BACKGROUND RECONSTRUCTION AND OBJECT EXTRACTION BASED ON COLOR AND OBJECT TRACKING

In YCbCr colorspace, a method is proposed to reconstruct the background and extract moving objects based on the Gaussian model of chroma components. Background model is updated according to changes of chroma components. In order to eliminate the disturbance of shadow, a shadow detecting principle is proposed in YCbCr colorspace. A Kalman filter is introduced to estimate objects＇ positions in the image and then the pedestrian is tracked according to its information of shape. Experiments show that the background reconstruction and updating are successful, object extraction and shadow suppression are satisfactory, and real-time and reliable tracking is realized.

Real-time moving object detection for video monitoring systems

Moving object detection is one of the challenging problems in video monitoring systems, especially when the illumination changes and shadow exists. Amethod for real-time moving object detection is described. Anew background model is proposed to handle the illumination varition problem. With optical flow technology and background subtraction, a moving object is extracted quickly and accurately. An effective shadow elimination algorithm based on color features is used to refine the moving obj ects. Experimental results demonstrate that the proposed method can update the background exactly and quickly along with the varition of illumination, and the shadow can be eliminated effectively. The proposed algorithm is a real-time one which the foundation for further object recognition and understanding of video mum＇toting systems.

Video Frame’s Background Modeling: Reviewing the Techniques

Background modeling is a technique for extracting moving objects in video frames. This technique can be used in ma-chine vision applications, such as video frame compression and monitoring. To model the background in video frames, initially, a model of scene background is constructed, then the current frame is subtracted from the background. Even-tually, the difference determines the moving objects. This paper evaluates a number of existing background modeling techniques in term of accuracy, speed and memory requirement.

Adaptive learning rate GMM for moving object detection in outdoor surveillance for sudden illumination changes

A dynamic learning rate Gaussian mixture model（GMM）algorithm is proposed to deal with the problem of slow adaption of GMM in the case of moving object detection in the outdoor surveillance,especially in the presence of sudden illumination changes.The GMM is mostly used for detecting objects in complex scenes for intelligent monitoring systems.To solve this problem,a mixture Gaussian model has been built for each pixel in the video frame,and according to the scene change from the frame difference,the learning rate of GMM can be dynamically adjusted.The experiments show that the proposed method gives good results with an adaptive GMM learning rate when we compare it with GMM method with a fixed learning rate.The method was tested on a certain dataset,and tests in the case of sudden natural light changes show that our method has a better accuracy and lower false alarm rate.

基于Object Proposals并集的显著性检测模型

针对当前常见的显著性检测模型得到的结果会包含大量的背景区域的缺点,本文提出了基于Object Proposals并集的显著性检测模型。该模型首先对于输入图片生成一系列Object Proposals,并通过其并集计算得到背景图;然后结合纹理特征和全局对比度得到初始显著图;最后,用得到的背景图对初始显著图进行背景抑制得到最终显著图。实验结果表明,在通用MSRA1000数据集上,本文提出的显著性模型与其他5种方法相比取得了很好的效果。

Automatic Video Segmentation Algorithm by Background Model and Color Clustering

In order to detect the object in video efficiently, an automatic and real time video segmentation algorithm based on background model and color clustering is proposed. This algorithm consists of four phases: background restoration, moving objects extract, moving objects region clustering and post processing. The threshold of the background restoration is not given in advanced. It can be gotten automatically. And a new object region cluster algorithm based on background model and color clustering to remove significance noise is proposed. An efficient method of eliminating shadow is also used. This approach was compared with other methods on pixel error ratio. The experiment result indicates the algorithm is correct and efficient.

Algorithm Research on Moving Object Detection of Surveillance Video Sequence

In video surveillance, there are many interference factors such as target changes, complex scenes, and target deformation in the moving object tracking. In order to resolve this issue, based on the comparative analysis of several common moving object detection methods, a moving object detection and recognition algorithm combined frame difference with background subtraction is presented in this paper. In the algorithm, we first calculate the average of the values of the gray of the continuous multi-frame image in the dynamic image, and then get background image obtained by the statistical average of the continuous image sequence, that is, the continuous interception of the N-frame images are summed, and find the average. In this case, weight of object information has been increasing, and also restrains the static background. Eventually the motion detection image contains both the target contour and more target information of the target contour point from the background image, so as to achieve separating the moving target from the image. The simulation results show the effectiveness of the proposed algorithm.

High efficient moving object extraction and classification in traffic video surveillance

Moving object extraction and classification are important problems in automated video surveillance systems. A background model based on region segmentation is proposed. An adaptive single Gaussian background model is used in the stable region with gradual changes, and a nonparametric model is used in the variable region with jumping changes. A generalized agglomerative scheme is used to merge the pixels in the variable region and fill in the small interspaces. A two-threshold sequential algorithmic scheme is used to group the background samples of the variable region into distinct Gaussian distributions to accelerate the kernel density computation speed of the nonparametric model. In the feature-based object classification phase, the surveillance scene is first partitioned according to the road boundaries of different traffic directions and then re-segmented according to their scene localities. The method improves the discriminability of the features in each partition. AdaBoost method is applied to evaluate the relative importance of the features in each partition respectively and distinguish whether an object is a vehicle, a single human, a human group, or a bike. Experimental results show that the proposed method achieves higher performance in comparison with the existing method.

Hardware Design of Moving Object Detection on Reconfigurable System

Moving object detection including background subtraction and morphological processing is a critical research topic for video surveillance because of its high computational loading and power consumption. This paper proposes a hardware design to accelerate the computation of background subtraction with low power consumption. A real-time background subtraction method is designed with a frame-buffer scheme and function partition to improve throughput, and implemented using Verilog HDL on FPGA. The design parallelizes the computations of background update and subtraction with a seven-stage pipeline. A stripe-based morphological processing and accounting for the completion of detected objects is devised. Simulation results for videos of VGA resolutions on a low-end FPGA device show 368 fps throughput for only the real-time background subtraction module, and 51 fps for the whole system, including off-chip memory access. Real-time efficiency with low power consumption and low resource utilization is thus demonstrated.

基于YOLOv5的复杂背景下植物叶片检测研究

对植物叶片进行检测是研究植物表型性状的基础,但真实环境下叶片间相互遮挡、叶片边缘特征不明显、幼叶目标过小以及外部环境如光照条件等因素影响会对叶片检测效果造成很大的障碍。针对复杂背景下的叶片检测,该研究提出了一种基于改进YOLOv5模型植物叶片检测方法。通过在骨干网络中引入空洞卷积,使得网络可以捕获到更广阔范围的上下文信息;利用双向连接的加权特征金字塔网络,以增强目标叶片特征提取并更好地融合特征信息;利用注意力机制,通过动态地调整注意力分布,以提高边缘特征表达能力。测试结果表明,在Plant Village数据集筛选的葡萄叶片图像以及自拍摄葡萄生长叶片上测试改进算法的可行性,改进的YOLOv5模型其叶片检测mAP比原生模型提高了5.8%,遮挡叶片检测精度提高了7.09%。叶片检测效果有显著提升。该研究提出的方法可以有效解决复杂背景下植物叶片检测效果不佳的问题,为植物表型研究提供技术支撑。

Real-time detection of moving objects in video sequences

An approach to detection of moving objects in video sequences, with application to video surveillance is presented. The algorithm combines two kinds of change points, which are detected from the region-based frame difference and adjusted background subtraction. An adaptive threshold technique is employed to automatically choose the threshold value to segment the moving objects from the still background. And experiment results show that the algorithm is effective and efficient in practical situations. Furthermore, the algorithm is robust to the effects of the changing of lighting condition and can be applied for video surveillance system.

针对浒苔目标检测的全局背景强化的位置蒸馏方法

浒苔检测是目前海洋环境智能监测领域研究的重要课题之一。为了有效解决传统浒苔检测方法存在的训练样本需求大的问题,本文提出了一种全局背景强化的位置蒸馏模型(GBS-LD)。通过引入全局上下文模块和背景蒸馏损失分支,解决了原始位置蒸馏方法在建模背景特征上的不足,在复杂海洋环境下有效提高了浒苔检测系统的稳健性。在浒苔检测数据集中,本文模型具有较高的准确性和实时性,为海洋智能监测提供了重要参考。

基于区域映射深度优化的动态多特征RGBD-SLAM算法

针对RGBD-SLAM在特征提取环节,由深度相机得到的深度图易出现深度不连续以及突然出现动态物体造成轨迹误差增大的问题,提出一种基于区域映射深度优化的动态多特征RGBD-SLAM算法。利用轻量级的语义分割网络获取RGB帧中的语义信息,将得到的语义掩码映射到其对应的深度图像中,并在掩码映射后的区域内进行近邻修复以完成深度图的优化。为减少动态物体对SLAM系统轨迹精度的影响,通过迭代最近点求解相机位姿,结合多视图几何得到物体位姿、运动估计矩阵以及动态视觉误差,进而估计物体运动状态并剔除动态物体。根据双向映射背景修复模型补全剔除区域的静态信息,并进行点线面特征的提取以完成定位与建图的任务。在公开数据集TUM中进行验证,实验结果表明所提算法相较于ORB-SLAM3、RGBD-SLAM、DS-SALM的平均绝对轨迹误差分别减少了78.2%、81.4%以及17.6%,表现了良好的轨迹精度与构图能力。

一种提升鬼影抑制性能的改进视觉背景提取算法

运动目标检测为视频帧生成指示运动目标像素的二值图,因此,正确判别待检测场景中的运动目标像素和背景像素是运动目标检测算法的主要任务。本文针对传统场景提取算法背景模型初始化导致的“鬼影”问题,提出了时域区间参考模块,该模块通过时域区间内像素的统计值生成背景参考值,在像素判别阶段利用候选前景像素与该参考值的差异进一步确定前景像素。实验结果表明,本文提出的方法对传统视觉背景提取算法性能进行了提升,对“鬼影”有较强的抑制作用,提高了检测准确度,具有较好的检测性能。

基于区域的弱纹理零件三维跟踪方法

为提升增强现实装配系统中对于弱纹理零件的跟踪效果,提出一种基于区域的三维跟踪方法。首先,采用全新的平滑阶跃函数优化基于水平集函数的图像分割方法,提高了轮廓边缘的分割效果;然后,设计了像素前背景颜色后验概率统计模型,增强了连续帧之间的时间一致性,提高了对运动模糊的鲁棒性和准确性;最后,采用高斯牛顿方法进行姿态优化,利用其快速收敛和数值稳定的性质,保证算法的实时性和稳定性。实验结果表明,所提出的三维跟踪方法能对弱纹理零件进行精确的跟踪。同时,在面对运动模糊或背景杂乱等干扰时,图像分割与位姿估计表现出更强的鲁棒性,满足了工业场景中对弱纹理零件跟踪的要求。

融合上下文感知和背景探索的伪装目标检测方法

伪装目标检测(camouflaged object detection, COD)旨在检测出与周围环境高度相似的伪装目标。针对目前COD方法中检测结果不完整、边缘细节模糊的问题,提出了一种融合上下文感知和背景探索(CABENet)的伪装目标检测模型。首先,该模型利用Swin-Transformer模型作为骨干网络,在多个尺度上提取全局上下文信息;其次,利用提出的注意力联级上下文感知模块扩大感受野,并从通道和空间两个维度增强网络的特征提取能力,再通过全连接解码器捕获隐藏对象的粗略位置图;最后,通过融合注意力机制的背景探索模块从背景信息中挖掘目标的边缘线索,加强伪装目标边缘特征的提取。在CHAMELEON、CAMO以及COD10K数据集上的实验结果表明,该方法在4个评估指标上的性能优于其他10个具有代表性的模型,在COD10K数据集上,平均绝对误差降至了0.026。

基于改进ViBe的自适应运动目标检测算法

针对ViBe算法无法去除动态背景,易出现鬼影及不能自适应光照变化的问题,提出一种复杂环境自适应的ViBe改进算法。通过计算区域的复杂度、闪烁波动度,对分类半径R和更新率T进行动态调整,对样本点进行有效性权重的计算,更高效地过滤背景噪声和适应光照渐变;在检测物体状态变化时,动态调整R和T,通过融合前景点计数和帧差法优化鬼影消除;通过识别最小外接矩阵区域差异加快去除鬼影;利用帧差法实时检测光照突变,及时进行重新初始化,避免大量误检。实验结果表明,改进ViBe算法在适应动态背景、光照变化及抑制鬼影等方面比原算法均有更好检测效果,检测精度平均提升了40.7%。

基于复杂背景的多尺度特征融合手-物交互检测方法

针对手与物体在交互过程中不同场景的背景噪声、光照变化等复杂背景,以及手-物相互遮挡、分辨率低等问题对手-物交互检测精度的影响,提出采用一种两阶段的多尺度特征融合手-物交互检测方法。首先,引入基于特征金字塔的残差网络Resnet50作为特征提取主干网络,实现深层语义信息和浅层细节特征的多尺度融合,提高小目标检测的精度;然后,利用检测到的手部区域与物体区域的几何信息来判断是否交互,过滤非交互的物体;最后,在大规模的室内外包括11个类别的手接触物体的人类互动视频帧数据集进行实验,提高网络的泛化性能。实验结果表明,本文所提方法和两阶检测方法相比,在提高检测精度的同时没有增加网络模型复杂度,同时在数据集不同类别的检测精度相对稳定,有效提升了网络的泛化性能。

搜索跟踪转台高分辨率图像的小目标检测技术研究

针对搜索跟踪转台高分辨率图像的要求,结合搜索跟踪转台的运动特点,提出了一种基于背景建模和YOLOv7的图像目标检测方法。首先,根据转台的运动参数,将提前采样的背景图像,由投影透视校正得到连续的背景图像,进而由背景差分法得到运动前景目标;其次,以YOLOv7作为该算法的目标检测器,结合转台背景信息半固定的特点,将背景图像作为训练数据进一步训练得到优化的网络参数;最后,搭建实验平台,建立了行人检测实验数据集。实验结果表明,本模型相比于原YOLOv7模型的召回率提升了6.56%,精确率提升了5.36%,可有效应用于高分辨率搜索跟踪转台的目标检测任务中。

基于背景抑制与改进多尺度LSD的声呐小目标检测算法

针对声呐小目标检测由于水下环境复杂、目标回波信号弱等因素造成虚警率和误检率较高的问题,文章提出基于背景抑制和改进直线分割检测(Line Segment Detection,LSD)的检测算法。首先对原始声呐数据截取序列片段,构建多周期累积历程图,凸显运动目标轨迹线特征;其次设计边缘滤波算子,有效滤除部分背景噪声,并结合投影变换进行线特征增强,不仅实现了断裂直线重连,还抑制了剩余噪声;然后基于图像金字塔改进了多尺度LSD直线分割检测算法,有效缓解了过检测问题,大幅增加了直线平均长度;最后为了合并冗余检测信息,利用运动轨迹时空一致性特征设计后处理模块,提高了检测定位精度。通过多组无人遥控潜水器(Remotely Operated Vehicle,ROV)、潜水员、空心球靶小目标序列的湖试、海试数据的定量与可视化结果定性分析,实验结果显示,文中算法与传统LSD相比,误检率和漏检率分别降低了11.2和3.9个百分点,定位误差下降了1.495个像素。结果表明,文中所提算法大幅提高了声呐小目标检测精度,为后续水下目标识别、跟踪等任务奠定重要基础。