The Research Of Gesture Recognition System Based On Visual

To meet the demand of more natural and humanized human-computer interaction, inthis paper, we study the natural scene and the fixed camera gesture recognition problems.After researched and analysed domestic and foreign research status and commontechnology of gesture recognition, it is do concrete analysis and validation for thefollowing several aspects: gesture segmentation, feature extraction, gesture recognition andtracking.The noise produced by the acquisition device and the environmental impact, filter out bymedian filtering method. After filtering noise, the system changes image collected by camerafrom RGB space to YCbCr space, combine the adaptive threshold segmentation method basedon skin color with the background difference method segment the image.Detecting image edge and extracting gestures shape feature is made in Feature extractionpart on the segmentation of image. After analysis found that static gestures feature extractionmethod based on image contours, has small amount of calculation and good robustness tointerfere with existing in the environment. As a result, the Canny edge detection algorithm isadopted to extract the edge after image segmentation. Given the Hu invariant moment ofrotation, translation and scaling invariance, this article use Hu moment to extracting gesturesshape feature.Template library is created in the context of single background and stable light. Thegesture images collected under ideal conditions, after artificial segmentation and denoising, useCanny edge detection algorithm to extract the edge character, use Hu invariant moments forstorage, as the matching template. Gestures extracted from the natural environment, afterextracting feature using the above methods, calculate the Euclidean distance between templatefeatures and test image features, and use the neighbor principle as recognition criteria. In orderto better achieve human-computer interaction, this paper also studied the gesture tracking. Implements the gesture tracking method based on the center of gravity and gesture trackingmethod based on CamShift algorithm.Finally, the median filtering, space transformation, Canny edge detection, featureextraction based on Hu moment, shape matching, focus tracking method and CamShift trackingmethod is verified. The results of gesture recognition and tracking are analyzed. The followingconclusions: Collected in the natural scene gestures can complete the segmentation, and removemost of the image noise. About the ten different gesture of1to10, recognition rate is different.Gestures1, gesture7and10recognition rate is low. The rest of the gesture recognition rate isabout80%. Gesture tracking effect is stable, can realize tracking lost back. But there are still hassome problems. Standard gesture library is not very perfect. Not training samples to a lot ofgestures. The representation of the sample gestures is not strong. For more complex scenarios ofgesture recognition remains to be improved.