Robust Vehicle Speed Measurement Based on Feature Information Fusion for Vehicle Multi-Characteristic Detection

A robust vehicle speed measurement system based on feature information fusion for vehicle multi-characteristic detection is proposed in this paper. A vehicle multi-characteristic dataset is constructed. With this dataset, seven CNN-based modern object detection algorithms are trained for vehicle multi-characteristic detection. The FPN-based YOLOv4 is selected as the best vehicle multi-characteristic detection algorithm, which applies feature information fusion of different scales with both rich high-level semantic information and detailed low-level location information. The YOLOv4 algorithm is improved by combing with the attention mechanism, in which the residual module in YOLOv4 is replaced by the ECA channel attention module with cross channel interaction. An improved ECA-YOLOv4 object detection algorithm based on both feature information fusion and cross channel interaction is proposed, which improves the performance of YOLOv4 for vehicle multi-characteristic detection and reduces the model parameter size and FLOPs as well. A multi-characteristic fused speed measurement system based on license plate, logo, and light is designed accordingly. The system performance is verified by experiments. The experimental results show that the speed measurement error rate of the proposed system meets the requirement of the China national standard GB/T 21555-2007 in which the speed measurement error rate should be less than 6%. The proposed system can efficiently enhance the vehicle speed measurement accuracy and effectively improve the vehicle speed measurement robustness.     

An infrared salient object stereo matching algorithm based on epipolar rectification

Due to the higher noise and less details in infrared images, general matching algorithms are prone to obtaining unsatisfying results. Combining the idea of salient object, we propose a novel infrared stereo matching algorithm which applies to unconstrained stereo rigs. Firstly, we present an epipolar rectification method introducing particle swarm optimization and K-nearest neighbor to deal with the problem of epipolar constraint. Then we make use of transition region to extract salient object in the rectified infrared image pairs. Finally, disparity map is generated by matching salient regions. Experiments show that our algorithm deals with the infrared stereo matching of unconstrained stereo rigs with better accuracy and higher speed.     

一种基于仿射不变区域约束的立体匹配算法

针对现有立体匹配算法匹配速度的不足,提出一种基于区域约束的快速立体匹配算法。选取4对稳定的特征点作为初始匹配点对,在左右图像中分别构建四边形,然后将四边形划分成4个三角形,以每个三角形对为待匹配区域,利用仿射不变性进行区域内特征点匹配。获得特征点匹配对后,再以每个特征点为中心进行开窗,利用传统的基于灰度信息的方法进行精度验证。实验表明,该算法在保证一定精度的情况下能获得较高的匹配速度,可满足实时性要求。     

改进Fast-RCNN的双目视觉车辆检测方法

针对不同空间尺度的车辆表现出显著不同的特征导致检测算法效率低、准确性差且单目难以准确获取车辆距离信息的问题,提出了一种改进Fast-RCNN的汽车目标检测法,利用双目视觉对车辆进行测距。首先利用双目立体相机采集前方图像并进行预处理,加载深度神经网络Fast-RCNN的训练数据,再针对汽车不同空间尺度引入多个内置的子网络,将来自所有子网络的输出自适应组合对车辆进行检测,然后利用SURF特征匹配算法进行左右图像的立体匹配,根据匹配数据进行三维重建并确定车辆质心坐标,从而测量出车辆与双目相机之间的距离。实验结果表明,所述算法可以实现对车辆的快速检测,检测时间比传统的Fast-RCNN缩短了42 ms,并且实现了对5 m范围车辆距离的准确测量,其误差仅为2.4%,精确度高,实时性好。     

基于灰度投影的数字近景摄影立体影像匹配

根据数字近景摄影测量的特点,针对建筑物三维建模的需要,提出了基于灰度投影的影像匹配算法。该算法将影像行、列两个方向上的一维灰度投影,拓展为行、列、主对角线和次对角线四个方向上的一维灰度投影,计算左右影像的四个一维投影向量间的相似性测度的加权平均值作为最后的匹配测度。利用相对定向线性变换（RLT）算法来确定同名核线,以改进的动态定界法来控制待匹配点序列中后续点沿核线匹配的搜索范围,以减少匹配的搜索空间并提高匹配的准确度。利用该算法对建筑物数字近景摄影影像进行了实验,结果表明该算法具有较高的匹配速度和匹配准确率。     

场景轮廓的动态规划立体匹配算法

立体匹配算法是三维重建的关键步骤。由于实际场景中经常存在大片灰度相近的区域,稠密三维重建存在计算时间长、实时性差的问题。采用了场景的轮廓来重建场景的方法。基于场景中相邻点之间的视差应当是连续的假设,解决了轮廓在匹配时存在的"噪点"的问题,利用动态规划法对轮廓上各个点的视差进行约束以及求解最优解。由于提取轮廓后需要匹配的点数大为减少,用时可减少为原来的10%,得到与场景一致的轮廓视差图。     

基于OpenCV的车牌图像定位

现有的车牌定位方法几乎都需要先对白天夜晚的场景进行分类,在图像中存在其他灰度剧烈变化区域时,这种场景分类容易影响到车牌定位的准确率;为了对获取到的车辆图像进行准确地定位,提出了一种改进的灰度跳变车牌定位算法,首先加入了光照补偿,使得不需要对白天和晚上的场景进行分类;然后进行了长竖线的噪声去除,可以在背景比较复杂的情况下准确定位车牌;并对精确定位出的车牌进行边框的去除,有利于下一步的字符分割;另外,对OpenCV计算机视觉库的库函数的利用,算法的复杂度得到了简化,从而更好地满足了车牌识别系统的实时性;通过对出入口处采集到的分辨率为704×576的300张图片进行测试,定位率高于95%,对白天和夜晚获取的图像有很好的适应性,满足设计要求。     

基于BP神经网络的车牌字符识别算法研究

车牌字符识别是车牌识别系统中的关键环节。采用图像处理和神经网络相结合的方法设计新的车牌字符识别算法,先对分割出的车牌字符进行归一化处理,然后进行SOBEI．边缘检测和角点特征提取,最后输入BP神经网络进行训练、识别,其中BP神经网络模型属于改进型神经网络。通过一系列神经网络训练和仿真实验,车牌识别速度和正确率得到了明显的提高。     

Calibration for stereo vision system based on phase matching and bundle adjustment algorithm

Calibration for stereo vision system plays an important role in the field of machine vision applications. The existing accurate calibration methods are usually carried out by capturing a high-accuracy calibration target with the same size as the measurement view. In in-situ 3D measurement and in large field of view measurement, the extrinsic parameters of the system usually need to be calibrated in real-time. Furthermore, the large high-accuracy calibration target in the field is a big challenge for manufacturing. Therefore, an accurate and rapid calibration method in the in-situ measurement is needed. In this paper, a novel calibration method for stereo vision system is proposed based on phase-based matching method and the bundle adjustment algorithm. As the camera is usually mechanically locked once adjusted appropriately after calibrated in lab, the intrinsic parameters are usually stable. We emphasize on the extrinsic parameters calibration in the measurement field. Firstly, the matching method based on heterodyne multi-frequency phase-shifting technique is applied to find thousands of pairs of corresponding points between images of two cameras. The large amount of pairs of corresponding points can help improve the accuracy of the calibration. Then the method of bundle adjustment in photogrammetry is used to optimize the extrinsic parameters and the 3D coordinates of the measured objects. Finally, the quantity traceability is carried out to transform the optimized extrinsic parameters from the 3D metric coordinate system into Euclid coordinate system to obtain the ultimate optimal extrinsic parameters. Experiment results show that the procedure of calibration takes less than 3 s. And, based on the stereo vision system calibrated by the proposed method, the measurement RMS (Root Mean Square) error can reach 0.025 mm when measuring the calibrated gauge with nominal length of 999.576 mm.     

双目视觉用于鱼苗尺寸测量

为了有效地提高鱼苗尺寸测量的效率及精度,搭建了平行式双目立体视觉系统。根据双目视觉原理,首先基于改进后的由多组同心圆构成的标定板对双目视觉测量系统进行定标;然后利用Harris角点提取算法获取鱼苗图像的关键特征点,并基于归一化互相关(Normalized Cross Correlation,NCC)立体匹配算法对关键特征点进行匹配,提取坐标数据;最后根据坐标数据进行计算,得到鱼苗的空间坐标,实现对鱼苗尺寸的精确测量。结果表明,由所搭建的平行式双目立体视觉系统测量的相对误差在8%以内,为双目视觉应用于鱼类养殖业的可行性提供了依据。     

基于Census变换和引导滤波的立体匹配算法

在双目立体视觉系统中,立体匹配是关键步骤之一,其精度对后续的研究有着重大影响。Census算法由于具有简单明晰、运行效果好、实时性强等优点,被广泛采用。但Census立体匹配算法存在变换窗口中心点易受外界条件干扰、深度不连续区域匹配精度低等缺点,由此提出了一种新型的基于Census变换及引导滤波器的立体匹配算法。在Census变换阶段通过计算变换窗口周围的像素的平均值,降低了外界干扰的影响,同时在代价聚合阶段引入具有包边特性且计算量不依赖于滤波核大小的引导滤波器作为自适应权重。实验结果表明:所提算法在Middlebury测试平台上平均误匹配误差为6.03%,相较于目前Census立体匹配算法16.2%的平均误匹配率,匹配效果明显提高,且算法效率较高,具有较好的辐射不变性。     

A stereo vision-based obstacle detection system in vehicles

Obstacle detection is a crucial issue for driver assistance systems as well as for autonomous vehicle guidance function and it has to be performed with high reliability to avoid any potential collision with the front vehicle. The vision-based obstacle detection systems are regarded promising for this purpose because they require little infrastructure on a highway. However, the feasibility of these systems in passenger car requires accurate and robust sensing performance. In this paper, an obstacle detection system using stereo vision sensors is developed. This system utilizes feature matching, epipoplar constraint and feature aggregation in order to robustly detect the initial corresponding pairs. After the initial detection, the system executes the tracking algorithm for the obstacles. The proposed system can detect a front obstacle, a leading vehicle and a vehicle cutting into the lane. Then, the position parameters of the obstacles and leading vehicles can be obtained. The proposed obstacle detection system is implemented on a passenger car and its performance is verified experimentally.     

地震振动台实验三维全场位移测量的研究

提出一种基于高速相机双目立体视觉技术的大视场全场三维位移测量方法,用来测量地震振动台实验过程中的位移变化。给出了一种鲁棒的标志点匹配算法,基于VS2010开发环境,研发了用于振动台实验三维全场位移测量系统,设计了精度评估实验方案,验证该方法在大幅面位移测量中的精度,并利用该实验系统对高边坡模型振动台实验进行测量。结果表明:在3 m1.5 m视场范围,静态位移测量误差优于0.4 mm,动态位移测量误差优于0.5 mm,可以满足振动台实验的要求;该方法可以方便、直观地测量地震振动台实验中高边坡模型的位移场,并且测量得到X、Y、Z 3个方向位移曲线以及总位移曲线过渡自然、数据合理,是测量振动台实验全场位移变化的一种有效方法。     

基于四目系统的真实场景三维建模

为实现敦煌莫高窟文化遗产数字化保护,设计了一套基于四目系统的真实场景三维建模方法。针对双目立体视觉系统特征点匹配精度不高的问题,构建了一种四目立体视觉系统,并设计了与之相适应的高精度立体匹配策略,运用四目系统的冗余设计和自检验功能,可以获得高精度的空间点云数据。针对Delaunay三角化结果中不相关目标的错误连接问题,增加三角形边长分别在图像坐标系和世界坐标系中的约束条件,抑制错误三角形,该条件可根据点云数据分布特征自适应获取。该系统已被运用到莫高窟第172窟的三维重建中。结果表明:该系统可以重建多目标、深度变化大的复杂场景,172窟的有效点云为41 649点,三维建模时间约为0.5 h。     

单线阵CCD相机定位精度评估模型及几何误差研究

为满足单线阵CCD相机立体测绘定位的需求,研究了相机几何误差对定位精度的影响,并建立了它们之间的数值模型,为相机的几何指标分析和量化提供理论支持.首先针对单线阵立体成像特点,分析了几何标定和像点量测误差对定位结果的影响;然后以前方空间交会公式为基础,推导出定位精度评估模型.该模型给出了定位精度随几何标定和像点量测误差的...     

用于像移探测的位平面匹配改进算法

像移探测作为电子稳像的重要的步骤,直接影响稳像的精度和速度。针对传统位平面匹配算法的局限,对各阶位平面的探测效果进行了综合测试,并提出了一种改进方案,即对位平面4,5经过下采样后组成一个合成位平面来进行匹配。研究了匹配块大小变化对位平面算法的影响,比较了相邻帧和固定帧两种参考帧选取方法的优劣。实验表明,改进的位平面匹配算法提高了位平面算法的稳健性,在没有增加计算量的前提下,算法精度达到了与全搜索块匹配法相当的程度。     

基于邻域特征点提取和匹配的点云配准

为解决噪声干扰、数据丢失情况下迭代最近点算法的鲁棒性差、配准精度低等问题,提出一种基于邻域特征点提取和匹配的点云配准方法.首先定义一个由点的k邻域曲率、点与邻近点的法向量内积均值以及邻近点与邻域拟合平面的欧氏距离方差等三部分组成的邻域特征参数,结合在移动最小二乘表面构造的曲率特征参数对点云进行两次特征点提取;其次依据直方图特征定义三个匹配条件,并用双重约束获得正确的匹配点对;最后在配准阶段,采用双向构建k维树的迭代最近点算法实现精确配准.实验结果表明,该算法的配准精度较迭代最近点算法提高了90%以上,并且能够在噪声环境下有效地完成缺失点云的配准,在鲁棒性和精确配准方面有明显优势.     

一种稳健的特征点配准算法

为了能准确快速提取特征和可靠匹配特征点对,提出一种稳健的基于特征点的配准算法。首先改进了Plessey角点检测算法,有效提高所提取特征点的速度和精度。然后利用相似测度归一化互相关(Normalized cross correlation,NCC),通过双向最大相关系数匹配的方法提取出初始特征点对,用随机采样符合法(Random sample consensus,RANSAC)来剔除伪特征点对,实现特征点对的精确匹配。最后用正确匹配特征点对实现图像的配准。实验表明,该方法能够快速准确地提取两幅图像间的对应特征点,大大降低了误匹配的概率,两幅图像光照不一致、重复性纹理、旋转角度比较大等较难自动匹配情形下,仍能有效地实现图像的配准。     

直升机旋翼三维动态变形测量与可视化分析

针对直升机旋翼动载荷飞行试验测试需求,提出一种直升机旋翼三维动态变形测量与可视化分析方法。首先,根据直升机旋翼的结构特点和高速旋转特性,设计了以一组双目高清摄像头为核心的旋翼动态变形影像测量与监控及分析系统;其次,基于双目立体视觉测量理论,论述了测量系统标定、实时单点变形测量、散斑影像匹配、旋翼表面三维重建与三维动态变形可视化分析原理/方法;最后,在地面进行了模拟实测环境的仿真试验,实现了试验系统搭建、试验数据采集、处理与分析。试验结果证明,该方法可获得最大误差优于4 mm的定位测量精度,能很好地实现直升机旋翼三维动态变形测量,为飞行试验旋翼载荷测试数据分析等提供直观、可靠的数据与技术支撑。     

A real-time image matching algorithm for integrated navigation system

In this paper we introduced a kind of real-time image matching navigation algorithm based on ORB which can meet the requirement for high speed aircraft. We combined image matching with INS to fix the accumulative error of INS and get high accuracy position data of the aircraft. In order to improve the accuracy and robust of system, we used SIFT feature as a supplement of ORB and introduced the Kalman filter to optimize the output of image matching and INS. Based on maneuverability law, we did wrong point deleting with RANSAC. Finally, we designed a software system to simulate the image matching/INS navigation system. From the result of experiments, we can draw a conclusion that the algorithm has good real-time performance, play a good matching effect and get high accuracy position value.