Research And Implementation Of Video-oriented Finger Detection Algorithm

Traditional human-machine interaction is limited to the application of keyboard,mouse and joystick for a long time. Although the rage touch screen technology haschanged this tradition, the friendly and natural interface still can’t meet people’srequirement. As a natural and intuitive interaction, gesture has been a hot issue fordomestic and overseas scholars. However, due to gestures’ diversity, ambiguity anddifferences between time and space, combined with hand is a complex deformable body,there still hasn’t make a breakthrough progress. Currently, the main drawbacks ofVision-based Gesture Recognition Technology are:(1) low recognition rate;(2) sensitive toscale variance and plane rotation;(3) poor real-time performance. As one of keytechnologies of gesture recognition, fingertips detection is facing the same problems.Therefore, this article is mainly devoted to explore a fingertip detection algorithm whichmeets the requirement of detection rate, real-time and robustness on fingers undermonocular vision. Then a finger detection system is designed and implemented.On the basis of the prior technology, the works have done in this article are shown asfollows:(1) To overcome the deficiencies of the existing fingertip framework, a new fingertipframework is proposed. The framework proposed in this paper realized robustness of thesystem by introducing palm width information on the basis of the original framework,enhanced the positive detection rate of fingertips by introducing finger width, finger length,and finger orientation information to lower false rate and miss rate which made the systemnot has high positive detection rate of fingertips, but also has good robustness.(2) For the issues that existing fingertip detection algorithms are sensitive to scalevariance and plane rotation, an adaptive fingertip detection algorithm is proposed. Twoadaptive fingertip detection algorithms are proposed by applying fingertip detectionframework proposed in this paper and combining different characteristics of the gesturewith different post-dealing methods. Firstly, skin region with motion can be detected byutilizing motion and skin color feature. Then gesture can be segmented out by connecteddomain segmentation algorithm. Thirdly, boundary tracking method is applied to storeedge points of target by sequence. Finally, different adaptive fingertip detection algorithmsare used to detect fingertips, and then the result is filtered by using different characteristicsof fingers, such as finger width, finger length and finger orientation. Experiment results show that the algorithms lead to high detection rate, good robustness performance and canadapt to changes in scale variance and plane rotation.(3) For the real-time issues of the finger detection system, the algorithm of everymodule are optimized, a fast and efficient region-based segmentation algorithm and tworeal-time and efficient fingertips detection algorithms are proposed, which makes thesystem meet the requirement of real-time performance.