Consistent depth maps estimation from binocular stereo video sequence

In the paper, an approach is proposed for the problem of consistency in depth maps estimation from binocular stereo video sequence. The consistent method includes temporal consistency and spatial consistency to eliminate the flickering artifacts and smooth inaccuracy in depth recovery. So the improved global stereo matching based on graph cut and energy optimization is implemented. In temporal domain, the penalty function with coherence factor is introduced for temporal consistency, and the factor is determined by Lucas-Kanade optical flow weighted histogram similarity constraint (LKWHSC). In spatial domain, the joint bilateral truncated absolute difference (JBTAD) is proposed for segmentation smoothing. The method can smooth naturally and uniformly in low-gradient region and avoid over-smoothing as well as keep edge sharpness in high-gradient discontinuities to realize spatial consistency. The experimental results show that the algorithm can obtain better spatial and temporal consistent depth maps compared with the existing algorithms.     

混合交通流视频检测中区域选择更新混合高斯背景模型研究

混合交通是我国交通的主要特征，利用视频检测技术可以获取混合交通流参数，实现混合交通的有效管制，由于检测过程中天气、光线等环境变化，实时有效的自适应背景提取模型尤其重要。本文在混合高斯模型的基础上，根据运动分割与Kalman运动跟踪，结合象素的时间与空间特性，提出区域选择更新混合高斯模型来抽取背景，克服了交通控制信号或交通阻塞等造成的长时间停车，以及高峰期大量运动物体长期充满当前图像等情况对背景抽取造成的影响，该模型通过对交叉口和路段视频进行背景提取，实验效果良好，证明了本方法具有较强的鲁椿性和自适应性。     

一种基于MRF的多层遗传递减视频对象分割算法

提出一种视频对象分割方法,将时域信息、空域信息和图像颜色特征有效地结合起来,采用MRF遗传递减式分类来提取视频对象.利用视频帧的颜色特征,在初始分割中采用时空域分水岭方法,建立一个时空域毗连图(ST-RAG);基于时空域毗连图建立马尔可夫随机场(MRF)模型,在分割过程中采用遗传递减式方法对区域进行合理分类,利用形态学进行后处理,分割出感兴趣的视频对象.实验结果表明所获取的分割方法具有灵活性,且精度高.     

Study on Moving Pedestrian Tracking Based on Video Sequences

The study on tracking of moving pedestrians based on video sequence has been presented in the paper. The video sequences have been processed by the background subtraction and adaptation, detection of moving objects, shadow detection, and image eroding and dilating in order to create the moving pedestrian contour. The change of location of the moving pedestrian contour and area has been considered in the method presented here. First, the method computed the centroid coordinates and radius of moving objects through the pedestrian contours. Second, the matching matrix has been established by the change of locations and radius between different moving pedestrians at two successive frames and the moving objects that matched each other have been found. The method has been tested using the video sequences of actual measurement, and the experimental results have been analyzed. Under the experiment condition, the contrails of moving pedestrians have been clearly drawn with a little noise at different image sampling intervals. Finally, the application condition of the method has been described.     

������Ƶ���е����˶�Ŀ����ٷ����о�

本文研究了基于视频序列的行人多目标跟踪方法。首先利用视频图像处理技术实现背景提取、背景更新、移动目标检测。然后利用移动行人在相邻两帧图像上位置和面积的变化特点，建立匹配矩阵，完成移动目标的匹配，实现多目标跟踪；最后利用实测的视频数据检验算法，实验结果表明，该方法能清楚地绘出行人的移动轨迹，分析了跟踪轨迹中噪声产生的原因。     

Human and bicycle detection system using range sensor

This paper presents a full-scale solution to the detection of the traffic data using laser device. Range images, gathered by a particular laser camera, are used in the multi-threshold segmentation. The multi-threshold segmentation is based on the height of the moving objects. In order to get the precise height of the moving objects, mapping of the original terrain is performed on the first step. On each layer, the clustering algorithm called iteration-self organizing data analysis techniques algorithm (ISODATA) is conducted afterwards. Kalman filtering technique is applied to recognize and track the moving objects. Extensive experiments show that these algorithms are effective in object recognition and tracking, as well as robust in the applications.     

双注意力引导的跨层优化交通场景语义分割

针对交通场景目标分割边缘不平滑以及小目标难以准确分割等问题，本文提出一种双注意力引导的跨层优化交通场景语义分割算法。首先，构建多分支特征提取编码网络，并利用串行非比例式空洞卷积实现空间上下文信息提取，进而改善小目标信息的丢失；其次，构建基于空间对齐的跨层特征融合解码网络，实现语义信息和细节信息的融合，增强不同尺度目标的表达能力；最后，提出通道和空间注意力机制，建模全局通道相关性和长距离位置相关性，增强网络对关键特征的学习能力。交通场景数据集Cityscapes和CamVid上的实验结果表明，所提特征提取编码网络、跨层特征融合解码网络以及注意力机制模块是有效的；所提语义分割算法获得了77.79%和78.66%的平均交并比，能够平滑目标分割边缘，尤其对细长条形目标具有鲁棒性。     

The Implementation of Mirror-Image Effect in MPEG-2 Compressed Video

Straightforward techniques for spatial domain digital video editing (DVE) of compressed video via decompression and recompression are computationally expensive. In this paper, a novel algorithm was proposed for mirror-image special effect editing in compressed video without full frame decompression and motion estimation.The results show that with the reducing of computational complexity, the quality of edited video in compressed domain is still close to the quality of the edited video in uncompressed domain at the same bit rate.     

һ�����͵��˶�Ŀ��ʶ��������㷨�о�

提出了一种应用于智能交通监控系统的运动目标识别和跟踪方法。针对帧间差分提取运动目标的缺陷与不足,提出了一种基于冗余小波变换的运动目标识别算法,即直接在冗余小波变换域提取运动区域,从而检测出运动目标。对于检测出来的运动目标,本文对mean-shift算法进行了改进,采用自适应mean-shift算法,对目标进行跟踪。实验结果表明,本文提出的算法可以有效地提取运动目标,即使目标与背景具有较高的相似度,也可以较准确的提取出前景运动信息,效果要好于传统的帧差法;跟踪目标准确度高,不受目标大小变化的影响。本算法具较高的实用价值和应用前景。     

融合边缘检测与区域生长的交通图像分割方法

在交通监控中,如何从复杂的背景中分割运动物体是至关重要的一步,针对车辆的运动阴影对图像分割产生的不利影响,提出了一种新的融合边缘检测与区域生长的彩色图像分割算法,算法同时考虑了图像的彩色信息和空间信息.该算法首先对彩色图像边缘检测,并根据检测结果设置种子像素;再基于颜色相似性生长准则,结合边缘检测结果,对每个种子点进行区域生长;最后,利用区域合并算法对剩余的像素进行合并.实验结果表明该算法很大程度上克服了阴影给图像分割带来的不利影响.     

利用MAS实现车辆动态识别与智能跟踪

针对视频图像车辆智能跟踪问题,提出了利用帧间差异积累动态矩阵进行自适应背景建模算法,采用背景差提取运动目标区域,设计了一种基于知识的多Agent智能系统进行目标分割、轮廓提取和空域滤波,增强了抗背景干扰能力,使获得的目标区域具有更好的空域连通特性;通过自适应核窗宽改进了MeanShift算法的收敛速度,利用SSD算法实现了快速初始定位。实验结果表明,该方法自治能力强,跟踪目标快速准确,实时有效。     

A Bintree Energy Approach for Colour Image Segmentation Using Adaptive Channel Selection

A new hierarchical approach called bintree energy segmentation was presented for color image seg-mentation. The image features are extracted by adaptive clustering on multi-channel data at each level and used as the criteria to dynamically select the best chromatic channel, where the segmentation is carried out. In this approach, an extended direct energy computation method based on the Chan-Vese model was proposed to segment the selected channel, and the segmentation outputs are then fused with other channels into new images,from which a new channel with better features is selected for the second round segmentation. This procedure is repeated until the preset condition is met. Finally, a binary segmentation tree is formed, in which each leaf represents a class of objects with a distinctive color. To facilitate the data organization, image background is employed in segmentation and channels fusion. The bintree energy segmentation exploits color information involved in all channels data and tries to optimize the global segmentation result by choosing the "best" chan-nel for segmentation at each level. The experiments show that the method is effective in speed, accuracy and flexibility.     

基于异构图学习的交通场景运动目标感知

为了提高无人车在交通场景中的运行效率和运输安全,研究了基于异构图学习的交通场景运动目标感知; 考虑实际交通场景中运动目标之间的复杂交互关系对目标运动的影响,基于异构图学习提出了交通场景中多目标检测-跟踪-预测一体化感知框架; 结合YOLOv5和DeepSORT检测并跟踪运动目标,获得目标的运动轨迹; 使用长短期记忆(LSTM)网络从目标历史轨迹中学习目标的运动信息,使用异构图学习目标间的交互信息,以提高运动目标轨迹预测准确度; 使用LSTM网络对目标运动信息与交互信息解码得到目标未来轨迹; 为了验证方法的有效性,在公共交通数据集Argoverse、Apollo和NuScenes上进行了评估。分析结果表明:结合YOLOv5和DeepSORT可实现对运动目标的检测跟踪,对交通场景中的运动目标实现了75.4%的正确检测率和61.4%的连续跟踪率; 异构图能够有效捕捉运动目标之间复杂的交互关系,并且捕捉的交互关系能够提高轨迹预测精度,加入异构图捕捉交互关系后,运动目标的平均位移预测误差降低了63.0%。可见,考虑交通场景中运动目标之间的交互关系是有效的,引入异构图学习运动目标之间的交互关系可以感知运动目标的历史与未来运动信息,从而帮助无人车更好地理解复杂交通场景。      

Exploiting Local and Global Characteristics for Contrast Based Visual Saliency Detection

Visual saliency is an important cue in human visual system to identify salient region in the image;it can be useful in many applications including image retrieval,object recognition,image segmentation,etc.Image contrast has been used as an effective feature to detect visual salient region.However,the conventional contrast measures either in spectral domain or in spatial domain fail to give sufficient consideration towards the local and global characteristics of the image.This paper presents a visual saliency detection algorithm based on a novel contrast measurement.This measurement extracts the spectral information of image block using the 2D discrete Fourier transform(DFT),and combines with the total variation(TV)of image block in spatial domain.The proposed algorithm is used to perform salient region detection in the image,and compared with state-of-the-art algorithms.The experimental results from the MSRA dataset validate the effectiveness of the proposed algorithm.     

基于时序融合的自动驾驶多任务感知算法

采用连续图像帧作为输入,挖掘连续图像帧之间的时序关联信息,构建一种融合时序信息的多任务联合驾驶环境视觉感知算法,通过多任务监督联合优化,实现交通参与目标的快速检测,同时获取可通行区域信息;采用ResNet50作为骨干网络,在骨干网络中构建级联特征融合模块,捕捉不同图像帧之间的非局部远程依赖关系,将高分辨率图像通过卷积下采样处理,加速不同图像帧的特征提取过程,平衡算法的精度和速度;在不同的图像帧中,为了消除由于物体运动产生的空间位移对特征融合的影响,且考虑不同图像帧的非局部关联信息,构建时序特征融合模块分别对不同图像帧对应的特征图进行时序对齐与匹配,形成融合全局特征;基于共享参数的骨干网络,利用生成关键点热图的方法对道路中的行人、车辆和交通信号灯的位置进行检测,并利用语义分割子网络为自动驾驶汽车提供道路可行驶区域信息。研究结果表明:提出的感知算法以多帧图像代替单一帧图像作为输入,利用了多帧图像的序列特性,级联特征融合模块通过下采样使得计算复杂度降低为原来的1/16,与CornerNet、ICNet等其他主流模型相比,算法检测精确率平均提升了6%,分割性能平均提升了5%,并保持了每秒12帧图像的处理速度,在检测与分割速度和精度上具有明显优势。      

A bintree energy approach for colour image segmentation using adaptive channel selection

A new hierarchical approach called bintree energy segmentation was presented for color image segmentation. The image features are extracted by adaptive clustering on multi-channel data at each level and used as the criteria to dynamically select the best chromatic channel, where the segmentation is carried out. In this approach, an extended direct energy computation method based on the Chan-Vese model was proposed to segment the selected channel, and the segmentation outputs are then fused with other channels into new images, from which a new channel with better features is selected for the second round segmentation. This procedure is repeated until the preset condition is met. Finally, a binary segmentation tree is formed, in which each leaf represents a class of objects with a distinctive color. To facilitate the data organization, image background is employed in segmentation and channels fusion. The bintree energy segmentation exploits color information involved in all channels data and tries to optimize the global segmentation result by choosing the ＂best＂ channel for segmentation at each level. The experiments show that the method is effective in speed, accuracy and flexibility.     

传感器智能芯片与阵列光纤对接平台运动顺序优化

从运动平台空间运动可能存在的720种运动顺序配置入手, 针对智能芯片与阵列光纤对接过程各运动单元产生的几何误差进行敏感性分析, 通过区分和归类各运动单元的敏感误差和不敏感误差, 将运动平台运动顺序配置数减少到90;考虑到运动平台各运动单元具有均匀分散、齐整可比的特性, 运用正交试验设计方法将敏感误差和不敏感误差确定为3个水平, 将6个运动单元确定为6个影响因素, 建立了对应的正交试验表, 得出了5条运动顺序配置的试验路径; 借助MATLAB仿真平台对5条运动顺序配置的试验路径进行了仿真试验, 获得了运动平台运动顺序最优配置; 在封装系统多自由度精密运动平台上进行了实测试验, 检验了仿真试验结果。试验结果表明: 传感器智能芯片与阵列光纤对接的运动平台在空间直角坐标系中最优的运动顺序为先沿横轴平动, 再绕横轴转动, 再绕纵轴转动, 最后沿纵轴平动; 该方法可优化光纤扫描雷达传感器智能芯片与阵列光纤对接的运动平台的空间运动顺序, 还可预测和规划其他多自由度运动平台的配准路径。      

复杂背景下多目标图像分割方法研究

真实路况中的运动车辆图像进行图像分割时,图像中往往存在多个车辆车牌信息,且这些车牌信息具有尺度不一,位置随机等特点,加之光照及复杂背景的影响,如何兼顾多个车辆车牌的分割效果是车辆检测和跟踪领域亟待解决的问题.为了解决这类工程应用中的问题,需要在尺度空间下对多目标图像进行分析.因本文在前期多尺度分割模型的基础上引入视觉注意机制,利用不变性特征实现多目标的定位及最优分割尺度的选取.经大量实验测试结果表明,该算法较好地实现了图像中多个车牌图像的分割并且具有较好的分割效果.     

人物交互检测研究进展综述

作为目标检测、行为识别、视觉关系检测的交叉学科，人物交互（human-object?interaction，HOI）检测旨在识别特定应用场景下人与物体的相互关系. 本文对基于图像的人物交互检测研究成果进行了系统总结及论述. 首先，从交互关系建模的原理出发，把人物交互检测方法分为基于全局实例和基于局部实例两类，并对代表性方法进行了详细阐述和分析；进而，根据所采用视觉特征的差异将基于全局实例的方法进行细分，包括融合空间位置信息、融合外观信息与融合人体姿态信息；然后，讨论了零样本学习、弱监督学习以及Transformer模型在人物交互检测中的应用；最后，从交互类别、视觉干扰以及运动视角三方面出发，总结了人物交互检测面临的挑战，并指出领域泛化、实时检测和端到端网络是未来发展的趋势.      

上海市出租汽车出行时空分布规律研究

为测定大样本视角下出租汽车出行时空规律,采用探索性空间数据分析（ESDA）技术进行相关研究。从上海市出租汽车调度系统的GPS数据提取单个运营日9921辆出租汽车的OD样本,按街道和时段统计出租汽车OD的时空发生量,定义并计算各区域各时段的载客空间密度指标,运用ESDA技术分时段计算、分析全局和局部莫伦值（Moran＇s I）。全局分析结果表明,上海市出租汽车出行在全局分布上具有较强的空间正自相关性,这与城市中心商业区高度集中的活动分布有关。局部分析结果表明,上海市出租汽车出行分布呈＂句＂字形聚集结构。全局结构和局部结构在时间序列上变化较小,表明上海市出租汽车出行分布的时空结构明显且稳定。