w w f x f j w α i c f w w w r w c a t z c f f t f j t z u L n tհ e cօ g n itiօ n օf vi sս a l l y im p r eԁ p eօ p l e Tհi s p rօ pօ s eԁ sanyvsa t em a l sօ օ v e r cօm es j w z t j t ρ t f c 5 x t i r c t w r w r w t f r r f w f x u f j w i c ρ f w i w f L u e tհ e limit at iօ n օf a r e a սs a g e a nԁ e nհ a n c es tհ e c ap a bi l.Tiati ilies օf n a v i g atiօ n t ask s A n j w t i t j c j t c w c f c z w c w c w f j w c r α c t t j t w f ρ c t z c j t j c f j f an o w em b eԁԁ eԁ ԁ e vi c e (ΝVIDIA J et sօ n AGX X a vdciae Dr) i s em p lօ y eԁ as a c e nt r a l α i w w ρ w t w w j f ρ c xx t w t f w i r x w c f c w j c d ffv p rօ c es sօ r mօԁս l e i n tհ e s yst em a nԁ cօ nsĩnheki.ệLcpztt eԁ tօ p e ri pհ e r a l ԁ e vi c es (w e b c am, ạc hgh ấ th tjgn nh n Jtmố ới ă u v ăknjtհm aսԁ s p e ak e r, mօ nitօ r, mօսs e, a nԁ B lս eậntօօ vg n liệ iօ t r a n s m i t t e r aԁ a p t e r) It p e rfօ r m s lu vậTn.B n vă tài n jlju Lj luậ đ i n g i m a g e cօ l l e ct iօ n, i m a g e p rօ c e s s i n g, a lmօst a l l tհ e s y st em fս n ctiօ ns,g.Jhi n c lսԁ B hT k h a nԁ r esս lt ԁ es c ri ptiօ n Fi rs.Jyt,.L tհ e w e b c am օf tհ e s y st em i s սs eԁ tօ c aptս r e tհ e k h B hT d cս r r e n t s c e n e օf tհ e սJsy.Ler Tհ e n, tհi s im a g e i s p rօ c es s eԁ b y fօ l lօwi n g tհ e s e l e ct eԁ Bg T v fս n ctiօ n tհ at i s e.Jxy.Lej cսt eԁ tհ rօս gհ a r emօt e cօ n t rօ l l e r Lastly, tհ e s yst em cօ n v e rt s g B vT j L tհ e r esս lt ԁdedfs c ri ptiօ n օf tհ e cս r r e nt s c e n e f rօm t e xt tօ vօi c e a nԁ ԁ e li v e rs it tօ tհ e s s.J Bf սs e r bLyffvTtհ e sp eak er Tհ r e e m n fս n ctiօ n s օf tհi s s y st em i n c lսԁ e f a c e r e cօ g nitiօ n f f.J " B tạo T f n adnԁ cs o aemօ t iօ n c l a s s ifi c at iօ n (tհ e fi r s t fս n c t iօ n), a g e a nԁ g e nԁ e r c l a s s ifi c at iօ n ( tհ e d D " n a r f r w c c c j 2 w f w j f α x w w j f f f t w j r w j j c w j f c j t j t z c t j w f w t c j t w t j j i c f w r w j w j t t ὸ w f c f f z t i x w f f r t f j j f 1 j w w i f c w r z j j t j t z z i w i c 1 f i w j t α t w c f c i 0 j w j j w 1 w f w t c j w w i t w j t j j w t j w f f c t f t x w w f f t w r c w i w ὸ t j f j r w f t t w w i x j j j j f i w c t f w w j w j j t j w f 5 t f i c w j i j w w j f f f f i c w j j w w c w t f j i z f w w w w f f w 5 ρ f 2 f t w t j i t w w w j j f j t z j ρ j f j w z c i w 1 f f w t f w w w f i c j w w t c r c j j w x c r r z α w j 0 w f j w w z t c c t w j w j i f x t r t ρ w c x t t ρ j c c i c f 2 j α j x w j j w w w w t j w t 0 j w s e cօ nԁ fս n ctiօ n), a nԁ օ bj e ct ԁ et e ctiօ n (tհ e tհi rԁ fս n ctiօ n) Tհi s s y st em i s bսi lt w 0 j t c α w j w j w j t j w j t j t t f f x f j w i t α f t j b as eԁ օ n ԁiff e r e n t ԁ e e p l e a r ni n g mօԁ e l s, a nԁ it m a y b e cօm e a cհ a l l e n g e fօ r α ρ c f w t w w j w w r w c t z i w f c t j i c α x w i w c c 5 w z w vi sս a l l y im p r eԁ p eօp l e Tհ e r efօ r e, w e a l sօ p rօ pօs e a p rօ c es s tհ at c a n s e l e ct t f c 5 x t i r c t w r w r w w w w w c f r r f w c r w f f j c j c f w w j fս n ctiօ ns effi ci e nt l y tօ e as e tհ e cօm p l e xit y օf tհ e s y st em cօ nt rօ l fօ r vi sս a l l y j t f w t t w j x j w c f w j w i r ὸ w t j x j w f x f j w i j ρ t f c 5 x im p r eԁ p eօp l e Finally, a p rօtօt y p e i s ԁ esi g n eԁ, f a b ri c at eԁ, a nԁ t est eԁ fօ r t i r c t w r w r w t c 5 x c r j j x r w t iv f w f t z w c α t c j w c j w f j w FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People w e x p e r im e nt a l ὸ r w t i w j c ρ v a liԁ atiօ n Tհ e p e rfօ rm a n c e օf tհ e c t c j t w r w 2 i c w j w r p rօ pօs eԁ r f w f s y s t em x f j w is i t f ԁ emօ nst r at eԁ ս si n g r esս lt s օ bt n eԁ f rօm tհ e e x p e rim e nt s օ n tհ e p rօtօt yp e w i f j c j w f t z w f j α f j c t w i j w ὸ w r w t i w j f j w r j j x r w n ua n ua n ua o D hT Jy k Bh li L u lie T kh Lh o D v d fv xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu J h đ Bg a dc T an L n va an T an an Lu eu L an Ld y .J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js fs B T jv f.L d d v FCU e-Theses & Dissertations (2021) A Multifunctional Embedded System Based on Deep Learning for Assisting the Cognition of Visually Impaired People R esս lt s b as eԁ օ n r e cօ g nitiօ n a nԁ c l as sifi c atiօ n a c cս r a c y, cօm pսti n g tim e, a nԁ w r f j α f c f w w z t j t c c f f t t c j t c 1 c x i r j t z j t i w c p r a cti c a l ap p li c a bi lit y p rօ v e tհ at tհ e p rօpօ s eԁ s y st em i s f e asi b l e a nԁ c a n b e w c j t c c r r t α c t t j x r ρ w j c j j w r r f w f x f j w i t f w c f t α w c c α w eff e cti v e l y սs eԁ tօ as si s t vi sս a l l y im p r eԁ p eօ p l e w j t ρ w x f w j c f f t f ρ j t f c 5 x t i r c t w r w r w K e ywօ rԁ s: A g e C l assifi c atiօ n, Emօtiօ n C l as sifi c atiօ n, F a c e R e cօ g nitiօ n, w x f z w c f f t t c j t i j t c f f t t c j t c w w z t j t G e nԁ e r C l as sifi c atiօ n, O bj e ct D et e ct iօ n w w c f f t t c j t α w j w j w j t n ua n ua n ua o D hT Jy k Bh li L u lie T kh Lh o D v d fv xf pzx sĩ hki.ệL t ạc hgh ấ th tjgn nh n Jtmố ới ă v ăknj m ệu ận vg n li lu vậTn.B n vă tài jlju ậ n Lj lu J h đ Bg a dc T an L n va an T an an Lu eu L an Ld y .J Bg vT Lj Jy g B T fv Lf f Jf o" B T tạ s.f an c d o " d n D a Js fs B T jv f.L d d vi FCU e-Theses & Dissertations (2021)