Optimized meanshift target reference model based on improved pixel weighting in visual tracking

The generic Meanshift is susceptible to interference of background pixels with the target pixels in the kernel of the reference model, which compromises the tracking performance. In this paper, we enhance the target color feature by attenuating the background color within the kernel through enlarging the pixel weightings which map to the pixels on the target. This way, the background pixel interference is largely suppressed in the color histogram in the course of constructing the target reference model. In addition, the proposed method also reduces the number of Meanshift iterations, which speeds up the algorithmic convergence. The two tests validate the proposed approach with improved tracking robustness on real-world video sequences.     

基于高度有效驱动注意力与多层级特征融合的城市街景语义分割北大核心CSCD

针对DeepLabv3+网络在进行城市街景图像分割任务时,没有充分利用到网络中多层级特征信息,导致分割结果存在大目标有孔洞、边缘目标分割不够精细等不足;并且考虑到城市街景数据具有天然的空间位置特殊性,本文提出在DeepLabv3+网络的基础上引入高度有效驱动注意力机制(height-driven efficient attention model,HEAM)与多层级特征融合模块(multi-stage feature fusion model,MFFM),将HEAM嵌入特征提取网络与空洞空间金字塔池化(atrous spatial pyramid pooling,ASPP)结构中,使其对目标关注更多垂直方向上的空间位置信息;MFFM通过融合多层特征图,在网络中形成多条融合支路依次连接到网络解码端,采用逐次上采样提高解码时像素上的连续性。将改进的网络通过CamVid城市街景数据集验证测试,实验结果表明,该网络能有效改善DeepLabv3+的不足,并且合理运用了数据集的位置先验性,增强了分割效果,在CamVid测试集上平均交并比(mean intersection over union,MIoU)达到了68.2%。     

A lateral position sensing system for automated vehicle following

This paper investigates the use of photoresistive light sensor arrays as a possible lateral position sensing system for a vehicle, or several vehicles, following a lead vehicle under separate control. A brief motivation for the use of such a sensing array is followed by a derivation of an appropriate sensor layout. It is shown that the sensing array has several benefits, including cost, simplicity, redundancy, and near linearity. Experimental results are given for a scaled vehicle following manoeuvre using a simple lateral control strategy. The results show that the sensor array may be attractive for lateral alignment in vehicle-following manoeuvres     

一种基于变化检测技术的SAR图像舰船目标鉴别方法

该文引入变化检测思想,利用SAR图像中海杂波和目标之间的灰度差异,通过对潜在舰船目标切片的目标像素和背景像素进行分离,计算目标像素聚集度(TPAM)特征,实现对高亮像素在图像切片中聚集程度的定量评估,从而鉴别目标切片中是否包含有舰船目标,有效去除杂波虚警。首先,基于感兴趣区域(ROI)切片中心为目标像素及四周为海杂波的合理假设,构建似然比变化检测量获取差异图像;然后,利用KSW熵阈值选择方法实现差异图像中目标像素和海杂波像素的自动分离,生成二值图像;最后,利用切片中心像素为种子点,对二值图像进行区域生长,计算目标像素聚集度特征,并判断目标切片是否包含舰船目标。基于RADARSAT-1 SAR实测数据的实验结果表明,该文方法得到的目标像素聚集度特征计算简单、稳健性好、可区分度高,具有良好的鉴别性能,能够去除大部分海杂波干扰产生的虚警,有效地降低目标检测虚警率。     

一种目标响应自适应的通道可靠性跟踪算法

为解决基于时空正则项的目标跟踪算法(STRCF)在目标短时遮挡时定位精度低和目标旋转时尺度估计不准确的问题，该文提出了一种目标响应自适应的通道可靠性跟踪算法。该算法在目标模型训练时加入了目标响应正则项，通过在求解过程中更新理想目标响应函数，使得目标被短时遮挡后可重新跟踪目标；加入通道可靠性评价各特征通道的可靠性，提高了模型对目标的表达能力；将目标图像转换至对数极坐标系下训练尺度滤波器，提高在目标旋转时的尺度估计精度。实验结果表明，该文所提算法较STRCF在平均中心位置误差中降低了28.54个像素，在平均重叠率中提高了22.8%，在OTB2015数据集下成功率曲线下面积较STRCF提高了1.5%。     

基于改进Camshift算法的NAO机器人目标跟踪

针对Camshift算法应用于NAO机器人目标跟踪过程中,当目标受到相似颜色背景干扰或被物体遮挡时跟踪失败的问题,提出一种基于ORB特征检测和Kalman滤波多算法结合的目标跟踪方法。首先检测目标ORB特征点初始化搜索窗口,然后利用Kalman滤波作为目标运动状态的预测机制,以预测的位置初始化Camshift算法。利用Bhattacharyya距离判断跟踪窗口的收敛性,若受到背景干扰,则利用ORB算法对当前帧中的Kalman预测区域和目标模型进行特征点匹配,重新检测目标在视频帧中的位置。根据Kalman滤波预测目标被物体遮挡后可能的位置来更新预测器参数。实验结果表明,改进的算法能够在相似颜色背景干扰和目标遮挡的复杂环境下,连续稳定地跟踪运动目标。     

基于场景感知的运动目标检测方法

背景减除法是一种主要的运动目标检测框架,但在复杂环境中构建一种初始模型建立周期短、可靠性高、鲁棒性好的模型仍是一大难题.本文从场景感知的角度出发,在背景减除框架的基础上提出一种目标检测方法.该方法根据前两帧中稳定的结构信息感知背景中潜在的前景区域,在第二帧建立初始模型时利用最近邻域背景像素点代替可能的前景像素点,提高了初始模型可靠性;结合颜色信息和二进制特征提出了像素点二级分类判决机制,并通过感知像素点邻域内的纹理复杂度自适应调整局部判决阈值和更新频率;在模型更新阶段提出处理误判的反馈机制.在公开视频序列上同几种流行检测算法的实验对比结果证明了本文算法的有效性和优越性.     

改进的核相关滤波跟踪算法

针对核相关滤波算法在目标跟踪过程中尺度特定和遮挡判断失败的问题,文中提出一种利用自适应特征融合的位置滤波器来判断目标是否被遮挡的方法。该方法检测到峰值旁瓣比异常时,停止模型自适应更新,启动在线重检测;并结合尺度金字塔中的尺度滤波器来确定目标尺寸,从而得出精准的目标位置。实验通过复杂背景下的10组运动视频来评估改进算法的性能。与基础核相关滤波算法相比,改进算法的平均中心位置误差降低了36.683 pixel;在像素阈值设为20 pixel时,平均距离精度提升了44.632%;在边界框重叠阈值设为0.5时,重叠精度提升了46.453%。     

Visual control of an autonomous vehicle (BART)-thevehicle-following problem

The authors consider the problem of vehicle-following including automatic steering and speed control of an autonomous vehicle following the motion of a lead vehicle. A visual control system for vehicle-following is presented. The system consists of the following modules: image processing, recursive filtering, and a driving command generator. First, the range and heading signal of the lead vehicle are obtained by visually identifying a unique tracking feature on the lead vehicle. Based upon this information, appropriate steering wheel and speed commands for driving are generated, which are then downloaded and executed on a microprocessor controller. The visual control system was tested on BART (Binocular Autonomous Research Team), a testbed vehicle developed at Texas A&M University for autonomous mobility. Successful full-scale test runs have been accomplished for speeds up to 20 mi/h     

一种改进的图像域SAR目标散射中心特征提取方法

散射中心是SAR图像目标识别的重要特征。本文基于属性散射中心模型，在文献[6]的基础上，提出了一种改进的图像域SAR目标散射中心特征提取方法。在该方法中，通过引入参数规则化处理步骤，解决了属性散射中心特征提取方法的收敛问题，提高了属性散射中心特征参数估计的精度和效率；提出了一种能同时实现散射中心数目确定和结构判别的方法，实现了散射中心类型的可靠判别。仿真数据和MSTAR实测SAR图像数据的实验结果，验证了本文改进的图像域SAR目标散射中心特征提取方法的有效性。     

基于红外特征的三维目标识别算法研究

基于三维特征的目标识别存在相似点云域容易误判、总数据运算量大等问题,而造成目标检出率低和误判率高。为了提高目标识别准确度与速度,提出了基于红外特征的三维目标识别算法。系统同时获取目标区域的二维红外图像与三维点云数据,利用目标红外特性的显著特征获得目标的投影范围,并计算系统与目标的位姿关系。根据红外特征映射关系计算点云数据中目标的限定范围,由此大幅缩减需要匹配计算的点云总量。在相同背景条件下对同一目标车辆进行测试,记录分析了3种不同测试角度条件下的识别数据。结果显示,传统点云识别算法的目标检出率均值为93.4%,误判率均值为19.5%,收敛耗时4.77 s。本算法的目标检出率均值为98.7%,误判率均值为1.5%,收敛耗时1.23 s。由此可见,基于红外特征的目标识别算法的检出率和误判率都更有优势,且处理速度更快。     

基于最简视觉显著性的红外目标快速提取方法

在红外深空目标跟踪系统中,为了能够从深空红外图像中快速提取微小目标,通过分析红外深空图像的特点,提出一种基于最简视觉显著性的红外目标快速提取方法。该方法在传统的视觉显著性的基础上,通过计算局部灰度最大值和目标像素的灰度平均值与邻域像素的灰度加权值的对比度组成特征向量,构造显著性模型,抑制背景并凸显目标,使之不但能够减少运算耗时,而且能够保证提取精度。通过对红外深空图像进行处理,实验结果表明该算法的运算时间仅为传统的视觉显著性算法的28%,且有较好的处理结果,证明了该算法的有效性。     

RSSI-based Localization without a Prior Knowledge of Channel Model Parameters

In target node localization problem, conventional methods based on received signal strength indicator (RSSI) assume a prior knowledge of a channel model and values of its parameters specific for an environment. This limits the conventional localization system to be set up quickly and effectively due to a necessary pre-measurement step to determine both the channel model and the values of its parameters. To address the limitation, a two-stage iterative algorithm which allows to localize a target node without any prior knowledge of the parameter values has been propose. Each stage of the algorithm can be implemented using different estimation methods, such as maximum likelihood (ML) and least square (LS) estimation which provides four different combinations. To determine the best combination, the location estimation performance for all four combinations is evaluated using experimental data collected in measurement campaigns on various indoor locations. The results reveal that the combination of ML estimation method implemented in both stages provides the best location estimation accuracy and the fastest convergence rate.     

目标机动对扩展弹道成型制导系统脱靶量影响分析

构造了基于剩余飞行时间指数函数的扩展目标罚函数,给出了对常值机动目标最优的扩展弹道成型制导律簇。利用无量纲化方法和伴随技术,研究了含有一阶动力学滞后的制导系统在目标常值机动、随机机动和终端落角约束作用下的无量纲位置和角度脱靶量随无量纲末导时间变化特性。采用直接法对伴随法仿真结果进行了验证。研究结果表明,当末导时间分别为制导系统动力学滞后时间常数的15倍、20倍左右时,落角约束和目标常值机动引起的位置和角度脱靶量均趋近于零;随着指数n的增大,导航系数也越大,无量纲位置和角度脱靶量振荡加剧,收敛时间也加长。目标随机机动引起的位置和角度脱靶量随着末导时间的增大维持在某个稳定的值,并不趋近于零,而且指数n越大,其脱靶量的稳定值也越大。     

基于免疫克隆高斯过程隐变量模型的SAR目标特征提取与识别

作为一种非线性维数约减算法,高斯过程隐变量模型(Gaussian process latent variable model,GPLVM)由于其适合处理小样本、高维数据,因而在模式识别、计算机视觉等领域得到了广泛应用.基于此,提出一种基于改进GPLVM的SAR图像目标特征提取及自动识别方法,其中利用改进的GPLVM进行特征提取,高斯过程分类进行目标识别.传统GPLVM使用共轭梯度法对似然函数进行优化,为避免梯度估值易受噪声干扰、步长对算法影响严重等缺点,提出基于免疫克隆选择算法的GPLVM,利用其具有快速收敛到全局最优的特性提高算法性能.实验结果表明,该算法不仅降低了特征维数,且提高了识别精度,从而验证了算法用于SAR图像目标识别的有效性.     

Pixel feature-weighted scale-adaptive object tracking algorithm

An effective object tracking method using weighted pixel features was proposed to deal with all kinds of complicated tracking situations,such as target movement,rotation,background interference and scaling and so on.First,the color feature and location information of the pixels in the target area were used to build the object model.Then the average weight image was used to estimate the scale variation coefficient.The aim was to adapt to the scale changes of the target.Finally,an update model was proposed,which was able to renew the object model and background model.The experimental results show that the proposed algorithm could make full use of the differences between pixels in the target area,so it can track more quickly and more effectively with strong robustness.     

Multi-feature Based Ensemble Classification and Regression Tree (ECART) for Target Tracking in Infrared Imagery

This paper considers target tracking as a binary classification problem to label pixels as either belonging to the target or the background. We present a novel robust algorithm, the multi-feature based ensemble classification and regression tree (ECART), for target tracking in infrared imagery (IR). In the first frame, a region of interest (ROI) containing target and background is initialized manually. Based on the multiple features of pixels, the ECART is trained online to distinguish between the target and the background. In the subsequent frames, the position and size of the ROI are predicted by the position and size of the target in the previous frame, respectively. Then, the ECART is used to label pixels within the predicted ROI, giving the label map. The new position and size of the target are finally found in the label map. Experimental results indicate that the proposed algorithm is effective and robust.     

基于Markov随机场K-Means图像分割算法

 传统的K-Means算法在图像分割中只与特征向量有关,从而忽略了像素间的空间位置关系,因而分割模型是不完整的.本文利用Markov随机场描述图像像素间的邻域关系,引入拒绝度的概念到聚类目标函数中的同时,提出了初始类别及初始中心点的确定方法,提出了较为完备的基于Markov随机场图像分割算法.并通过实验验证该分割方法在效果及效率上的有效性.     

基于角点增强改进的TLD目标跟踪算法

针对TLD算法的特征点无法有效表述目标问题,提出了一种基于角点增强改进的TLD目标跟踪算法。改进算法在跟踪模块加入了对目标表述能力更强,具有光照不敏感性和旋转不变性的Shi-Tomas角点作为跟踪特征点。跟踪器运行时,在角点经光流法跟踪和双向误差检测后,利用剩余的稳定角点定位目标窗口。对照结果表明,改进算法在面对目标抖动和形变时可以稳定跟踪;有效抑制因跟踪平滑点造成的漂移现象;提高了跟踪的稳定性。针对TLD算法跟踪过程中因在线模板积累造成的计算量持续增大、实时性持续降低的问题,提出了一种依据相似度中值的模板判断删除机制。该删除机制在模板积累到设定阈值时运行,根据模板与当前目标的相似度,删除不再具备代表性的模板;调整模板空间并更新模板数目。实验表明,该删除机制在应对模板更新快、持续时间长的跟踪情景时有效降低算法计算量,实时性可提高约20%。     

改进的SSD算法在智慧交通中的应用

针对智慧交通管理系统中交通车辆监控、车流量统计、违法车辆追踪等问题,为了提高目标车辆检测的准确率和效率,提出了一种改进的SSD( Single Shot MultiBox Detector)目标检测算法.该算法将相邻的卷积层进行特征信息融合,提高准确率;通过减少部分卷积层的深度,提高计算效率;为了提高泛化能力,在减少1...