Research On Gesture Feature Extraction Based On Kinect

Sign language recognition technology based on visual sense is one of the hotspots study at present.Traditional gesture recognition technology needs wearing data gloves,and has a high demand for data collecting background,thus could not get a wide application.The birth of Kinect effectively makes up the imperfection;it could effectively collect flat image and its depth data under complex background with the assistance of RGB camera and depth of focus camera.This dissertation takes advantage of this feature of Kinect,and makes a study on the feature extraction for static state and dynamic sign language gesture.For extracting the static state gesture features based on Kinect depth information,this dissertation makes a following study: at the very beginning,obtain depth image by Kinect,mapping the human skeleton into the depth image,locating the hand arthrosis points by using the skeleton tracking system,thereby achieve the real-time tracking;then,after getting the hand location,extract a maximum probability area in the hand as the region of interest(ROI);next,proceed pretreatment such as filtering and morphological operation in the region of interest;in the end,carry through a comparison among diverse feature extraction methods such as HU moment feature,FAST feature,SIFT feature,and SURF feature for the pretreated hand image.The experimental results indicate that the 7 features of HU moment have rotation scaling invariance;the FAST algorithm has a fast speed in feature detection,yet the detected feature points do not contain direction information;the SIFT algorithm could extract 128 dimension vector descriptors,the accuracy rate of feature extraction is quite high but with a long execution time;SURF algorithm is quite accurate and fast in calculating speed,hence,this study has selected SURF algorithm to conduct the gesture feature extraction under static state.For extracting the dynamic gesture features based on Kinect depth information,this dissertation makes a following study: first of all,the coordinate points of gesture movement track have been obtained by the Kinect camera coordinate system,and the start-stop position coordinates also get confirmed;secondly,conduct the track matching by using the track points coordinates obtained by cubic spline interpolation algorithm,a smooth curve of gesture movement track is gained;finally,take the location and angle of gesture movement track as the motion gesture features,carry out a unitary processing for the gesture movement track location of the sign language,ascertain the angle and location features of the sign language gesture movement track.