邵 肖偉 特任講師 空間情報基盤研究部門 2007-
My research interests mainly focus on spatial information acquisition techniques and corresponding applications, especially for moving targets based on laser range scanners. Nowadays video camera based systems are popular in many applications, but their spatial resolution is usually limited. By introducing new kinds of sensing technology such as laser range scanner, the research is trying to achieve a better accuracy of spatial information, e.g. the exact position of each person in a crowd. Such kind of high-resolution spatial information is fundamental, and it can be considered as a kind of spatial data infrastructure and be applied for various applications.
Major research topics are summarized as follows:
1. Highly Accurate People Tracking
People tracking has been studied for decades and many video based surveillance systems are able to track a small number of people can be done by. However, the task of tracking several hundred of people simultaneously in crowded scene is beyond the ability of existing video camera based systems. By introducing laser range scanner, accurate distance information is obtained and occlusion is greatly reduced. A fully automatic tracking system was developed for locating variable number of people based on the distance information and its performance was verified inside a subway station at rush hour. Future research in this topic includes extension from 2D distance data to 3D distance data, and potential applications based on the spatial information, such as behaviour analysis of people.
2. Sensor Network, Data Fusion and Information Inference
To acquire a better understanding of environmental information, using multiple sensors of the same kind is a common selection. However, integration of different types of sensors could be complementary and bring more benefits to data analyzing. Therefore it is more likely to achieve a better performance than using one single sensing technique. Meanwhile, it also introduces some difficulties in this field. Except traditional synchronization and calibration problems, it has a high requirement on data fusion and information inference techniques. How to combine information of different kinds at various levels/resolutions efficiently and effectively is the main objective in this research.
3. Motion Capture
The research focuses on reconstructing the pose of people by using multiple laser scanners set at different heights. Unlike active sensing techniques which require attendees to wear equipments on the body, passive sensing has no such requirements for attendees and can measure at a distance. Therefore, the captured data is more natural and accurate. Currently, the research is focusing on obtaining detailed pose information of walkers/runners.