電機工程學系 ; This paper presents a compressed-domain fall incident detection scheme for intelligent home surveillance applications. For object extraction, global motion parameters are estimated to distinguish local object motions and camera motions so as to obtain a rough object mask. Then, we perform change detection and/or background subtraction on the DC+2AC images extracted from the incoming coded bitstream to refine the object mask. Subsequently, an object clustering algorithm is used to automatically extract the individual video objects iteratively. After detecting the moving objects, compressed-domain features of each object are then extracted for identifying and locating fall incident. Our experiments show that the proposed method can correctly detect fall incidents in real time.