融合边缘信息多运动目标检测识别系统设计  

Design of Multi Moving Target Detection and Recognition SystemBased on Edge Information Fusion

作　　者：严飞[1,2] 徐龙 孟川 李楚 刘佳 YAN Fei;XU Long;MENG Chuan;LI Chu;LIU Jia(Automation College,Nanjing University of Information Science&Technology,NanJing 210044,China;CICAEET,Nanjing University of Information Science&Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学自动化学院,南京210044 [2]江苏省大气环境与装备技术协同创新中心,南京210044

出　　处：《计算机测量与控制》2024年第5期72-79,共8页Computer Measurement &Control

基　　金：国家自然科学基金项目(61605083);江苏省产业前瞻与关键核心技术重点项目(BE2020006-2)。

摘　　要：运动目标检测算法易受到噪声和亮度变换的影响,从而出现虚假目标,很难满足在实际应用中准确性和实时性的需求;针对上述问题,提出了一种融合边缘信息的多运动目标实时检测算法,并在FPGA平台上构建了视频图像处理系统;该算法首先使用高斯滤波增强算法来提高抗噪性能;然后,采用三帧差分法减少虚假目标的出现概率;同时,对图像进行边缘检测,并将边缘信息与帧差法的结果相融合;最后,运用多目标识别算法完成对运动目标的识别和标记;该算法实现充分利用了FPGA的并行计算和流水线操作,经实验测试检测识别系统能够在1280×720@60 Hz视频场景下实现对最多16个运动目标的实时检测。Motion target detection algorithms are easily affected by noise and changes in brightness,leading to the presence of false targets,making it difficult to meet the requirements of accuracy and real-time performance in practical applications.To address these issues,this paper proposes a real-time multi-object motion detection algorithm that incorporates edge information and constructs a video image processing system on FPGA platform.Firstly,the algorithm utilizes a Gaussian filtering enhancement algorithm to improve noise resistance.Subsequently,the three-frame difference method is employed to reduce the probability of false targets.Simultaneously,the edge detection is applied to the images,and the edge information is fused with the results of the frame difference method.Finally,the multi-object recognition algorithm is used to identify and label the moving targets.The algorithm fully utilizes the parallel computation and pipeline operations of FPGA.Experimental tests demonstrate that the detection and recognition system can achieve the real-time detection of up to 16 moving targets in 1280×720@60 Hz video scenes.

关 键 词：FPGA 图像检测 多目标 边缘检测 实时性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]