【作者】 李强；

【导师】 范立南； 何友国；

【作者基本信息】 沈阳大学 ， 控制科学与工程， 2016， 硕士

【摘要】 红外图像及视频中目标区域检测一直是目标检测的重点和难点。目前,市场上存在的车载红外夜视辅助系统仅仅是将红外摄像机拍摄到的图像呈现到显示器,图像质量的好坏完全取决于摄像的质量和性能,并未对获取到的红外图像进行处理得到进一步的信息。因此传统车载辅助驾驶系统渐渐地无法满足人们的需求。随着计算机视觉技术的发展,图像处理算法在目标检测及目标区域检测方面不断被人们重视,逐渐成为模式识别与智能信息处理领域的研究热点之一。本文采用自建的车载红外图像及视频数据库,首先对车载红外图像的特点进行分析,对图像中信息进行提取。然后,围绕光流场目标检测算法进行分析和比较,并对车载红外图像目标区域检测的方法进行分析,同时提出改进光流场算法。最后,对算法进行仿真验证并将其移植到DSP开发板进行物理验证,取得较好实验效果。图像预处理部分,在比较了传统去噪算法后,由于车载红外夜视图像所具有的特征,提出了基于模板统计去噪算法。图像增强部分为了突出红外图像中各部分组成尤其是道路及运动的物体,本文提出方向扩散改进算法对车载红外夜视图像进行增强处理,该方法在对图像目标增强的同时可以很好的突出边缘信息,有利于我们采用光流场方法对目标及目标区域进行检测。仿真部分,利用MATLAB工具对车载红外夜视图像进行算法验证,结果显示本文算法对红外图像识别效果较好。然后将算法移植DSP开发板进行物理仿真,实验结果表明,该算法在车载红外夜视图像中对目标区域及目标的检测具有较好效果,对车载夜视辅助驾驶系统具有较大理论和实践意义,且具有广阔应用前景。更多还原

【Abstract】 By area detection in infrared image and video has been the emphasis and difficulty in target detection. At present, the market of vehicular infrared night vision assistant system of using infrared camera to just image presented to the display, the image quality depends on the quality and performance by camera, did not get to the infrared image processing to get further information. So the traditional vehicle auxiliary driving system gradually cannot satisfy people’s needs. Along with the development of the computer vision technology, image processing algorithm in target detection and test has constantly been appreciated by area, gradually become a hot spot in the field of pattern recognition and intelligent information processing.In this paper, using self-built vehicular infrared image and video database, first on the characteristics of vehicular infrared image analysis, to extract the image information. Then, around the optical flow field of target detection algorithm is analyzed and compared, and the vehicular infrared image can be analyzed by the method of area detection, and puts forward a new optical flow field algorithm. Finally, the algorithm simulation and transplanting it to DSP development board for physical verification, obtain good experiment effect.Image preprocessing part, after comparing the traditional denoising algorithm, due to vehicular infrared night vision image features that are characteristic of denoising algorithm is proposed based on template statistics. Image enhancement part to highlight in the infrared image of each parts especially roads and movement of the object, this paper put forward the direction of diffusion algorithm to enhance the vehicular infrared night vision image, the method of target image enhancement can be very good at the same time highlight the edge information and help us adopt optical flow method to the target and can through the testing area.Simulation part, using the MATLAB tools for vehicular infrared night vision image algorithm validation, the result shows that this algorithm for infrared image recognition effect is better. Then transplanted algorithm DSP development board physical simulation, the experimental results show that the algorithm in the vehicular infrared night vision image through regional and target detection has a good effect, for automotive night vision auxiliary driving system has a larger theoretical and practical significance, and has a broad application prospect.更多还原