The Researches Of Human Body Positioning And Hand Gesture Actions Recognition Based On Video Images

The widespread application of video made people’s life more colorful thus madepeople need higher technology request and more research interests on video imageprocessing. And as the primary organizer of social activities, human is regarded asan important concerned object of many video data naturally. So, using the computervision to simulate the human eyes to achieve the understanding of image is becominga hot research spot in recent years. The human motion analysis based on video is oneof the important topics in computer vision, and many researchers pay much attentionon this advanced research directions in the field of computer vision.This thesis focuses on human body positioning and hand action recognitionbased on computer vision, its mainly includes such sections: moving target detectionand segmentation, head positioning and detection, human limbs positioning anddetection, human motion tracking, human hand action training and recognition andso on.The main works of this thesis are as follows:(1) We use the Gaussian Mixture Models to implement the detection andsegmentation of the moving objects, and a series of image processing is implementedon the moving objects. Then, we adopt a new method based on Omega shape andHough Circle detection algorithm to position and detect the head. The thesis uses thefeatures that the head shape is similar to the Omega Shape, so we check the cornerpoints between the head and the shoulder to implement a rough head positioning, andthen we detect the head by using the Hough circle detection algorithm. This methodis faster than other methods in head positioning, and lower false detection rate.(2) Base on the head detection algorithm, we adopt a new method based on theU points to segment the limbs. By detecting the human body’s U points, we cansegment the human limbs successfully. Then, we adopt an easy and simple method todetect the human shoulder joints, the wrist joints. Then, we use the Kalman filteringto trace the human motion.(3) We use the human hands action model parameters getting from the previousstages, and we use the Support Vector Machine to train them. The experimentalresults show that our methods can recognize the human hand gestures accurately andbetter in robust and recognition rate.