改进Camshift算法实时目标跟踪实现  

Improved Camshift algorithm for real-time object tracking

作　　者：严飞[1,2] 徐龙 陈佳宇 姜栋 刘佳[1,2] YAN Fei;XU Long;CHEN Jia-yu;JIANG Dong;LIU Jia(Automation College,Nanjing University of Information Science and Technology,Nanjing 210044,China;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing University of Information Science and Technology,Nanjing 210044,China)

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

出　　处：《计算机工程与设计》2025年第1期314-320,F0003,共8页Computer Engineering and Design

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

摘　　要：为解决Camshift目标跟踪算法在跟踪目标遮挡时陷入局部最大值、跟踪目标快速移动导致跟踪丢失以及光照变化影响跟踪精度一系列问题,提出一种改进Camshift目标跟踪算法。利用自适应权重与H通道特征提取模板,融合Kalman滤波算法并引入巴氏距离遮挡判别法。非遮挡时,使用Kalman预测调整跟踪搜索区域;遮挡时,使用Kalman预测跟踪。实验结果表明,将改进后算法部署于FPGA硬件平台能够准确地跟踪快速运动、遮挡干扰目标,在1920×1080分辨率下理论跟踪帧率为98.17帧/s,对1080p@60 Hz以及多种分辨率视频输入下平均跟踪重叠率达到84.68%。To deal with a series of problems such as falling into local maximum when tracking target occlusion,tracking loss caused by rapid movement of the tracking target,and the impact of lighting changes on tracking accuracy,an improved Camshift target tracking algorithm was proposed.Adaptive weights and H-channel feature extraction templates were utilized.The Kalman filtering algorithm was integrated and the Bhattacharyya distance occlusion discrimination method was introduced.When the target was not occluded,Kalman prediction was implemented to adjust the tracking search area.When the target was occluded,Kalman prediction tracking was implemented.Experimental results show that deploying the improved algorithm on the FPGA hardware platform can accurately track fast moving and occluding interference targets.In 1920×1080 resolution ratio,the theoretical tracking frame rate is 98.17 frames per second.Additionally,the average tracking overlap rate is found to be 84.68%for 1080p@60 Hz and multiple video input resolutions.

关 键 词：目标跟踪 实时 图像处理 硬件加速 卡尔曼滤波 直方图 现场可编程逻辑门阵列 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]