Hand Shadow Gestures To Rotate And Scale Sample Image Tim Taylor, Colorado School Of Mines Introduction • Setup • Projector • Webcam • Application Projector Web Cam Previous Work • 3-D Contour Modeling • Position Features • Fourier Descriptors • Finger Segmentation • Rule Classifiers • Hu Invariant Moments • Finger-Earth Mover’s Distance • Time-Series Curve Methods Applied • Hu Invariant Moments π₯ πππ = π¦ π₯ π₯−π₯ π π¦ − π¦ π π(π₯, π¦) π¦ π(π₯, π¦) (1+(π+π)/2) πΌ1 = π20 + π02 πΌ2 = π20 + π02 2 + 4π11 2 πΌ3 = π30 − π12 2 + (3π21 − π03 )2 πΌ4 = π30 + π12 2 + (π21 + π03 )2 πΌ5 = π30 − 3π12 π30 + π12 [ π30 + π12 2 − 3 π21 + π03 2 ] + 3π21 − 3π30 π21 + π03 [ π30 + π12 2 − (π21 + π03 )2 ] πΌ6 = π20 − π02 [ π30 + π12 2 − π21 + π03 2 ] + 4π11 π30 + π12 π21 + π03 πΌ7 = 3π21 − 3π03 π30 + π12 [ π30 + π12 2 − 3 π21 + π03 2 ] − π30 − 3π12 π21 + π03 [3 π30 + π12 2 − (π21 + π03 )2 ] Methods Applied • Hu Invariant Moments Limitations • Experimental Usage Methods Applied • Earth Mover’s Distance π· = ππ , π€π , … ππ , π€π πΈ = { ππ , π€π , … (ππ , π€π )} πΈππ· π·, πΈ = π πππ πππ π π π πππ Methods Applied • Earth Mover’s Distance Limitations • Partial Matching Method • Gesture Recognition • Find Hand • Choose Image from ROI • Time Series Curve • Distance Radially from Center of Mass • Segmentations to Find Signature • Threshold at certain radius • Calculate Best Gesture Match • Gesture Usage • Nearest Neighbor Rule Method • Find Hand 5 4 3 2 1 Method • Time Series Curve • • • • Find Center of Mass Calculate increments Find Boundary of ROI Remove Wrist and Normalize 0 0 1 0 Method • Segmentations to Find Signature • Thresholding Method • Segmentations to Find Signature • Thresholding π» = { ππ , π€1 , … , ππ , π€π } ππ = {π‘ππ , π‘ππ } ππ ππ π 0.01 1.06 21.90 1.93 2.34 7.76 2.74 3.03 6.12 3.18 3.66 9.07 3.74 5.28 25.25 Method • Segmentations to Find Signature • Thresholding πππππππππ‘π¦ = πΈππ·(ππ , ππ ) π π Method • Video of Operation http://youtu.be/RkCfXE1OEOk Analysis • Gesture Recognition of Three Gestures Flat Flat Open Pincers Five 74 7 1 Open Pincers 0 26 8 Five 0 14 47 Analysis • Speed of Operation • Average 0.79s per frame • 38500 calls to Time Series Curve function • 4518 calls to EMD function Future Work • FEMD vs. EMD • Efficiency of Operation • Application References A. Licsar, and T. Sziranyi, “Hand Gesture Recognition in Camera-Projector System,” LNCS, vol. 3048, pp. 83-93, 2004 H. Ha, and K. Ko, “A Method for Image-Based Shadow Interaction with Virtual Objects”, Journal of Computational Design and Engineering, vol. 2, pp. 26-37, 2015 J. Martin, M. Santos, and J. de Lope, “Orthogonal Variant Moments Features in Image Analysis”, Information Sciences, vol. 180, pp. 846-860, 2010 Y. Rubner, C. Tomasi, and M. Lindenbaum, “The Earth Mover’s Distance as a Metric for Image Retreival,” International Journal of Computer VIsion, vol. 40, no. 2, pp. 99-121, 2000 Y. Yao, and Y. Fu, “Contour Model-Based Hand-Gesture Recognition Using the Kinect Sensor,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 24, no. 11, 2014 Z. Chen, J. Kim, J. Liang, J. Zhang, and Y. Yuan, “Real-Time Hand Gesture Recognition Using Finger Segmentation,” The Scientific World Journal, 2014 Z. Ren, J. Yuan, J. Meng, Z. Zhang, “Robust Part-Based Hand Gesture Recognition Using Kinect Sensor,” IEEE Transactions on Multimedia, vol. 15, no. 5, 2013 Questions