Image Segment Algorithm Based On Scene Matching Aided Navigation System

During a car with a camcorder is drived from some place to the destination along afixed route, the scene matching aided navigation system can perform some function for real-time orientation. Firstly, after a series of scene images along some route are processed which are taken by the camcorder of the car, the characteristics of these images are extracted and saved into the computer in turn as a characteristic database with the input of the car's velocity and the start time. Then when the car is drived along the route again and arrives to the certain position, immediately the image is taken which is called the real-time image.At last, the characteristics of the real-time image are extracted and matched to some characteristic in the database in the car's proceeding turn. In such way, we can immediately get the current position of the car. Based on the real-time need of the system, the images in the memory are collected every several seconds. It is not nessary that each frame image is processed in order to reduce the processed data quantum and time. In fact, for each image, not each pixel is useful. On the aim of navigation, it generally draws some characteristics as the matching factors to realize the classification. The characteristics should be unique, independent, consistency and calculable etc. The purpose of the image processing is to extract some characteristics for subsequent matching end recognition. The mid-value filter algorithm is used to denoising which is generated during the process of image taken and transferred. The advantages of the mid-value filter are that it will cost less time and memory than the frequency filter. Although this method's weakness is the influence to the details, because this task's aim is to extract the fine edge of the big piece area, it can be neglected.The equilibrium algorithm is used to expand contrast. The algorithm which Delong Zhou put forwards based on the Pal and King's fast calculating magus method is used to outstand the edge and obtain the characteristic points. Image segment is a difficult point of the research of image processing algorithm. How to improve the speed of the segment is the emphasis of the research. In recent twenty years, GA developed rapidly and is widely used to all kinds of fields such as image segment,enhancement and pattern recognition. In the thesis, the image segment algorithm based on GA is improved to converge more rapidly and exactly and avoid running into the local optimization. The mathematical morphology is used to extract edges of the area which are cut as the target. The key problem of the system consists in the research of combining algorithm of scene-matching. It is influenced greatly by season, weather and light. Even in the same place, under the different conditions, the images take on different prospect characteristics. According to the analyses which are based on the characteristic of the scene and the error of the system, the characteristics of image edge, contour and texture etc. are used to the matching of the scene. The extract of the textures is based on the gray-deviation array.And the extract of the shape characteristic is based on seven moment invariants .It can prove the robustness of the matching algorithm to choose RHD as a kind of measurement function. The paper draws such conclusions: 1. The flow chart of software and hardware are drawn and the basic thought for the matching and recognition of the system is put forward. 2. The theories of image processing develop very fast which have already been more mature to come into theory system and applied very extensively to all kinds of fields. But nature prospects still have no perfectly processed methods due to their complexity. In the paper, I have a beneficial try about image processing and improve image segment of GA to adapt the real-time need of the system. 3. The matching algorithm of the paper on a certain extent avoids the influence of the bogus edge after image processing. And the source of the errors is analyzed in detail. All are beneficial to the combination of the image processing and matching algorithm to adapt real-time navigation. The image processing as next figure, Texture extraction from t gray-deviation array Features matching Shape extraction from the t edge output Improved Edge extraction gray-magus edge HD features points matching T image of HD which is the smallest 2n+1datum image Real-time image GA image segment Gray-deviation array Mid-value filter equilibrator...