Research On Video Vehicle Monitoring System Based On The Combination Of Color Spaces And Space-time Domain

The thesis presents an intelligent video vehicle monitoring system based on the combinationof color spaces and space-time domain information. This system is comprised of a monitor andfront detectors. The latter includes a Video detection card and a PC with two software,--DOWNITS running in WINDOWS and DSP vehicle detection program running in the videodetection card. The monitor is composed of the software of UPITS running in windows and a PC.The UPITS has a user-friendly man-machine interface, through which thedetection parameters can be modified remotely if needed.Through the WinSocketinterface,a monitor and up to 64 front detectorscan form a real time remote monitoring system,by which we can look over the video and detection results of any front detector, such as thevelocity and volume of vehicles. The monitor, if used together with picture capture card andlicense plate number identification software, can identify the license plate number. Each frontdetector can detect traffic stream in both directions.The major content of the thesis includes the design of UPITS, DOWNITS and DSPvehicle detection program. The camera calibration, vehicle parameters detection, traffic lightdetection and accident detection algorithms are researched for them, and innovationsinvolved in the algorithms are as follow:1. It proposes a vehicle detection algorithm based on a group of virtual detection lines,which optimized with information of space-time domain. It runs in real time, and itscomplexity is close to that of the algorithm using single virtual detection line, whileits robustness is better.2. In shadow-eliminating arithmetic, information of Y, Cr, Cb is efficiently used.3. It presents a algorithm of traffic light detection combining the information of bothYCrCb and HSV color spaces and space-time domain. Its accuracy is better thanthat of algorithms only making use of Y signal.4. It puts forword a background establish algorithm which based on the correlationof Y signal in time domain, and its threshold refresh algorithm is self-adaptive.The intelligent video vehicle monitoring system gets a good detection accuracy with thesealgorithms, which indicates a promising future of the algorithms.