Research On Key Techniques Of Information Extraction In Video-Based Traffic Scene

Computer vision, which is based on image processing, has been widely applied in industry, while it hasn't received enough concern in the area of transportation until recent years. Computer vision provides a direct and convenient way to understand traffic environment. As is well known, there has so much information about traffic such as traffic sign, traffic signal and the status of vehicles in traffic is derived from vision. Therefore introducing computer vision technique into transportation system is a nature choice and has a promising prospect.This research is aimed at to extract some important traffic parameters by using traffic video sequences, which are acquired by a static monocular camera in specific traffic scenes. These traffic parameters include traffic flow, vehicle velocity and the vehicle type identification, etc. In this thesis, some key techniques in moving object detection are focused on, such as background image update and moving cast shadow detection. Based on these techniques, a prototype of traffic information measurement system in YUV color space is finished. The main contents can be listed as follows:1. In order to extract moving vehicles from traffic image efficiently, two frame-difference techniques are analyzed (differenced by the last frame in the same sequence or by the estimated background reference). An experiment is given to prove that the latter is better for detecting moving object reliably in traffic video;2. Three kinds of background extraction and self-adapting algorithms (Time Average Method, Time Median Method and Non-parametric Model-based one) are discussed in detail. The experiment indicates that the Non-parametric Model is the best one among the algorithms mentioned above. Considering the Non-parametric Model-based method is costly in time for computation and memory for storage, an improved algorithm in YUV color space is proposed in this thesis. By the way, in order to save the time of background extraction algorithm by the use of histogram analysis, a block based histogram analysis background estimation algorithm is also...