Embedded Gesture Recognition Algorithm Research Based On Skin Color Segmentation

With growing needs of the society, gestures are popular for communication now withconvenience, nature and intuition. Gestures have a broad prospect on robot control, medicaldiagnosis and virtual reality, which are the key research content in the field ofhuman-computer interaction. Embedded gesture recognition based on skin colorsegmentation is studied and transplanted to S3C6410processor of an ARM11architecture,which achieves accuracy and real-time for static gesture recognition.Through careful analysis and study of the technology during the gesture recognition,the main research work is divided into the following five parts: Firstly, color space isselected for image conversion in the YCbCr space, which separates luminance andchrominance. Secondly, image is preprocessed to reduce noises that image contains andhighlight the edges and minutiae. Thirdly, gestures are segmented and statistics of skincolors, which distribute in CbCr space, is done to calculate skin color similarity values ofeach pixel. In addition, traditional Otsu algorithm is improved. By dynamic threshold, thewhole image is segmented into hand target and background region. Then morphologicalprocessing is used to eliminate the remaining glitches and empty. Fourthly, gesture featuresare extracted. The eight neighborhood search method is adopted for edges of the closedcontour and its features are extracted through invariant moment, which include gesturecontour centroid, perimeter, area, convex envelope and so on. Fifthly, gesture feature isrecognized. Similarity comparison of the collected gesture image and the loaded templateimage is done to output corresponding gesture template numbers. Among them, adaptivethreshold algorithm for segmentation of skin color and Hu invariant moments algorithm forcontour feature matching are core research work of this article.Linux-specific interface V4L2is used to capture video images and embeddedcross-development environment is built. Then, open source computer vision libraryOpenCV and Qt GUI library is transferred to the development board. The cross-compiler ofrelated programs achieves gesture recognition system in the embedded platform, andenables to complete the accurate identification for gestures of the figures, whose correctrate is up to91.3%.