基于DSP的红外运动目标检测技术研究

发布时间：2018-01-31 19:21

本文关键词： 红外图像目标检测三帧差 surendra 高斯建模 DSP　出处：《北京理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着红外技术的发展,红外探测器以其良好的抗干扰能力,可以全天候连续工作,出色的隐蔽性,适用范围广等优点,在军事和民事领域都得到了广泛的应用。而红外运动目标检测是红外技术的一个重要分支,为后续的目标识别、跟踪、分析等提供了必备的准备,是一项有着重大研究意义的项目。本文首先对红外图像的特点、组成模型和处理过程进行了介绍,并对运动目标常用的三种检测算法,光流法,帧差法,背景差法进行了详细的讨论和研究。通过分析我们了解到,帧差法动态适应性较强,算法简单实用,但是目标内部容易产生空洞;一般的背景差法,可以比较完整的提取出运动目标,但是抗干扰能力差,而经过合适的背景建模和更新则可以改善最终结果。本文主要从四部分,对目标检测算法进行了研究和论证,包括:(1)三帧差法,(2)surendra背景建模,(3)高斯背景建模,针对高斯背景建模在处理和背景接近物体时提取物体轮廓不清楚的特点,在处理可见光视频时将其有三帧差进行结合,在处理红外视频时先将图像进行直方图均衡化后再进行高斯建模处理。(4)提出了可以减少运算量的运动区域帧差或者运动区域背景差法的方法,并且对运动区域帧差法进行了验证。以上算法在处理本课题中的红外视频和可见光视频中,均表现了良好的效果。为了满足本课题对实用性的要求,我们需要将算法移植到嵌入式平台DSP中,以实现脱机处理。这里采用了两块开发板,ADI的Bf548以及TI的DM6437,主要分为视频采集,视频处理,结果显示模块。整个处理流程与软件平台的设计相一致,大体分为:灰度化,背景差法、阈值分割、跟踪加框。为了更加明显的将目标表明出来,特意增加了跟踪加框这一环节。最终成功实现了可见光和红外运动目标检测的准确、实时的检测与跟踪。
[Abstract]:With the development of infrared technology, infrared detector with its good anti-jamming ability, can work all day, excellent concealment, wide range of application and so on. Infrared moving target detection is an important branch of infrared technology, which provides necessary preparation for subsequent target recognition, tracking, analysis and so on. This paper first introduces the characteristics of infrared image, composition model and processing process, and three commonly used detection algorithms, optical flow method, frame difference method. The background difference method is discussed and studied in detail. Through analysis, we know that the frame difference method has strong dynamic adaptability and simple and practical algorithm, but it is easy to generate holes inside the target. The general background subtraction method can extract the moving target completely, but the anti-jamming ability is poor, and the final result can be improved by the appropriate background modeling and updating. This paper mainly from four parts. The algorithm of target detection is studied and proved, including: 1) 3 frame difference method and 2% surendra background modeling and 3) Gao Si background modeling. In view of Gao Si background modeling in the processing and the background close to the object the extract body contour is not clear, when processing the visible light video, it has three frames difference to carry on the union. In the processing of infrared video, the image is equalized by histogram and then modeled by Gao Si. (4) the method of frame difference in moving region or background difference in moving region which can reduce the computation is put forward. And the motion region frame difference method is verified. The above algorithm has shown good results in the processing of infrared video and visible light video. In order to meet the practical requirements of this subject. We need to transplant the algorithm to the embedded platform DSP in order to achieve offline processing. Here we adopt two development boards: Bf548 and TI DM6437 which are mainly divided into video capture. Video processing, result display module. The whole processing process is consistent with the software platform design, generally divided into: grayscale, background subtraction, threshold segmentation, tracking box. In order to show the target more clearly. Finally, the detection and tracking of visible and infrared moving targets are realized accurately and in real time.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】