Font Size: a A A

Model Based Object Extraction And Its Applications To ITS

Posted on:2007-05-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z W HeFull Text:PDF
GTID:1118360182990561Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
With the development of science and technology, more and more information are obtained through the processing and analysis of images and videos. The vehicle Identity code information and the traffic information in the urban road, which are the two important kinds of information in intelligent transportation systems, can also be obtained through the analysis of image sequences. The main focus of this dissertation is on the key problems and their solutions for automatic extraction of static and motive objects, especially the vehicle Identity code in an image and motive vehicles in a traffic scene, in sophisticated backgrounds.With different kinds of noises on the image, the vehicle Identity code can hardly be extracted. Initially, the vehicle image is filtered with both adaptive linear and nonlinear filters in order to reduce noises so that the candidate text lines can be properly located. Then, a series of standard align templates has been brought forward according to the standard align modes of the vehicle identity (ID) codes. Finally, the align mode of each candidate text line is obtained and then matched with those standard templates, and the vehicle ID codes can be extracted automatically. Through this method, other characters or marks in the vehicle image can be automatically abandoned and have no influence on the extraction of the ID codes.Background subtraction method is the widely used method to detect motive objects. The key technique to background subtraction is background modeling. A good background model can reflect the true background and can change from time to time according to the real scene. Two kinds of effective background modeling methods are proposed in this dissertation. One is iterative decision-based background modeling and the other is approximated median filter based background modeling. In the first method, the decision of whether a current pixel is a background pixel or a foreground pixel is first made, then only those background pixels are used for the updating of the new background. By doing this, the pixels which belong to a motive object will not be blended to the new background, resulting in a more accurate background model. In the second method, median value of a pixel histogram is used for the estimation of the background. To obtain the median value of a pixel histogram, we need to save the last few frames of the image sequences and sort them in order, which consumes both time and memory spaces. So in the second method, we use some iterative techniques to obtain the median value approximately. By using this method, both the time complexity and space complexity are small, and the obtained background model is relatively good. In order to obtain a more accurate detection result, color information is also applied in the background modeling. A color space called "rgs" color space is used here for its simplicity of conversion with RGB color space. When color information is utilized in the two background modeling methods, they can detect the motive vehicles effectively and efficiently. Another advantage of using color information is that it can help to eliminate the influence of the vehicle shadows.At the end of the dissertation, the role of motive vehicle detection in the trafficinformation gathering system is stated. In order to build a real-time system with high efficiency, several optimization techniques are proposed for an embedded DSP system, which is used for gathering the traffic information.Experiment results show that the proposed methods have good performance.
Keywords/Search Tags:Intelligent Transportation Systems, Image Processing, Pattern Recognition, Identity Codes, Align Template, Background Subtraction, Background Modeling, Embedded System, Traffic Information Gathering
PDF Full Text Request
Related items