Computer Security Symposium 017 3-5 October 017 1,a) 1,b) Microsoft Kinect Kinect, Takafumi Mori 1,a) Hiroaki Kikuchi 1,b) 1. 017 5 [1] 1 Meiji University Graduate School of Advanced Mathematical Science a) cs1705@meiji.ac.jp b) kikn@meiji.ac.jp Microsoft Kinect v EER 0.5 0.50. - 7 -
1 m : n 1 : n n 1 : n m n m n m : n n 1 3. Han 006 (GEI: Gait Energy Image)[3] 1 1 Kinect v RGB 10 1080 RGB 30fps 51 44 30fps 6 6 5 0.5 4.5m Shiraga GEINet[1] (CNN: Convolutional Neural Network) GEI 015 Anderson Kinect [4] 013 Igual [5] 4. 4.1 4. 4..1 Kinect v Kinect v Microsoft NUI(Natural User Interface) [8] Kinect RGB 3 Kinect SDK Kinect v [7] Kinect v 1-73 -
.3m 1m Kinect.m 1 Kinect v [6] 3 017/08/05 017/08/17 10 6 D A B C 1 シークエンス D E 4 165cm 170cm 0 51 3 1 4.3 4.3.1 3 10 U 1 U 10 ID 4 3 6 4.3. 1 1 1 1 3 1 1 3 1 A 3 E 1 S S4 S1 S0 S1 S18 S3 S17 S11 S16 S15 S14 S13 S1 S10 S S6 S5 S8 S7 S S4 S3 S1 4 4.3.3 4 4.3.3.1 1 5 5 4-74 -
5 ID S 1 FootL-AnkleL 11.76. 11.58 17.71 S FootR-AnkleR 11.88.08 11.86 17.05 S 3 AnkleL-KneeL 37.60 3.57 38.00 45.6 S 4 AnkleR-KneeR 37.46 3.60 37.86 45.33 S 5 KneeL-HipL 35.77 3.76 35.63 4.77 S 6 KneeR-HipR 35.6 3.7 35.76 51.38 S 7 HipL-SpineBase 8.1 1.44 8.5 11.66 S 8 HipR-SpineBase 7.71 1.56 8.15 10.76 S SpineBase-SpineMid 31.10 1.01 31.05 3.50 S 10 SpineMid-SpineShoulder.80 0.68.76 8.10 S 11 SpineShoulder-ShoulderL 17.4 1.15 17. 1.61 S 1 SpineShoulder-ShoulderR 17.58 1.43 17.6.46 S 13 ShoulderL-ElbowL 5.65.0 5.65 31.1 S 14 ShoulderR-ElbowR 5.57.3 5.41 31.60 S 15 ElbowL-WristL 3..18 3.43 31.37 S 16 ElbowR-WristR 4.8.46 3.66 3.33 S 17 WristL-HandL 7.5 1.71 7.6 1.56 S 18 WristR-HandR 7.71 1.7 7.77 15.08 S1 HandL-HandTipL 7.01 1.3 7.37 11.50 S 0 HandR-HandTipR 7.08 1.88 7.40 15.67 S 1 WristL-ThumbL.18.54.01 1.8 S WristR-ThumbR.43.45.3 18.73 S 3 SpineShoulder-Neck 7.46 0.1 7.45.10 S 4 Neck-Head 14.68 1.0 14.70 18.5 D6 A A1 D5 D4 A6 A5 D3 D D1 A4 A3 5 6 4.3.3. 6 5 4.3.3.3 ( ) 7 6 4.3.4 4.3. 4.3.3 3 4.3.5 i k f {S 1,..., S 4, D 1,..., D 6, A 1,..., A 6 } µ(f i,k ), median(f i,k ), max(f i,k ) - 75 -
6 ID D 1 FootL-FootR 3.40 14.61 3.10 74.51 D AnkleL-AnkleRigh 3.37 1.75 33. 73.4 D 3 KneeL-KneeR 3.1 4.6 3.46 37.7 D 4 FootL-HandTipL 71.1.6 70.51 106.0 D 5 FootR-HandTipR 6.18 11.0 67.48 11.61 D 6 HandTipL-HandTipR 47.0.6 46.46 3.01 7 ID A 1 ShoulderL-HandTipL 13.41 6.73 13.00 40.16 A ShoulderR-HandTipR 1. 7. 1.33 64.0 A 3 HipL-AnkeL 18.5 10.46 16.87 45.4 A 4 HipR-AnkleR 18.6 10.6 16.6 43.44 A 5 HipL-KneeL 13.76 7.68 1.17 38.60 A 6 HipR-KneeR 13.83 7.1 1.03 41.30 θ µ(f i,k ) µ(f j,k ) T same(i, j) = F if µ(f i,k ) µ(f j,k ) θ otherwise f g T if (µ(f i,k ) µ(f j,k )) + (µ(g i,k ) µ(g j,k )) θ same(i, j) = F otherwise 3 0.5 0.0 0.5 Y Head HandTipLeft HandTipRight AnkleLeft AnkleRight.0 1.5 1.0 0.5 0.0 0.5 X 7 4.4 4.4.1 3 7 5 4.4. 1 8 4.4.3 D 5 8 U 6 D 5 Median µ(d 5 ) 10 4.4.4 µ(d 5 ) Distance 10 0 30 40 50 60 D D(MovingAverage) 0 5 10 15 0 5 30 35 Frame 8 1 D - 76 -
8 ID µ(d 5 ) median(d 5 ) max(d 5 ) U 1 1 65.4 63. 77.0 U 1 6.8 63.0 70. U 1 76.3 73.1 100.4 U 75.6 75.3 104.5 U 5 1 6.7 61.6 68.5 U 5 60.8 60.8 71. Value 50 60 70 80 0 Value 60 70 80 0 mean median max U 6 D 5 U7 U5 U1 U8 U U10 U U6 U3 U4 UserID 10 µ(d 5 ) 11 (FAR) (FRR) (ROC) 1 µ(s 6 ),median(d 1 ), max(a ) 3 3 median(s 6 ) Density 0.00 0.05 0.10 0.15 0.0 FAR 0.0 0. 0.4 0.6 0.8 1.0 1 Self Others 0 10 0 30 40 Distance 11 µ(d 5 ) ) 0.0 0. 0.4 0.6 0.8 1.0 FRR S6 mean D1 median A max µ(s 6 ),median(d 1 ), max(a ) ROC EER Top10 EER EER µ(d 5 ) 0. max(d 5 ) 0. µ(s 6 ) 0.30 median(d 5 ) 0.30 µ(s 5 ) 0.31 median(a 4 ) 0.31 median(s5) 0.31 median(s 6 ) 0.31 µ(d 4 ) 0.3 µ(a 4 ) 0.3 5. 5.1 EER 10-77 -
Value 0.30 0.35 0.40 0 10 0 10 7 7 7 7 7 7 5 5 8 5 8 1 10 8 1 1 1 5 8 10 5 6 10 10 36 6 1 1 8 3 8 6 10 3 3 4 4 10 6 4 33 4 6 4 4 1 3 4 5 6 Num of Feature 0 0 0 40 13 EER 15 Max MDS 14 Density 0.00 0.05 0.10 0.15 0.0 Self Others 0 5 10 15 0 5 Distance (Max) 6 5. EER 10 10 10 11 Max EER 13 EER 5 Max 6 14 5.3 A1 A6 (MDS: Multi-Dimensional Scaling) 15 6. Kinect Kinect v 10 1 10 11 A 5,A 6 EER 7. Kinect v EER 0.5 0.50-78 -
10 EER 10 EER µ(d 3 ), µ(d 5 ) 0.5 µ(d 3 ), µ(d 4 ), µ(d 5 ) 0.5 µ(d ), µ(d 3 ), µ(d 4 ), µ(d 5 ) 0.5 µ(d 3 ), µ(d 4 ), µ(d 5 ), µ(d 6 ) 0.5 µ(d 1 ), µ(d 3 ), µ(d 4 ), µ(d 5 ), µ(d 6 ) 0.5 µ(d ), µ(d 3 ), µ(d 4 ), µ(d 5 ), µ(d 6 ) 0.5 µ(d ), µ(d 3 ), µ(d 5 ) 0.6 µ(d 1 ), µ(d ), µ(d 3 ), µ(d 4 ), µ(d 5 ) 0.6 µ(d 4 ), µ(d 5 ), µ(d 6 ) 0.6 µ(d 1 ), µ(d ), µ(d 3 ), µ(d 4 ), µ(d 5 ), µ(d 6 ) 0.6 11 EER 10 EER µ(a 6 ) 0.50 median(a 6 ) 0.48 median(a 5 ), µ(a 6 ) 0.47 median(a ), 0.47 median(a ), median(a 5 ) 0.46 max(a 5 ) 0.45 max(a 5 ), 0.45 median(a ), median(a 5 ), median(a 6 ) 0.45 median(a ), median(a 6 ) 0.45 median(d 1 ) 0.45 (https://developer.microsoft.com/jajp/windows/kinect/hardware, 017 8 ) [8],, SCIS015, Jan 015. [1] K. Shiraga, Y. Makihara, D. Muramatsu, T. Echigo, Y. Yagi, GEINet: View-Invariant Gait Recognition Using a Convolutional Neural Network, In Proc. of the 8th IAPR Int. Conf. on Biometrics (ICB 016), pp.1-8, Halmstad, Sweden, Jun 016 [] S. D. Bakchy, M. R. Islam, A. Sayeed, Human identification on the basis of gait analysis using Kohonen selforganizing mapping technique nd International Conference on Electrical, Computer & Telecommunication Engineering (ICECTE), Dec 016 [3] J. Han and B. Bhanu, Individual recognition using gait energy image, IEEE Trans. Pattern Anal. Mach. Intell., vol.8, no., pp.3163, 006. [4] V. Andersson and R. Araujo, Person Identification Using Anthropometric and Gait Data from Kinect Sensor, In Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence. 015. [5] L. Igual, À. Lapedriza, R. Borràs, Robust gait-based gender classification using depth cameras, EURASIP Journal on Image and Video Processing 013, 013. [6] JointType Enumeration - MSDN - Microsoft, (https://msdn.microsoft.com/jajp/library/microsoft.kinect.jointtype.aspx,017 8 ) [7] Kinect - Microsoft Developer, - 7 -