无线嵌入式平台运动目标检测的研究与实现

发布时间：2018-05-30 05:10

本文选题：运动目标检测 + 混合高斯模型　；参考：《河北师范大学》2013年硕士论文

【摘要】：运动目标检测是计算机视觉领域的一个研究热点运动目标检测是将视频画面中的前景目标提取出来，得到前景目标的相关信息，是目标跟踪和识别安全监控等视频处理的基础，检测结果直接影响着后续处理的效果和性能其广泛应用于海关工业检测图像处理安防等领域，尤其是无线嵌入式平台的运动目标检测可以应用到高危处网络条件和地理环境较差的应用场景，能够克服线缆和空间的限制，应用范围更加广泛本文针对特殊场景下的运动目标检测和视频传输中的丢包问题进行了研究，对现行的运动目标检测和错误隐藏算法进行了改进，并实现了基于TMS320DM6446嵌入式平台的运动目标检测系统本文针对因无线网络和特殊应用场景导致的网络丢包问题，对解码端的错误隐藏算法进行了分析和改进空域的错误隐藏采用基于方向插值的错误隐藏算法，提出了一种检测边缘方向的方法；时域的错误隐藏利用宏块的分割模式对各分割块分别进行错误隐藏，本文对运动矢量的预测方法进行了改进，有效地改善了隐藏效果针对运动目标检测算法的检测效率问题，在基于混合高斯模型的背景减除法和帧间差分法相结合的基础上，提出了基于匹配次数的运动目标检测快速算法首先利用帧间差分法得到背景区域，然后根据每个像素的观测值与背景模型的匹配次数将场景划分为静态区和动态区，，最后取静态区与帧间差分的背景区域的交集，对交集中的像素点进行隔帧检测，对其他像素点进行逐帧检测实验表明该算法在保证检测质量的前提下，明显提高了检测效率，更好的满足了实时性为了适用于特殊的应用场景，本文实现了基于无线嵌入式平台的运动目标检测通过对TI公司的TMS320DM6446双核开发板的硬件结构和软件架构的学习，将x264源码和运动目标检测源码移植到该平台，并对程序进行C语言级和线性汇编级的优化，提高了系统的性能
[Abstract]:Moving target detection is a hot topic in the field of computer vision. Moving target detection is to extract the foreground target from the video picture and obtain the relevant information of the foreground target. It is the basis of video processing such as target tracking and security monitoring. The result of detection directly affects the effect and performance of subsequent processing. It is widely used in the field of image processing and security in customs industry. Especially, the moving target detection of wireless embedded platform can be applied to the network condition of high risk area and the application scene of poor geographical environment, which can overcome the limitation of cable and space. In this paper, the problem of moving target detection and packet loss in video transmission is studied, and the existing algorithms of moving target detection and error concealment are improved. A moving target detection system based on TMS320DM6446 embedded platform is implemented. Aiming at the problem of packet loss in wireless networks and special application scenarios, this paper analyzes the error concealment algorithm at decoding end and improves the error concealment algorithm based on directional interpolation in spatial domain. This paper presents a method for detecting edge direction, error concealment in time domain uses the segmentation mode of macroblock to hide each partition block separately, and the prediction method of motion vector is improved in this paper, which improves the hiding effect effectively. Aiming at the detection efficiency of moving target detection algorithm, the background subtraction method based on hybrid Gao Si model and the inter-frame difference method are combined. A fast algorithm for moving target detection based on matching times is proposed. Firstly, the background region is obtained by using the difference method between frames, and then the scene is divided into static and dynamic regions according to the observed values of each pixel and the matching times of the background model. Finally, the intersection of the static region and the background region of the difference between frames is taken to detect the pixels in the intersection. The experiment of frame by frame detection of the other pixel points shows that the algorithm improves the detection efficiency obviously under the premise of guaranteeing the detection quality. Better meets the real time. In order to be suitable for special application scenarios, this paper realizes the detection of moving targets based on wireless embedded platform by learning the hardware structure and software architecture of TI's TMS320DM6446 dual-core development board. The X264 source code and the moving target detection source code are transplanted to the platform, and the program is optimized at the C language level and the linear assembly level, which improves the performance of the system.
【学位授予单位】：河北师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.1;TP391.41

【参考文献】