| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 95 === In this thesis, two kinds of human body posture analysis methods are proposed. One is continuous human body posture recognition by a recurrent fuzzy neural network, and the other is human posture estimation by silhouette and skin color information. Before posture analysis, it is necessary to segment the human body from background. A moving object segmentation algorithm is proposed to distinguish the human body from background from a sequence of images. This algorithm uses an automatic threshold determination method with Euler numbers for frame and background differences. After segmentation, a series of image processing is used obtain a complete silhouette of human body. The objective of posture recognition is to recognize four types of main body postures, including standing, bending, sitting, and lying. The significant Discrete Fourier Transform (DFT) coefficients of horizontal and vertical histograms together with length-width ratio of the silhouette are used as features. Recognizer is designed by a recurrent neural fuzzy network. In posture estimation, our objective is to locate significant body points. We combined skin color information with the convex points of contour of human body to locate head, hands, and feet. Experiment results show that the proposed approach can recognize the four types of postures and locate the significant points of human body with good performance.
|
|---|
