Target Recognition And Detection Technology Based On Image Processing

Target recognition and detection technology involves many subjects, such as computer, signal processing and physiology. As one of the core technologies of machine vision system, it has significant application value in intelligent transportation systems, intelligent positioning and other research areas. This thesis is entrusted by the enterprise. The target recognition technology based on static image and target detection technology based on dynamic video image has been researched though the following parts on the basis of the classical algorithm :(1) Based on the traditional algorithms, such as K-L transform method, frame difference, optical flow and background-difference method, features of classical algorithm and its application environment is analyzed by experimental comparison.(2) The blurred license plate characters caused by light changes, stain and other conditions, has increased the difficulty of license plate recognition and seriously impacting the illegal vehicle tracking and effective inspection, the target detection technology based on classification is researched, such as license plate recognition as the representative. Therefore, a license plate recognition algorithm is proposed which combines Gaussian pyramid with histograms of oriented gradients(HOG). Firstly, utilizing the multi-scale expression of Gaussian pyramid, two layers of Gaussian pyramid model is established for Chinese character fuzzy license plate. The main feature is highlighted on the basis of describing details about the fuzzy characters; Then extracting HOG from two layers of Gaussian pyramid, the characteristic dimension of image is expanded and the ability of recognizing fuzzy Chinese character is enhanced. The simulation result shows that the recognition rate in this method is higher than HOG feature method and K-L transform method.(3) Dynamic video image, as the research object, detection problem based on target feature segment of geometry and statistical characteristics is researched in complex circumstance, such as light changes, shadows and other changes in a wide range of disturbances. For the traditional codebook algorithm in complex background circumstance, the process of indiscriminate recording pixel feature lead the background model which deviates from the actual background. Moreover, it is difficult to detect the situation that background color similar to foreground and serious color distortion leaded by severe light change. According to the above disadvantages, an improved CB moving target detection algorithm in complex background circumstance is proposed. In order to restrain the CB model deviation from actual background, the Codebook model(CBM) updating area which controlled by Dual-Surendra detection model is proposed. It can effectively reduce misjudgments of foreground in moving target detection. Then, the selection principles and methods of the threshold are proposed; In order to reduce the computational complexity of traditional CB algorithm and better distinguish the similar color between background and foreground, brightness changes of pixel is calculated under YUV space, and the space coordinate axis and brightness change is made in the same direction. Simulation results show that this algorithm can ensure the higher integrity and authenticity of detection area under complex background circumstance, compared with traditional CB algorithm.(4) In order to verify the improved Codebook algorithm validity, bran-new virtual image of Fedora14.0 operating system is structured by combining the virtualization software(VMwar workstation) with Fedora14.0 system in the upper computer(PC), and the research of moving target detection is implemented under Visual Studio 2010 development environment. Then video images gotten by video capture device is utilized to detect the moving target using Open CV2.3.1 on e mbedded operating system, and the detection result is displayed on QT graphic interface. The moving target detection of this thesis is finally proved effectively.