Video based vehicle detecting system is a profitable way in an intelligent automobile/high-way controlling system and traffic monitoring system. It is possible to have a large detection area and a flexible installation way. To improve the usability of the video-based real-time intelligent traffic monitoring system, a new idea by using 2.5 dimensions time-space image has been introduced. First, color changing and intensity transforming should modify images. Second, more features of objects including length, width, and fuzzy height information can be extracted by object's shape analysis。Thirdly, a supervised learning method using a neuro-network realizes a self-adaptive recognition system. Then combined with the color information it can give a more accurate result. All the improvement endows the system with a desirable performance.