| Summary: | 碩士 === 國立交通大學 === 電機與控制工程系所 === 95 === Human activity recognition from video streams has a wide range of application such as human-machine interface, security surveillance, home care system, etc. The objective of this thesis is to provide a human-like system to auto-survey and then to track people and identify their activities. When the foreground color is different from the background color, the foreground subject can be extracted easily by the luminance component. When the foreground color is similar to the background color, we cannot extract the foreground image completely by the luminance component. To solve this, we utilize the HSV color space to build the background model, in line with similar spirit of W4 segmentation algorithm, which can not only extract foreground image but also be helpful to shadow removal. Since H and S component are not reliable in some conditions, we make use of three criteria to obtain reliable and static hue values.
A foreground subject is first converted to a binary image and transformed to a new space by eigenspace and canonical space transformations. Recognition is done in canonical space. A three image frame sequence, 5:1 down sampling from the video, is converted to a posture sequence by template matching. The posture sequence is classified to an action by fuzzy rules inference. In our experiment, extracting the foreground image in the HSV space improves not only the accuracy of foreground image but also human activity recognition accuracy.
|
|---|
