基于矢量夹角的三维物体识别

提出了一种基于局部描述符的物体识别算法.算法根据点云位置信息得到其矢量和曲率信息,根据形状索引提取特征点,在每个特征点根据矢量夹角把点云物体分割成不同的曲面片,每个矢量夹角曲面片通过一个二维直方图描述.该图显示了特征点与邻域之间法线矢量夹角对特征点法线矢量与特征点到邻域矢量之间夹角的出现频率.对于给定的一个物体,通过比对预测物体和模型物体的曲面片集描述,可得到潜在的对应模型物体,再通过迭代最近点算法,得到最终的识别结果.     

利用异或运算和编码约束的降维LDP人脸识别方法

为了有效表示面部特征,在局部方向模式(LDP)的基础上,提出降维局部方向模式(RDLDP)。首先,修改LDP编码模式约束以完成模式的重构,通过对LDP码进行异或运算来计算每个块的单一码;然后,将所得编码图像划分为生成直方图,连接所有区域的直方图块以形成最终描述符。最后,计算特征向量间的卡方相异性度量值,并使用最近邻分类器完成最终的人脸识别。实验采用了三个公开的标准数据库FERET、扩展YALE-B和ORL。实验结果验证了提出算法的有效性。与其他基于局部描述符的先进方法相比,提出方法在准确度和错误识别率等方面更优。     

Pedestrian detection from thermal images: A sparse representation based approach

Pedestrian detection, a key technology in computer vision, plays a paramount role in the applications of advanced driver assistant systems (ADASs) and autonomous vehicles. The objective of pedestrian detection is to identify and locate people in a dynamic environment so that accidents can be avoided. With significant variations introduced by illumination, occlusion, articulated pose, and complex background, pedestrian detection is a challenging task for visual perception. Different from visible images, thermal images are captured and presented with intensity maps based objects’ emissivity, and thus have an enhanced spectral range to make human beings perceptible from the cool background. In this study, a sparse representation based approach is proposed for pedestrian detection from thermal images. We first adopted the histogram of sparse code to represent image features and then detect pedestrian with the extracted features in an unimodal and a multimodal framework respectively. In the unimodal framework, two types of dictionaries, i.e. joint dictionary and individual dictionary, are built by learning from prepared training samples. In the multimodal framework, a weighted fusion scheme is proposed to further highlight the contributions from features with higher separability. To validate the proposed approach, experiments were conducted to compare with three widely used features: Haar wavelets (HWs), histogram of oriented gradients (HOG), and histogram of phase congruency (HPC) as well as two classification methods, i.e. AdaBoost and support vector machine (SVM). Experimental results on a publicly available data set demonstrate the superiority of the proposed approach.     

基于OTSU分割和HOG特征的行人检测与跟踪方法

传统的HOG算法针对整幅图像进行行人特征提取,大量的非人窗口计算必然降低检测的准确率和效率。为此,提出一种基于OTSU分割和HOG特征的行人检测与跟踪方法。利用OTSU算法以最佳阈值分割图像,在分割区域的基础上进行Canny边缘检测,通过边缘的对称性计算确定行人候选区,继而采用经PCA方法降维后的HOG特征和隐马尔可夫模型对行人候选区进行检测验证。最后,以确定的行人区域为跟踪窗口,利用CamShift算法跟踪行人。多组实验结果证明,本文方法的行人检测效率和精度均有所提高,跟踪性能稳定、可靠。     

Classification of visible and infrared hyperspectral images based on image segmentation and edge-preserving filtering

The classification of hyperspectral images with a few labeled samples is a major challenge which is difficult to meet unless some spatial characteristics can be exploited. In this study, we proposed a novel spectral-spatial hyperspectral image classification method that exploited spatial autocorrelation of hyperspectral images. First, image segmentation is performed on the hyperspectral image to assign each pixel to a homogeneous region. Second, the visible and infrared bands of hyperspectral image are partitioned into multiple subsets of adjacent bands, and each subset is merged into one band. Recursive edge-preserving filtering is performed on each merged band which utilizes the spectral information of neighborhood pixels. Third, the resulting spectral and spatial feature band set is classified using the SVM classifier. Finally, bilateral filtering is performed to remove “salt-and-pepper” noise in the classification result. To preserve the spatial structure of hyperspectral image, edge-preserving filtering is applied independently before and after the classification process. Experimental results on different hyperspectral images prove that the proposed spectral-spatial classification approach is robust and offers more classification accuracy than state-of-the-art methods when the number of labeled samples is small.     

不均匀光照的路面裂缝检测和分类新方法

针对受光照不均影响的路面裂缝图像,提出一种基于Sobel算子和最大熵法的图像分割算法,并采用长线段与原图进行与操作和判断黑色像素所占比例的方法去除图像孤立噪声点.根据不同类型裂缝的几何形态,提取投影向量、分布密度和空洞数等特征值作为路面裂缝分类的依据,设计径向基函数神经网络的分类器实现对裂缝的准确分类.实验结果表明,较传统全局阈值算法,本文算法对光照不均图像的处理不仅能很好的提取裂缝边缘,且具有很强的抗噪能力,对路面裂缝的分类准确率高.     

Extraction and quantitative analysis of microscopic evacuation characteristics based on digital image processing

In this study, experiments of single-file pedestrian movement were conducted and the movement parameters of pedestrians were extracted with a digital image processing method based on a mean-shift algorithm. The microscopic characteristics of pedestrian dynamics, including velocity, density, and lateral oscillation, as well as their interrelations, were obtained and analyzed. Firstly, we studied the lateral oscillation phenomena of pedestrian movement. The result indicates that the trajectory of pedestrians presents a wavy form and the amplitude of the oscillation remains about 5.5 cm when the pedestrians move with free walking velocity, which is the velocity when there is no obvious interaction between sequential pedestrians; but when the movement velocity decreases to 0.27 m/s, the amplitude of oscillation increases to 13 cm. With increasing density, the velocity decreases and the amplitude of oscillation presents a linear increase trend. The increasing oscillation amplitude widens the occupation area of a pedestrian with low velocity, so as to make the moving efficiency even worse. Secondly, we studied the frequency of the oscillation; the result indicates that the frequency remains at 2 Hz when pedestrians move with a free walking velocity, but it presents a similar linear decrease trend when the velocity changes to a lower value. The decrease of oscillation frequency is also a negative feedback to the moving efficiency. Thirdly, it is found that with the increase of crowd density, the time interval between two sequential pedestrians increases, though the space gap between them decreases. The quantitative relation between time interval and crowd density is obtained. The study in this paper provides fundamental data and a basic method for understanding pedestrian dynamics, developing and validating evacuation models. The results are also expected to be useful for evacuation design.     

基于偏振测量的雾天图像场景分割

现有场景分割方法主要依赖于图像亮度、颜色和纹理等特征,然而在雾天图像中提取这些特征将变得困难且不稳定.基于此本文提出了适用于雾天图像场景分割的特征矢量,以及相应的特征提取算法.特征矢量由目标偏振度、深度和颜色三部分组成.特征提取算法分别为:用去相关的方法从图像偏振度分离出大气偏振度和目标偏振度;根据雾天退化模型和雾天图像偏振表示形式推导出场景深度信息;利用两幅偏振图像求出非偏振彩色图像,从而得到场景的颜色信息.将这些特征构成的特征矢量用于基于图的分割算法中,并从两个方面比较了仅使用颜色特征和使用本文特征矢量的分割结果.最后得出结论:对于雾天图像而言,这些特征比通常的颜色特征更加有效和鲁棒.     

基于机器视觉的圆锥滚子外观缺陷检测系统研究

回顾了近年来国内外工业生产中圆锥滚子外观缺陷分割领域的研究进展,为了实现更加简单高效的滚子外观缺陷检测,设计了一套合适的光学成像系统用于图像采集,提出了最大类间方差法和局部阈值算法的缺陷分割,对分割后图像进行特征提取并进行SVM分类。实验结果表明,该算法有较高的准确率和较好的适应性,优于传统的滚子外观缺陷分割算法,可以满足工业生产中的实际需求。     

Simulation of pedestrian counter flow through bottlenecks by using an agent-based model

Considerable research has been conducted on the topic of unidirectional evacuations from exits. However, few studies aim at simulating counter flow through a bottleneck with complex conflict. This paper proposes an agent-based model to investigate bidirectional flow evacuation. Pedestrian speed is determined by the speed of the leading agent and the surrounding agents. The moving direction of pedestrian originates from four forces, namely, gradient force, repulsive force, resistance force, and random force. These four forces dominate the main stream of the pedestrian moving trajectory, the interaction between pedestrians and their local environment, the resistance or disinclination to movement, and the random variations and chaotic nature of pedestrian dynamics. The novelty of this research is in the agent-based model that combines the agent and forces while providing insights for the simulation of the pedestrian dynamic on the cognitive level. The experiment results show that the behavior that arises from this model is consistent with the observations from Guangzhou Metro and that this model could help capture the essence of pedestrian behavior near egresses.     

一种基于特征不变描述的图像无缝拼接算法

针对尺度变化不明显场景的拼接,提出一种基于特征不变描述的图像无缝拼接算法.利用Harris特征检测算子进行特征点提取,在此基础上对提取出的特征点采用SIFT描述子进行处理,使其具备旋转不变性,然后利用k-d树算法进行遍历搜索定位,并用RANSAC算法进行图像间单应性变换矩阵计算,最后利用加权平均的融合方法进行图像无缝平滑,得到无缝拼接图像.实验结果表明:该算法在图像任意角度旋转的情况下,能有效地实现图像无缝拼接,并具有较强的鲁棒性.     

基于语义先验和几何约束的动态场景SLAM算法北大核心CSCD

实际场景中运动物体的特征点加入到相机位姿计算中,以及静态环境特征点过度稀疏都会导致移动机器人传统视觉同步定位与地图构建(simultaneous localization and mapping,SLAM)算法在位姿估计时精度低、鲁棒性差。设计了基于分支空洞卷积的双边语义分割算法,将环境区分为潜在运动区域和静态区域;结合几何约束进行静态特征点的二次判断及对没有先验动态标记而具有移动性的特征点的判断,并在事先均匀提取的全部特征点中进行移除,只应用静态特征点求解相机位姿和构建静态环境地图。在TUM公共数据集上进行实验,验证了提出算法在动态环境中SLAM的定位精度明显优于现有其他方法。在存在运动物体的真实环境下进行建图实验,与ORB-SLAM2算法进行对比,本文算法在动态场景中构建的地图更清晰。     

A method for moving objects segmentation based on human vision perception in infrared video

In this paper, a new region growing method to achieve the accurate and complete segmentation of the moving objects is introduced. Firstly, the ideal seeds of every moving object are extracted based on “hole” effect of temporal difference. Secondly, on the basis of the consideration that human vision system is most sensitive to the local contrast between targets and surrounding, we proposed a metric for “good” infrared target segmentation based on human vision perception. And according to this metric, a search method based on fine and rough adjustment is applied to determine the best growing threshold for moving objects. The segmented mask of every moving object is grown from the relevant seeds with the best growing threshold. At last, the segmented masks of all moving objects are merged into a complete segmented mask. Experimental results show that the proposed method is superior and effective on segmentation of infrared moving object.     

Cellular automaton model for evacuation process with obstacles

A bidimensional cellular automaton model is used to simulate the process of evacuation of pedestrians in a room with fixed obstacles. A floor field is defined so that moving to a cell with lower floor field means approaching an exit door. The model becomes non-deterministic by introducing a “panic” parameter, given by a probability of not moving, and by a random choice to resolve conflicts in the update of pedestrian positions. Two types of exit doors are considered: single (where only one person can pass) and double (two persons can pass simultaneously). For a double door, the longest evacuation time turns out to occur for a very traditional location of the door. The optimum door position is determined. Replacing the double door by two single doors does not improve evacuation times noticeably. On the other hand, for a room without obstacles, a simple scaling law is proposed to model the dependence of evacuation time with the number of persons and exit width. This model fails when obstacles are present, as their presence introduces local bottlenecks whose effect outweighs the benefits of increasing door width beyond a certain threshold.     

Segment-based region of interest generation for pedestrian detection in far-infrared images

We present a region of interest (ROI) generation method specialized for nighttime pedestrian detection using far-infrared (FIR) images. Because pedestrians typically appear brighter than background in FIR images, previous research efforts primarily attempted to extract ROIs based on the intensity threshold. However this approach has problems resulting from the intensity variances of pedestrians due to their clothing and, especially in urban scenarios, and other heat sources that emit more heat than the pedestrians. In this paper, we propose a novel ROI generation method that is based on combining image segments instead of using the intensity threshold. In order to minimize dependence on brightness, we utilize the low-frequency characteristics of FIR images. As a result, our proposed method generates a small number of ROIs at an acceptable miss rate and the generated ROIs provide advantages for classification because the pedestrians are satisfactorily arranged within a bounding box. Experiments conducted indicate that our proposed method performs reliably in urban scenarios.     

A semi-automated hybrid algorithm for accurate detection and counting of chemically etched tracks

In this paper, we present an accurate and comprehensive semi-automated algorithm for detection and counting the chemically etched tracks on polycarbonate nuclear Track detectors. The proposed algorithm consists of two main phases: a preprocessing phase for image preparation and quality improvement. It uses a format conversion into gray-scale image and contrast enhancement procedure. The contrast enhancement procedure employs the first-order fuzzy moment of the image and the Sugeno class of fuzzy complements to maximize the parametric indices of fuzziness of the image. In one feature extraction phase, two intrinsic characteristics of tracks are defined and formulated as topological feature (TF) and chromatic feature (CF). TF refers to tracks shape and geometry which are characterized and quantized by calculating the local Gray Level Spatial Correlation histogram of the tracks and CF refers to the color intensity difference of the tracks borders and tracks centric areas. The CF characterizes this difference specified to the tracks. After the feature extraction phase, the algorithm enters the detection and counting phase in which the fuzzy sets of background, noise (including small objects, notches, etc.) and track are constructed. Detected objects are then characterized according to their membership values to the fuzzy sets. The algorithm is designed in order to be applicable for different forms of tracks and track counting systems. Accuracy, linearity, sensitivity to the input parameters and comprehensibility of the method are validated in the simulation section.