Research On Methods Of Real-Time Tracking And Recognition Of Hand Gesture With Complex Backgrounds

With the development of human-computer interaction techniques, the methods of hand tracking and gesture recognition is becoming one of the key technologies, which can be used for interactive virtual environment, multi-channel, multi-media user interface, as well as sign language recognition and snatch by robot arm. Therefore, the study to hand dynamic target tracking and recognition is not only of great theoretical significance, but also of broad and high practical value.Based on the basic theories and algorithms on the topic in the video capture, preprocessing to gesture image , gesture dynamic segmentation with complex background and the methods of feature extraction,the bottom software used as hand tracking and recognition based on VFW were designed and realized. It has the functions for controlling the attributes of video source and the format of image captured, setting the rates of frame, and other functions, gaining static hand images, dynamic hand posture sequence of images, video files, as well as real-time key frame captured for real-time dynamic analysis.On the method of hand detection,a set of rectangle features were used to describe the hand characters, and the fast calculation methods of their features and the evaluation ways of the separability of hand gesture class were given. Then the Adaboost algorithm was improved to deal with the excessive training.The tests and the corresponding conclusions for the study are as follow: Fistly, The classifier training experiment with same kind of gestures and different rotational angle. The experimental results showed that rectangle features could obtain the reliable detector and had a good real-time performance and a strong adaptive capacity in complex backgrounds, but it's more sensitive to the changes of hand gesture. When hand gesture rotated small angle, the detection rates were 95% above. Secondly, the detective experiment with Single hand gesture . Experiment show that: To detected gestures captured from the video camera in real-time, the methods has a well real-time ability and a stronger adaptability to complex background and noises. 3. the experiments for Automatic tracking and recognition under the conditions with interference and multi-targets. The results proved that: when the samples contained various brightness, complex background in a wide range of scenes, the classification trained could have a good real-time performance and a strong adaptability in complex backgrounds, the same as noises. Finally, hand cursor preliminary study. At first, the proposed a new name in the fields of human-computer interaction: hand cursor ( hand as a direct computer input devices). This paper presents alternatives to the mouse for operations in game, HCI experimental results show that: the samples formed after trained classifier, "Hand-Cursor" can replace the mouse to work, but there are still shortcoming of positioning accuracy and the problem of slow .