UDK 004.93'1 Vestnik SibGAU 2014, No. 3(55), P. 162–167
M. M. Chudnovsky
Siberian Federal University 79, Svobodny prosp., Krasnoyarsk, 660028, Russian Federation E-mail: Chudnovskymax@gmail.com
The author presents the new real-time algorithm for hand gestures recognizing on a video sequence, based on the color clustering principles, interframe differences for video sequences and contour analysis principal. The first chapter of this paper contains the research of related work, which is based on Viola-Jones algorithm, the search for the skin color, wavelet transformation, localization of the centroid of the image, etc. The second chapter describes the new gesture recognition approach, and third includes experiment results. A proposed gesture recognition algorithm resolves two tasks: gesture localization and gesture recognition. The gesture localization task is resolved by compilation results of the image segmentation by human skin color clustering principle and motion search on the video sequences. The feature extraction task is resolved by invariant gesture contour moments analysis. A presented approach does not use any tutorial images. Gesture domain is not close and can be set up during initialization process. Key performance indicators (KPI) of the proposed approach are number of processed video sequence items per time unit and gesture recognition stability mark. The analysis of the performance of the proposed algorithm shows that the performance is at a high level. An average frame rate, which can function designed system is about 50 fps, which even exceeds the ability of the human visual perception. Recognition stability depends on environment condition changes, but the number of recognition errors can be reduced to a minimum by video device calibrating. The quality of recognition is 96%, which is a good indicator for such purpose algorithms. The KPI analysis, based on experimental data, is shown the applicability of the proposed approach in real-time gesture recognition systems, based on different hardware platforms. This fact allows to use the algorithm primarily in the aerospace industry to build effective management systems based on gestural interfaces.
gestural interface, gesture recognition, objects color clustering, interframe difference for video sequences, contour analysis, invariant moments.

1. Fan N. Kh., Bui T. T., Spitsyn V. G. [Real-time Gesture recognition on video sequence based on Viola-Jones method, algorithm Camshift, the wavelet transform and principal component analysis]. Vestnik TGU. 2013, no. 2 (23), p. 102–111. (In Russ.)

2. Murlin A. G., Piotrovskiy D. L., Rudenko E. A., Yanaeva M. V. [Real-time algorithm and methods for hand gesture recognition on video]. Nauchnyy zhurnal KubGAU, 2014, no. 97 (03). (In Russ.) Аvailable at http://ej.kubagro.ru/2014/03/pdf/20.pdf (accessed 01.06.2014).

3. Malima, A., Ozgur, E., Cetin, M. A Fast Algorithm for Vision-Based Hand Gesture Recognition for Robot Control. Proc. of the IEEE 14th. Antalya, 2006, pp. 1–4.

4. Chudnovskiy M. M., Rusanova O. A. [The object’s color clustering principle research for human-computer interaction systems implementation] Proc. 5th International conference “System analysis and information technologies”. Krasnoyarsk, 2013, p. 431–435. (In Russ.)

5. Shi L. Skin Colour Imaging That Is Insensitive to Lighting. Proc. of AIC Conference on Colour Effects & Affects, June 15–18. Stockholm, 2008, p. 102.

6. Martinkauppi, B. Detection of Skin Color under Changing Illumination: A Comparative Study. Proc. of the 12th International Conference on Image Analysis and Processing, September 17-19. Mantova, 2003, p. 652–657.

7. Fleck M. Finding Naked People. Proc. of Fourth European Conference on Computer Vision (ECCV'96), Vol. II, April 14–18. Cambridge, 1996, p. 592–602. 

8. Hu M. K. Visual Pattern Recognition by Moment Invariants. IRE Trans. Info. Theory. 1962, vol. IT-8,
p. 179–187.

Chudnovskiy Maksim Mikhailovich – postgraduate, Siberian Federal University. E-mail: chudnovskymax@gmail.com