【作者】 刘影；

【导师】 杜军；

【作者基本信息】 哈尔滨师范大学 ， 计算机技术（专业学位）， 2024， 硕士

【摘要】 运动目标检测与跟踪技术广泛应用于无人驾驶汽车、视频监控、自动交通控制、生物医学图像分析和智能机器人等多个领域,为各种领域的应用提供了关键的信息和决策支持,设计一个实时且准确的运动目标检测与跟踪系统具有极其重要的意义。本文以DIGILENT公司的Nexys Video开发板为核心处理平台,利用FPGA的丰富逻辑资源和并行性优势,实现运动目标检测与跟踪系统,主要工作包括:(1)首先研究了硬件开发平台、摄像头以及图形显示器。其次使用处理器软核Micro Blaze进行整个系统的控制,该处理器具有高度可定制性,设计完成后可以进行流片,实现系统的物理实现,并且其资源占用、功耗、实时性能方面表现优越,有助于确保系统在满足设计需求的同时,在硬件资源和性能方面取得更为优越的平衡。最后说明了系统的开发环境。(2)提出了一种动态调整检测区域的方法,以实现对不同大小运动目标的灵活处理。针对小目标,采用区域裁剪以减少检测区域,降低计算量,专注于感兴趣的区域,提高算法效率;对于大目标,采用双线性插值进行图像缩放,降低总体计算量,有效减少处理的像素数量。结合区域裁剪和双线性插值的方法极大地提高了检测速度,系统更好地平衡了计算效率和准确性,为不同规模目标的检测提供了更灵活的解决方案。(3)对帧间差分和背景差分两种目标检测算法进行对比分析,实验结果证明了两种算法各自适合的应用场景。此外,还引入卡尔曼滤波跟踪算法对运动目标位置进行跟踪。传统的检测算法在面对一些外界因素变化时常常表现出较低的鲁棒性,通过引入卡尔曼滤波跟踪算法,能够更准确地跟踪目标,使得算法在面对目标动态变化时依然能够保持稳定的检测性能,增强系统的可靠性。(4)在可编程逻辑部分,采用IP核进行互连,其中包括一些自定义IP核,如OV5640摄像头采集、HDMI(High-Definition Multimedia Interface,高清晰度多媒体接口)接口、双边滤波算法、帧间差分算法和区域裁剪算法等。这些IP核是使用Verilog HDL(硬件描述语言)进行编写并封装。Micro Blaze软核处理器系统部分利用SDK软件完成了卡尔曼滤波跟踪算法、摄像头寄存器的配置以及AXI VDMA(Advanced e Xtensible Interface Video Direct Memory Access,高级可扩展接口视频直接内存访问)读写控制等任务。通过软硬件协同设计的方式,成功完成了整个系统的设计。更多还原

【Abstract】 The technology of motion target detection and tracking is widely used in multiple fields such as autonomous vehicles,video surveillance,automatic traffic control,biomedical image analysis,and intelligent robots,providing critical information and decision support for various applications.Therefore,designing a real-time and accurate motion target detection and tracking system is particularly important.This article uses DIGILENT’s Nexys Video development board as the core processing platform,and utilizes the rich logical resources and parallelism advantages of FPGA to implement a motion target detection and tracking system.The main work includes:(1)Firstly,the hardware development platform,camera,and graphics display were studied.Secondly,the Micro Blaze processor core is used to control the entire system.This processor has high customizability and can be taped after design to achieve physical implementation of the system.Its resource utilization,power consumption,and real-time performance are superior,which helps to ensure that the system achieves a better balance between hardware resources and performance while meeting design requirements.Finally,the development environment of the system was explained.(2)A method for dynamically adjusting the detection area has been proposed to achieve flexible processing of moving targets of different sizes.For small targets,region cropping is adopted to reduce the detection area,reduce computational complexity,focus on the area of interest,and improve algorithm efficiency;For large targets,bilinear interpolation is used for image scaling to reduce overall computational complexity and effectively reduce the number of processed pixels.The combination of region cropping and bilinear interpolation greatly improves detection speed,and the system better balances computational efficiency and accuracy,providing a more flexible solution for detecting targets of different scales.(3)A comparative analysis was conducted on two object detection algorithms,inter frame difference and background difference,and the experimental results proved that both algorithms are suitable for their respective application scenarios.In addition,the Kalman filter tracking algorithm is also introduced to track the position of moving targets.Traditional detection algorithms often exhibit low robustness when facing changes in external factors.By introducing the Kalman filter tracking algorithm,the target can be tracked more accurately,allowing the algorithm to maintain stable detection performance even in the face of dynamic changes in the target,enhancing the reliability of the system.(4)In the programmable logic section,IP cores are used for interconnection,including some custom IP cores such as OV5640 camera acquisition,HDMI(High Definition Multimedia Interface)interface,bilateral filtering algorithm,inter frame difference algorithm,and region cropping algorithm.These IP cores are written and encapsulated using Verilog HDL(Hardware Description Language).The Micro Blaze soft core processor system utilized SDK software to complete tasks such as Kalman filter tracking algorithm,camera register configuration,and AXI VDMA(Advanced e Xtensible Interface Video Direct Memory Access)read and write control.The design of the entire system was successfully completed through software and hardware collaborative design.更多还原