| Summary: | 碩士 === 國立海洋大學 === 資訊科學學系 === 90 === In this thesis, we design a novel integrated system for vehicle speed estimation, and suggest an efficient and effective searching algorithm called adaptive windowing prediction (AWP) to estimate vehicle speed automatically. The estimation algorithms in the proposed system employ a sequence of video images from existing CCD cameras on road and perform combinatorial image analysis techniques, block matching algorithms, and inverse perspective transformation as well. The methodologies of image analysis for object detection include differential motion analysis technique, mathematical morphology, region growing, and connected components technique, which enhance the results of target extraction in prior to block matching algorithm. In terms of matching theory, a new introduced adaptive windowing prediction algorithm, based on the results of previous frame calculation is proposed in this thesis. According to the physical constraints in real world, such as limited vehicle speed, driven in the permitted road direction, and un-allowed direction changing instaneously, the AWP algorithm is then able to predict the moving distance for the next video frame analysis. Finally, in order to satisfy the computational requirements of a real time system, we propose a quick look-up table in advance for mapping screen pixel distance to real world distance and physical vehicle speed. Experimental results show that the proposed system exhibits good performance in both computational reduction and improved accuracy.
|
|---|
