【作者】 李兵；

【导师】 李哲英；

【作者基本信息】 北京交通大学 ， 微电子学与固体电子学， 2013， 硕士

【摘要】 随着信息技术的发展,图像目标跟踪技术已成为计算机视觉领域的重要课题,目前被广泛应用于军事、安全、工业和生物医学等各种领域。但是随着需求的不断增加,图像目标跟踪技术也面临越来越多的难点,如实时处理、鲁棒性、精确度、旋转摄像头采集到的图像目标跟踪等等。基于以上问题,本论文提出了一种基于分割的图像目标跟踪算法的FPGA实现。对于图像处理算法实现的主流方式计算机和DSP来说,图像处理算法的FPGA实现具有以下几个优点,一是硬件算法实现较之软件算法实现可以发挥硬件所固有的快速特性；二是FPGA可以实现并行处理和流水线的方式；三是不同的操作过程是并行的等等。本论文的主要贡献：一是通过算法到硬件电路的映射,完成了基于分割的图像目标跟踪算法的FPGA实现,二是将该算法移植到DSP上,针对同一样本对算法的软件实现(DSP)和硬件实现(FPGA)在处理速度、寄存器数量等方面进行对比。本论文首先引入了局部兴奋全局抑制振荡网络的图像分割算法,LEGION,该算法基于视觉皮层的神经网络机制原理,即一个兴奋状态的神经元可以激励起与其相似的一片区域的神经元,而其他神经元处于静默状态。该算法具有鲁棒性好,精确度高的优点,最重要的是振荡网络中的所有网络单元均可并行处理,非常适合于使用FPGA实现该算法。论文对算法进行了分析和验证,提出了基于该算法的FPGA逻辑模块划分和设计,将数字电路分为连接权重计算模块、振荡发起者判断模块和图像分割网络模块三部分,并分别用FPGA实现了这三个模块,其仿真结果与预期完全一致。另外,为了更好地对比算法的硬件实现和软件实现的优劣,论文中还将该算法移植到DM6437上,通过同一个样本将算法的FPGA实现与DSP实现进行处理速度、寄存器数量等方面的对比,并给出了对比结果。更多还原

【Abstract】 With the development of information technology, Image object tracking technology has become an important topic in the field of computer vision, and it is used in many fields such as military, security, industry and biomedicine and so on。 However, image object tracking technology is facing more and more difficulty with the increasing demands, such as real-time processing, robustness, precision, object tracking with rotating camera, etc. Based on the above issues, this paper presents the FPGA-implementation of image object tracking based on segmentation algorithm.Compare with the mainstream way, computer and DSP, to realize image processing algorithm, use FPGA to realize algorithm has such advantage:rapid inherent characteristics of the hardware, parallel processing and pipeline, the procedure is parallel in FPGA.The main contribution of this paper:first, through the mapping of algorithm to hardware circuit, FPGA-implementation of object tracking based on image segmentation algorithm is realized; second, we transplant this algorithm on DM6437and contrast processing speed and the number of registers between FPGA-implementation and DSP-implementation of the algorithm.This paper introduces an image segmentation algorithm, locally excitatory globally inhibitory oscillator networks, LEGION, this algorithm is based on the neural network of visual cortex mechanism principle, that an active neuron can excite another neural if they are in a similar region, and the others are in silence. This algorithm has the advantages of robustness and precision, and the most important is all the oscillation in the network can be processed in parallel, thus, FPGA implementation of the algorithm is very suitable. In this paper, we analyze and verify the algorithm in the first place, and then propose the division and design of FPGA logic modules based on the algorithm. There are three modules:connection-weight calculation module, leader cell selection module and image segmentation cell network module, and successfully use FPGA to implement them respectively, the simulation result is exactly the same as expected. Furthermore, in order to compare the algorithm implemented in hardware and software implementations, we transplant this algorithm on DM6437, contrast processing speed and the number of registers between FPGA-implementation and DSP-implementation of the algorithm, and give the comparison results.更多还原