MPEG provides a standardized way to describe, compress and stream synthetic and natural audio and visual objects in a scene. The MPEG4 related projects and research topics running mainly aim to support Internet multimedia applications, such as remote video surveillance management, videoconference, video-on-demand, audio-visual email, live television streaming, and in the field of Wireless multimedia applications such as embedded solution for wireless handsets, PDAs, PCs and set-top boxes. Currently, the following MPEG4 related projects are running for implementations for industry application and academic research.

Software-based MPEG-4 Video CODEC and toolkit
The research in this project addresses the implementation of real-time coding and testing of the video compression algorithms toolkit, in order to provide a common platform/system core for relative application and projects.

MPEG-4 streaming over low bit rate network
On the basis of investigation and study of the multimedia streaming techniques, this project aims to implement a client/server software-based MPEG4 Streaming System for the delivery of multiple elementary stream presentations containing audio-visual objects for PC and handheld devices.

Automatic real-time video segmentation
Most of the existing algorithms for video segmentation are less used practically due to one or more of the following constraints: the background is variable due to camera motion; the light conditions slightly change; new objects can appear at any time; objects may remain for a long time in the scene; many objects in a scene and possible occlusion.

In order to serve real-time automatic applications and address the above problems, we present a novel approach, which uses sprite-object-based adaptive background technique and statistical feature analysis-based change detector. In this approach, a sprite generator is used to generate a sprite object, which presents the background without moving objects. The reference frames for change detector can be produced from sprite background object. A change detector detects the video shot and produces the difference-frame, which represents the change between a reference frame and the current frame. The differences are analyzed based on a probabilistic method and the moving objects appearing in a scene at a given time are extracted. A key frame identifier can speed up the segmentation procedure. The segmentation is initialized automatically by using reference frame, which is updated adaptively. Finally, a post-processor is used to obtain more accurate segmented shapes.