融合尺度空间的智能机器人自适应跟随仿真  

Adaptive Tracking Simulation of Intelligent Robot Based on Scale Space

作　　者：韩菊 曹峰[2] HAN Ju;CAO Feng(Department of Computer Science and Technology,Taiyuan University,Taiyuan Shanxi 030012,China;College of Computer and Information Technology,Shanxi university,Taiyuan Shanxi 030001,China)

机构地区：[1]太原学院计算机科学与技术系,山西太原030012 [2]山西大学计算机与信息技术学院,山西太原030001

出　　处：《计算机仿真》2021年第12期324-327,439,共5页Computer Simulation

摘　　要：当前智能机器人越来越多的使用视觉跟随目标,但是由于场景变换复杂,很容易因为视觉模糊或者出现遮挡而导致目标跟随失败。提出融合尺度空间自适应跟随方法。采取高斯核对空间特征进行映射,基于DFT变换设计位置滤波器;采用高斯函数与卷积计算对学习粒子进行特征尺度映射,基于DFT变换设计尺度滤波器。根据区域内的目标特征分类与子区域权重,细化非规则目标与边缘的处理,改善目标与背景识别的精度。针对目标存在遮挡的情形,综合响应峰值与响应波动进行衡量,并设计了基于似然估计的分类器,针对不同的遮挡等级采取相应的处理策略。采用包含模糊、遮挡,以及尺度等变换属性的OTB视频集来模拟机器人的目标跟随,通过实验结果,验证了所提方法在复杂环境中的自适应跟随性,有效提高了目标跟随精度、速度,以及鲁棒性。At present, more and more intelligent robots use vision to follow the target, but due to the complexscene transformation, it is easy to cause the target following failure due to visual blur or occlusion. Therefore, an a-daptive tracking method based on scale-space fusion is proposed. Firstly, the spatial features of the Gaussian checkwere mapped, and the position filter was designed based on the DFT transform. The feature scale mapping of learningparticles was carried out by Gauss function and convolution calculation, and the scale filter was designed based on theDFT transform. Then, according to the classification of target features and the weight of sub-regions, the processingof irregular targets and edges was refined to improve the accuracy of target and background recognition. Finally, inview of the occlusion of the target, the response peak and response fluctuation were measured. A classifier based onlikelihood estimation was designed, and corresponding processing strategies were adopted for different occlusionlevels. The OTB video set including blur, occlusion, scale, and other transformation attributes was used to simulatethe robot’s target following. The experimental results show that the proposed method can be used to adapt to the com-plex environment, and improve the tracking accuracy, speed and robustness of the target effectively.

关 键 词：智能机器人 目标跟随 位置滤波器 尺度滤波器 分类器 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]