Research On Gesture Recognition Of Dynamic Target

This thesis mainly studies the gesture recognition of movement target method based on image processing. Target recognition has broad application prospects in areas of military, automation and vivil use, for instance, warning defense and attack of air targets, air traffic control. While moving object detection utilizes data image processing to acquire accurate moving objects, it is a key and difficult point in moving image coding, computer vision, pattern recognition and security monitoring. Feature selection is a key point in target recognition, the effectiveness of feature selection can influnce the recognition accuracy of target. Moreover, the selection of classifiers and parameter optimization has a great influnce on recognition accuracy.In this thesis, we mainly investigated methods of plane target detection and recognition under the background of simple in the sky. Firstly, We investigated the principles of wavelet and wavelet packet. In view of the moving target easily influenced by outside factors such as noise interference, a specific solution is selected. Compared with the traditional Canny edge detection, we presented the edge detection method using the combination of wavelet packet and based on local statistical threshold. This method can effectively extract the wee edge of the image, and carried on binarization processing. Then, We used the improved optical flow method, so as to obtain the optical flow field by analyzing the successive two plane images, and then detected the plane moving orientation. Considering the different geometric features,such as center point, size, shape, the ratio of plane body to wing, the ratio of the same bottom triangle area, we extracted target corresponding geometric features to train the plane target, and obtain the stable model with the method of K-neighbor, Bayesian, linear discriminant analysis and Support vector machine. After classifying the aircraft target, we verified the effectiveness of extracted feature vectors based on the kernel function of support vector machine. In the end, we used the line detection based on hough transform to detect the location of the crankshaft, so as to identify the plane’s attitude change in two-dimensional space.Through the simulation, edge detection method based on wavelet packet and local threshold can effectively extract the wee edge image, and the improved optical flow method has better detection rate for moving object detection. Moreover, classification based on the kernel function of support vector machine also has high recognition rate.