CN102073846B - Method for acquiring traffic information based on aerial images - Google Patents

Method for acquiring traffic information based on aerial images Download PDF

Info

Publication number
CN102073846B
CN102073846B CN2010105888800A CN201010588880A CN102073846B CN 102073846 B CN102073846 B CN 102073846B CN 2010105888800 A CN2010105888800 A CN 2010105888800A CN 201010588880 A CN201010588880 A CN 201010588880A CN 102073846 B CN102073846 B CN 102073846B
Authority
CN
China
Prior art keywords
vehicle
road
aircraft
motion vector
direction
Prior art date
Application number
CN2010105888800A
Other languages
Chinese (zh)
Other versions
CN102073846A (en
Inventor
刘富强
李志鹏
龚剑
崔建竹
张姗姗
刘晓丰
Original Assignee
同济大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 同济大学 filed Critical 同济大学
Priority to CN2010105888800A priority Critical patent/CN102073846B/en
Publication of CN102073846A publication Critical patent/CN102073846A/en
Application granted granted Critical
Publication of CN102073846B publication Critical patent/CN102073846B/en

Links

Classifications

    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0108Measuring and analyzing of parameters relative to traffic conditions based on the source of data
    • G08G1/012Measuring and analyzing of parameters relative to traffic conditions based on the source of data from other sources than vehicle or roadside beacons, e.g. mobile networks
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/0104Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125Traffic data processing
    • G08G1/0133Traffic data processing for classifying traffic situation

Abstract

一种基于航拍图像的交通信息获取方法,根据航拍图像进行分析,检测动态目标和静态目标,其中,动态目标检测方法为采用KLT算法获得若干特征点以及动态目标的运动参数,通过道路区域方向投影确定运动方向矢量,采用k-medoids算法将若干特征点聚类,从而通过特征点将动态目标分离出来;静态目标检测方法为采用道路区域限制和斑块分析方法限定出道路区域和静态目标,在交通信息获取的基础上,通过对该信息的综合分析与处理,可以得到综合交通状况以及详细的交通参数指标。 Based on traffic information aerial image acquisition method, according to the aerial image analysis, detect the dynamic and static target targets, wherein moving object detection method for obtaining a plurality of feature points and the motion parameters of a dynamic target KLT algorithm is employed, the projection direction of the road by region determining the direction of motion vectors, using the k-medoids several features clustering algorithm to dynamically isolated by the target feature point; target detection method using static road area limitation and plaque out of the way analysis region defines a static target, in traffic on the basis of information obtained through comprehensive analysis and processing of this information, we can get a comprehensive traffic conditions as well as detailed traffic parameter index.

Description

基于航拍图像的交通信息获取方法 Traffic information based on aerial image acquisition method

技术领域 FIELD

[0001] 本发明属于图像处理技术领域,涉及ー种基于航拍图像的交通信息获取方法及系统。 [0001] The present invention belongs to the technical field of image processing, traffic information relates ー species based on aerial image acquisition method and system.

背景技术 Background technique

[0002] 稀疏路网交通状态检测是路网交通监控与预警的基础。 [0002] sparse road network traffic status detection is the basis of the road network traffic monitoring and early warning. 为了对交通事件和其他的交通事故等进行科学的、迅速的评测和预警,首先要对交通视频检测系统设备进行功能升级,使之能够适应西部稀疏路网的特殊地理、自然条件。 In order to traffic accidents and other incidents and other scientific, rapid evaluation and early warning, we must first feature upgrade for traffic video detection system equipment to enable them to adapt to the special geographical and natural conditions, sparse road network in the West. 然后需要研究基于视频的交通状态检测与识别算法,实现交通事件的有效检测,从而为路网安全预警提供可靠的信息。 You then need to research on the detection and video traffic state recognition algorithms to achieve effective traffic incident detection to provide reliable information for the road network safety warning.

[0003] 由于稀疏路网交通状态比较特殊,车流量相对较小,又没有足够人员去现场监控,所以亟需可以监控该区域的系统解决以下问题: [0003] Due to the sparse road network special status, traffic is relatively small, they do not have enough personnel to monitor the scene, so the need to monitor the system in the region to address the following questions:

[0004] (I)对于无人监控的道路区域需要对其交通情况进行监控,尽量减小交通故障带来的损失。 [0004] (I) need to monitor their traffic conditions for road area unattended, minimizing the loss of traffic caused by the fault.

[0005] (2)通过智能控制,減少人力资源占用,以优化人力配置。 [0005] (2) through intelligent control, reduce human resource consumption, to optimize staffing.

发明内容 SUMMARY

[0006] 本发明的目的提供一种基于航拍图像的交通信息获取方法,克服传统的基于斑块检测与跟踪的背景差方法针对动态背景的不适用,可适用背景变化及交通稀疏的场所。 [0006] The object of the present invention to provide a traffic information aerial image acquisition method, the background subtraction method to overcome the conventional plaque detection and tracking based on the dynamic background for NA applicable sparse traffic and background change based on properties.

[0007] 为达到以上目的,本发明所采用的解决方案是: [0007] To achieve the above object, the solution is used in the present invention:

[0008] ー种基于航拍图像的交通信息获取方法,根据航拍图像进行分析,检测动态目标和静态目标,提取道路交通參数; [0008] ー kinds of traffic information, aerial image acquisition method, analyzed according to aerial images, static and dynamic target detection target, based on the extraction of road traffic parameters;

[0009] 其中,动态目标检测方法为采用KLT算法获得若干特征点以及动态目标的运动參数,通过道路区域方向投影确定运动方向矢量,采用k-medoids算法将若干特征点聚类,从而通过特征点将动态目标分离出来; [0009] wherein the dynamic object detection method for obtaining a plurality of feature points and the motion parameters of a dynamic target KLT algorithm is adopted, the projected area is determined by the direction of motion vector direction of the road, using the k-medoids several features clustering algorithm, so that by the features separate dynamic target point;

[0010] 静态目标检测方法为采用道路区域限制和斑块分析方法限定出道路区域和静态目标。 [0010] The target detection method using static road area limitation and plaque analysis region out of the way, and defines a static target.

[0011] 进ー步,所述KLT算法是在航拍图像相邻帧之间的特征区域灰度相似的条件下,在灰度图像序列中选取大量的特征点,进行ニ维特征跟踪从而得到特征点位置,据此求解出ニ维特征运动參数的方法;同时,根据运动矢量的投影方向,提取出道路方向,井利用霍夫变换检测出道路线。 [0011] Step ー into the KLT algorithm is similar to the feature region gradation between adjacent aerial image frame condition, a large number of feature points selected in the gray image sequence, the writing is performed to obtain dimensional feature tracking feature position, whereby Ni-dimensional features a method of solving the motion parameters; Meanwhile, the projection direction of the motion vector extracted road direction, using the Hough transform detection wells debut route.

[0012] 所述k-medoids算法,包括以下步骤: [0012] The k-medoids algorithm, comprising the steps of:

[0013] (I)从特征点中任意选取K个对象作为Inedoids(C)1, O2,...01...0k); [0013] (I) is selected from any of the K feature point object as Inedoids (C) 1, O2, ... 01 ... 0k);

[0014] (2)将余下的对象根据与medoid最相近的原则分到各个类中去; [0014] (2) assigned to each of the remaining object classes according to the principles of the closest medoid;

[0015] (3)对于每个类(Oi)中,顺序选取一个も,计算用Or代替Oi后的消耗E(0ふ选择E最小的那个ル来代替Oi ; [0015] (3) (Oi) in order to select a mo each class, is calculated by Oi Or instead of after consumption E (0 fu selecting the minimum E Hikaru instead Oi;

[0016] (4)返回步骤⑵循环计算,直到K个medoids固定下来。 [0016] (4) returns to step ⑵ calculation cycle until the K medoids fixed. [0017] 所述道路区域限制是采用颜色直方图分布的方式将道路区域标识处理,结合霍夫变化检测出的道路线共同限定出道路区域。 [0017] The road area is restricted by way of the color histogram distribution road area identification process, in conjunction with road line Hough variation detected together define road area.

[0018] 所述斑块分析是具体获取斑块的中心点坐标、斑块面积、斑块的RGB信息和斑块数目等信息,斑块的中心点坐标反映静态目标的位置;斑块面积可用于去除噪声;斑块的RGB信息是在将航拍得到的图像进行二值化处理(即由彩色图像转换为只有黑白二色的图像)后,在此图像)中确定斑块的位置后,到原航拍图像中找到相应的斑块位置,统计RGB信息以区分出目标物体和周围环境;斑块数目用于对整个航拍图像中的斑块统一处理。 [0018] The analysis is specific plaque plaque acquired center point coordinates, center point coordinates plaque area, the RGB information, and the number of plaques and other information patches, plaques reflecting the position of the static object; plaque area available to remove noise; plaque after RGB information is binarization processing in the aerial image obtained (i.e., only black color image converted to two-color image), after the image is determined in the plaque) in the position to original aerial images to find the corresponding position of the plaque, to distinguish between statistics RGB information of the target object and the surroundings; patch number of aerial images for the entire plaque unitary.

[0019] 所述提取道路交通参数包括提取车辆行驶方向、车辆速度和车身长度。 [0019] extracting the parameters comprises extracting road traffic traveling direction of the vehicle, vehicle speed and vehicle length.

[0020] 所述车辆行驶方向,是根据车辆运动矢量与背景点运动矢量的大小关系来确定车辆的行驶方向,所述车辆运动矢量是指动态目标相对于飞机的运动矢量,所述背景点运动矢量是指静态目标相对于飞机的运动矢量,如果车辆运动矢量大于背景点运动矢量;那么车辆与飞机逆向行驶;如果车辆运动矢量小于背景点运动矢量,那么车辆与飞机同向行驶。 [0020] The vehicle traveling direction, the traveling direction of the vehicle is determined according to the magnitude relation between the vehicle and the motion vector of the background motion vector points, the dynamic vehicle motion vector refers to a motion vector with respect to the target aircraft, the background motion point refers to a static target vector with respect to the motion vector of the aircraft, if the vehicle is greater than the background motion vector is the motion vector points; then reverse-way traveling vehicle and aircraft; if the vehicle is less than the background motion vector is the motion vector points, the vehicle with the aircraft traveling the same direction. [0021 ] 所述车辆速度根据下式计算, The [0021] calculated according to the vehicle speed,

[0022] 车辆与飞机逆向行驶时,V车=V相对-V飞机; [0022] When the reverse driving of the vehicle with the aircraft, V = V relative vehicle -V aircraft;

[0023] 车辆与飞机同向行驶时,V车=V相对+V飞机; [0023] When a vehicle traveling in the same direction with the aircraft, V = V relative to vehicle + V plane;

[0024] 其中,¥¥是指动态目标的速度,Viw是指静态目标的速度,是指飞机的速度。 [0024] where, ¥¥ is the speed of moving targets, Viw is the speed of static target, it is the speed of the aircraft.

[0025]所述车身长度根据 length = (ymax-ymin) *cos θ * (Vplane*0.04/vectorbackground)计算,其为车辆在飞机飞行方向上的位移,θ为飞机飞行方向与车辆行驶方向的夹角,Vplane是飞机的飞行速度,vect0rbaekgMund为图像上背景点的相对运动速度,0.04为每帧之间相隔时间。 [0025] The length of the body length = (ymax-ymin) * cos θ * (Vplane * 0.04 / vectorbackground) calculated, which is the displacement of the vehicle in the direction of flight of the aircraft, the aircraft flight direction [theta] is the traveling direction of the vehicle interposed angle, Vplane aircraft flying speed, vect0rbaekgMund relative moving speed of the image background points, is separated by 0.04 time between each frame. 根据道路目标的特点,分别对运动车辆和静止车辆进行检测。 According to the characteristics of the target path, the motion-detection vehicles and stationary vehicle.

[0026] 动态目标检测方法 [0026] The moving object detection method

[0027] 对于运动车辆,采用KLT算法进行检测,KLT算法是在图像相邻帧之间的特征区域灰度相似的条件下,在灰度图像序列中选取大量的特征点,进行二维特征跟踪从而得到特征点位置,据此求解出二维特征运动参数的方法。 [0027] For the motion of the vehicle, using the detection algorithm KLT, KLT algorithm under similar conditions wherein the gradation region between adjacent image frames, a large number of feature points selected in the gray image sequence, two-dimensional feature tracking whereby the position of the feature point, a method for solving two-dimensional feature whereby motion parameters.

[0028] 同时,根据运动矢量的投影方向,可以提取出道路方向,以减小飞机抖动带来的误差。 [0028] Meanwhile, according to the projection direction of the motion vector, the direction of the road can be extracted, in order to reduce the error caused plane jitter. 利用霍夫变换检测出道路线。 Hough transform to detect debut route. 霍夫变换可以检测出图像中的所有直线(霍夫变换检测直线的方法是行业内公知的?最好能够简单介绍一下),由此计算出所有直线的斜率k和截距b,根据斜率求出分布最大的斜率范围并求出平均值kavg,由此求出道路线的方向并筛选掉其他无关直线。 Hough transform can detect all straight lines in the image (Hough transform to detect straight lines are well known in the industry? Best to brief), whereby all the calculated slope k and intercept of the straight line B, the slope seek the distribution of the slope range, and the maximum average value kavg, thereby obtaining the direction of the road line and filter out other irrelevant line. 由所得截距求出最大的截距bmax和最小的截距bmin可以得出道路线的两个边界,由此可以粗略估计出目标区域。 Determined by the intercept of the resulting maximum and minimum intercept intercept bmin bmax can draw two road boundary line, whereby the target region can be roughly estimated.

[0029] 在聚类算法中,最常用的是k-means算法。 [0029] In the clustering algorithm, it is the most common k-means algorithm. k_means算法接受输入量k ;然后将η个数据对象划分为k个聚类以便使得所获得的聚类满足:同一聚类中的对象相似度较高;而不同聚类中的对象相似度较小。 k_means algorithm accepts input k; then η data object into k clusters in order to satisfy such clusters obtained: the same cluster objects high similarity; different cluster similarity small objects . 聚类相似度是利用各聚类中对象的均值所获得一个“中心对象”(引力中心)来进行计算的。 Clustering the similarity is the use of objects in each cluster mean a "central object" (center of gravity) is obtained by calculation.

[0030] k-means算法的工作过程说明如下:首先从η个数据对象任意选择k个对象作为初始聚类中心;而对于所剩下其它对象,则根据它们与这些聚类中心的相似度(距离),分别将它们分配给与其最相似的(聚类中心所代表的)聚类;然后再计算每个所获新聚类的聚类中心(该聚类中所有对象的均值);不断重复这一过程直到标准测度函数开始收敛为止。 [0030] The processes and k-means algorithm is described as follows: First, select objects from any k η data objects as initial cluster centers; the left for other objects, according to their similarity to the centers of the clusters ( distance), respectively assigned to them most similar () represents the center of the cluster of the cluster; then calculate the new cluster center of each cluster obtained (average of all the objects in the cluster); repeated this process began until the standard measure function converges. 一般都采用均方差作为标准测度函数.k个聚类具有以下特点:各聚类本身尽可能的紧凑,而各聚类之间尽可能的分开。 Are generally used as the standard variance measure function .k clusters has the following characteristics: each cluster itself as compact as possible, as far as possible among the separated clusters.

[0031] 通过KLT算法可以得到若干特征点以及目标的运动參数,然后通过道路区域方向投影确定运动方向矢量,而k-means算法则是将若干特征点聚类,从而通过特征点将运动目标分离出来。 [0031] can be obtained by a number of feature points KLT algorithm and motion parameters of the target, and the direction of movement is determined by the projected area of ​​a road direction vector, and k-means clustering algorithm sucked several features, so that by moving the target feature point separate from.

[0032] 静态目标检测方法 [0032] Detection target static

[0033] 对于静止车辆,采用道路区域限制加斑块分析的方法,由于霍夫变换检测出来的直线不一定能够准确地描述道路线,因此需要结合其他算法共同完成道路区域的检测。 [0033] For a stationary vehicle, using the road area limit plus plaque analysis method, since the linear Hough transform can be detected is not necessarily accurate description of the road line, it is necessary in conjunction with other algorithms together to complete the detection of the road area. 一般来说,路面顔色分布相对比较固定,根据图像处理中已有的求取顔色直方图分布的方式,可以将道路区域标识出来,结合前边由霍夫变换方法检测出的道路线信息共同限定出道路区域。 In general, the color distribution of the road surface is relatively fixed, according to conventional image processing methods of deriving a color histogram distribution, it can be identified road area, the front binding detected by Hough transform information together define road line road area.

[0034] (I)斑块分析:在四连通环境下对斑块进行分析。 [0034] (I) Plaque Analysis: plaques analyzed under four communication environment. 所谓四连通,是指ー个象素点的上下左右方向四个紧邻的点与这个点是相邻关系,而左上、左下、右上、右下四个点与这个点不属于相邻关系。 The so-called four communication means ー a vertical and horizontal directions of the four pixel dots immediately adjacent to this point is a point adjacent relation, and the upper left, lower left, upper right, lower right four points to this point does not belong to the neighbor relations. 斑块信息越丰富,车辆检测和跟踪就越精确。 Information plaques richer, more accurate vehicle detection and tracking. 斑块分析提取的基本信息主要有:斑块的最上、下、左、右坐标。 Basic information extracted plaque analysis are: the most plaque, down, left and right coordinates. 利用这四个坐标,可以方便地计算斑块的中心点坐标和斑块的矩形面积。 Use the four coordinates, you can easily calculate the center point coordinates of a rectangular area of ​​plaque and plaque.

[0035] (2)斑块的中心点坐标 [0035] (2) the center point coordinates plaque

[0036] 中心点坐标可以粗略地反映车辆的位置,是车辆检测和跟踪的主要依据之一。 [0036] coordinates of the center can be roughly reflect the position of the vehicle, it is one of the main basis for vehicle detection and tracking.

[0037] (3)斑块面积 [0037] (3) patch area

[0038] 斑块面积包括实际面积和矩形面积。 [0038] The patch area comprises a rectangular area and the actual area. 利用面积信息可以去除一些噪声的影响。 Use area information may remove the effects of some of the noise. 进一步精确地得到斑块的位置。 More precisely obtain the position of the plaque.

[0039] (4)斑块的RGB信息 [0039] (4) RGB information plaque

[0040] 在ニ值图中确定斑块的位置后,到RGB图中找到相应的斑块位置,然后统计RGB信息。 [0040] After determining the position of plaque in FIG ni values ​​to find the corresponding RGB FIG plaque location, RGB information and statistics. RGB信息也是车辆检测和跟踪的主要依据之一。 RGB vehicle detection and tracking information is one of the main basis. 通过统计RGB信息,可以较为准确地区分出目标物体和周围环境,成为检测出静止车辆的重要步骤之一。 RGB statistical information, can be more accurately distinguish separate the target object and the surrounding environment, has become one of the important step of detecting a stationary vehicle.

[0041] (5)斑块数目 [0041] (5) The number of plaques

[0042] 整个ニ值图中,所有斑块的数目。 [0042] FIG whole values ​​ni, the number of all the patches. 得到斑块数目便于对整个图像中的斑块进行统 The number of plaques obtained facilitates the entire image system blobs

一处理。 A process.

[0043] 通过斑块分析,利用RGB顔色信息,可以初步确定静止车辆的位置,实现静止车辆的检测。 [0043] By analyzing plaque, using the RGB color information, can initially determine the position of the stationary vehicle, enable detection of a stationary vehicle.

[0044] 道路交通參数提取 [0044] road traffic parameter extraction

[0045] 对道路目标检测之后,可以提取相应參数来直观评价道路交通信息。 [0045] After the road for object detection, the corresponding parameter can be extracted intuitively evaluate traffic information.

[0046] 1、车辆的行驶方向 [0046] 1, the vehicle traveling direction

[0047] 根据车辆运动矢量与背景点运动矢量(即地面上的静止物体相对于飞机的运动矢量)的大小关系来确定车辆的行驶方向。 [0047] The vehicle motion vector and the background motion vector points (i.e., stationary object relative to the aircraft on the ground of the motion vector) to determine the magnitude relationship between the direction of the vehicle. 根据对视频的观察可以得到运动矢量的大小关系如下: Can be obtained based on observations of the motion vector magnitude relationship between the video as follows:

[0048] vector逆向行驶车辆> vector背景点> vector同向行驶车辆(I) [0048] vector reverse-way traveling vehicle> vector background points> vector with the same direction of the vehicle (I)

[0049] 首先,在所得到的特征点中,背景点是最多的,这样便可以得到vector背景点,然后再根据上式将每辆车的运动矢量与vector背景点进行比较确定车辆的行驶方向是与飞机同向还是与飞机逆向。 [0049] First, the feature points obtained, the background is the largest point, so that we can obtain a vector background points, and then compared to determine the traveling direction of the vehicle in accordance with the motion vector and vector background points of each vehicle It is the same or reverse direction with the aircraft and the aircraft. 如果车辆的运动矢量大于背景点运动矢量,那么车辆与飞机逆向行驶;如果车辆的运动矢量小于背景点运动矢量,那么车辆与飞机同向行驶。 If the vehicle is greater than the background motion vector is the motion vector points, the reverse-way traveling vehicle and aircraft; if the motion vector of the vehicle is less than the background motion vector points, the vehicle with the aircraft traveling the same direction.

[0050] 2、车辆速度值的提取 [0050] 2, a vehicle speed value extracted

[0051] 由于飞机的飞行速度是给定的,而背景特征点的运动矢量α飞机的飞行速度(给定),同时车辆特征点的运动矢量~车辆相对于飞机的相对速度,这样,便可以利用比例关系得出飞机的相对速度,再利用下式推算出车辆的实际行驶速度: [0051] As the aircraft airspeed is given, and the moving background feature point vector α aircraft flying speed (given), while the vehicle feature point motion vectors ~ vehicle relative speed of the aircraft, so that they can derived using the ratio between the relative speed of the aircraft, and then using the following formula calculate the vehicle's actual driving speed:

[0052] V车=V相对-V飞机(车辆与飞机逆向行驶) ⑵ [0052] V = V relative to the vehicle -V aircraft (reverse-way traveling vehicle and aircraft) ⑵

[0053] V车=V相对+V飞机(车辆与飞机同向行驶) ⑶ [0053] V = V relative to vehicle + V aircraft (aircraft and vehicles traveling the same direction) ⑶

[0054] 3、车型的提取 [0054] 3, extracted models

[0055] 车身长度: [0055] Body length:

[0056] length = (ymax-ymin) *cos Θ *(Vplane*0.04/vectorbackground) (4) [0056] length = (ymax-ymin) * cos Θ * (Vplane * 0.04 / vectorbackground) (4)

[0057] 其中ymax_ymin为车辆在飞机飞行方向上的位移,Θ为飞机飞行方向与车辆行驶方向的夹角,Vplane是飞机的飞行速度,vectorbaekg_d为图像上背景点的相对运动速度,0.04为每帧之间相隔时间。 [0057] wherein ymax_ymin vehicle displacement direction in the aircraft, the aircraft flight direction [Theta] is the angle of the vehicle traveling direction, Vplane aircraft flying speed, vectorbaekg_d relative moving speed of the image background points, 0.04 per frame separated by time between.

[0058] 上式同样是利用已知的飞机速度,将相邻两帧之间飞机飞行的实际距离与图像的像素单位对应起来,从而估算出车身的长度。 [0058] The above formula is also known the use of aircraft speed, the actual distance of adjacent unit pixels of the image of the aircraft association between two, to estimate the length of the vehicle body.

[0059] 以上过程都是根据航拍图像,利用计算机图像技术进行处理,针对不同参数通过不同方法提取而实现的,最终得到本系统所需的各项参数,包括车辆行驶方向、速度以及车身长度。 [0059] The aerial image is the above process, the use of computer image processing technology, extraction is achieved by different methods for different parameters, the parameters that are required to obtain the final system, comprising a vehicle running direction, speed, and length of the body.

[0060] 基于航拍图像的交通信息获取系统 [0060] acquisition system based on aerial images of traffic information

[0061] 该系统通过对无人机航拍视频进行处理,可以获取航拍路段上的交通量信息,和每辆车的世界坐标、行驶速度等信息。 [0061] The system by UAV aerial video processing, you can get traffic information on aerial sections, and each vehicle's information world coordinates, speed and so on. 在交通信息获取的基础上,通过对该信息的综合分析与处理,如对车辆数目的统计,对车速的测量等,得出该地区路面的综合交通信息以及详细的交通参数。 On the basis of the traffic information obtained on a comprehensive analysis and processing of the information, such as statistics on the number of vehicles, vehicle speed and other measurements, obtained comprehensive traffic information in the area of ​​road transport as well as the detailed parameters.

[0062] 由于采用了上述方案,本发明具有以下特点:不仅能够应用在一般交通流量下路况的分析,特别能够针对西部地区的稀疏路段进行交通状况分析,得到所需交通参数和综合路况。 [0062] As a result of the above, the present invention has the following characteristics: not only can be used in the analysis of general road traffic, especially the traffic situation can be analyzed for the sparse sections of the western region to give the desired parameters and integrated road traffic. 且由于所得图像来源于无人机航拍,本发明更能够适应采集图像的变化。 And because the resulting aerial image derived from a UAV, the present invention is more able to adapt to changes in image acquisition.

附图说明 BRIEF DESCRIPTION

[0063] 图1是本发明方法的一种实施例的流程示意图。 [0063] FIG. 1 is a flow diagram of a method according to embodiments of the invention.

[0064] 图2是本发明方法的一种实施例的运动目标特征检测跟踪方法流程图。 [0064] FIG. 2 is a moving target detection and tracking features of the method of the present invention a method of an embodiment of a flowchart.

[0065] 图3是本发明方法的一种实施例的静止目标检测方法流程图。 [0065] FIG. 3 is a flowchart of a stationary target detection method of the present invention, a method embodiment.

具体实施方式 Detailed ways

[0066] 以下结合附图所示实施例对本发明作进一步的说明。 [0066] Hereinafter, the present invention will be further described in conjunction with the embodiment shown in the accompanying drawings.

[0067] 本系统通过摄像头在无人机上航拍获取视频,对视频进行相关预处理后得到需要的输入视频。 [0067] The system in aerial video acquired by a camera on the UAV, to get the desired input video related to the video preprocessing. 如图1所示,即为采集视频的过程。 As shown in FIG. 1, that is, the video acquisition process. 接下来,需要对视频进行一系列相关处理以得到所需内容。 Next, we need to perform a series of related video processing to achieve the desired content.

[0068] 首先,将视频截成帧,每一帧相当于一幅图像,然后针对每一幅图像进行处理。 [0068] First, the cut video frames, each frame corresponding to an image, and then processed for each image. 下面分别阐述针对运动目标检测和静止目标检测的不同方法。 Following different methods for the stationary target detection and moving object detection are described. [0069] 对于运动目标,第一歩需要得到其特征点。 [0069] For the moving object, a first feature point ho need thereof. 对于每ー帧图像,如图2,利用KLT算法对图像进行扫描获取特征点,KLT算子首先计算每个特征点的8矩阵的特征值X 1、X 2,如果min(入p A2) > H(阈值)(一般来说阈值为经验值)则该点为有效特征点。ー for each frame image, FIG. 2, the image is scanned using a feature point acquiring KLT algorithm, KLT operator first calculates the feature of each feature point 8 matrix values ​​X 1, X 2, if min (the p A2)> H (threshold) (General empirical threshold value) the feature point is a valid point. S矩阵定义为 Matrix S is defined as

[0070] [0070]

Figure CN102073846BD00071

[0071] 式中:IX为ー阶X方向导数,Iy为ー阶y方向导数。 [0071] wherein: IX is ー order derivative X direction, Iy is ー order derivatives y direction.

[0072] 随着相机的移动,图像的强度以复杂的方式发生着变化。 [0072] With the movement of the camera, the intensity of the image is changing in a complex manner. 如果摄像机捕获图像的速度足够快,那么对于相邻帧而言,由于各种影响灰度变化因素的相似性,在局部区域内的灰度变化是极其相似的,因此可以认为,相邻两帧的局部区域之间存在沿X和Y方向的位移,这就是所谓的ニ维特征平移运动模型。 If the image captured by the camera speed fast enough, for adjacent frames, variations due to the similarity factors gray, gray-scale variation in the local region is very similar, it is considered that two adjacent frames the presence in the X and Y directions between the local area displacement, which is called ni-dimensional feature translational motion model. 这意味着t时刻图像上的某个特征点X= (x,y)在t+1时刻运动到了X' = (x-dx,y_dy),其中d = (dx,dy)为ニ维特征的平移运动參数向量,该特征点的灰度值在运动前后是近似相等的,即 This means that a time t on the image feature points X = (x, y) at time t + 1 is moved to the X '= (x-dx, y_dy), where d = (dx, dy) of the Ni-dimensional feature parameter translational motion vector, the feature point gray value before and after exercise is approximately equal, i.e.,

[0073] J ⑴=I (Xd) +n (X) (6) [0073] J ⑴ = I (Xd) + n (X) (6)

[0074] 其中:J(X) = I (X,t+1)为t+1时刻的特征点X的灰度值,I (Xd) = I (Xd, t)为t时刻该特征点的灰度值,n(X)为相应的噪声。 [0074] wherein: J (X) = I (X, t + 1) is characterized by time t + 1 of point X gradation values, I (Xd) = I (Xd, t) of the feature point in time t gradation value, n (X) is a corresponding noise.

[0075] 显然需要选择合适的运动參数向量d使在特征点X周围的某个特征窗ロW内如下的二重积分得到的残差最小 [0075] Obviously to select the proper parameters of the motion vector d in that the minimum residual wherein a window around the feature points X ro W follows the double integration obtained

[0076] e = / ff(I (Xd)-J(X))2 GJdX (7) [0076] e = / ff (I (Xd) -J (X)) 2 GJdX (7)

[0077] 式中Co为对特征区域内不同象素点的加权方程。 [0077] Co in the formula is a weighted equation for different points within the feature region pixels. 如果相邻两帧之间的运动比较小的情况下,可以将I (X- d)在X点进行ー阶泰勒展开 If the motion between two adjacent relatively small, may be I (X- d) for ー order Taylor expansion point X

[0078] I (Xd) = J ⑴ _g • d (8) [0078] I (Xd) = J ⑴ _g • d (8)

[0079] g是梯度向量,于是可以将式(7)重新写成如下的形式 [0079] g is the gradient vector, can then be re-written in the following form of the formula (7)

[0080] e =/ w(I (Xd)-JOO)2 wdX = / w(h_g • d)2 wdX (9) [0080] e = / w (I (Xd) -JOO) 2 wdX = / w (h_g • d) 2 wdX (9)

[0081] 其中h = I (X)-J(X)。 [0081] where h = I (X) -J (X). 可以看出,残差是平移向量d的二次方程,这个最优化问题可以得到闭合形式的解。 As can be seen, the translation is quadratic residual vector d, closed-form solutions of the optimization problem can be obtained. 为了使残差最小,对式(9)等号两边,求其对d的ー阶导数,得到 To minimize residual, equal on both sides of the formula (9), find the first derivative of the ー d, to give

[0082] / w (hg • d) go dX = 0 (10) [0082] / w (hg • d) go dX = 0 (10)

[0083] 由于(g*d)g= (ggT)d,而且在特征窗ロ区域中假设d为常量,因此得到 [0083] Since (g * d) g = (ggT) d, and d is assumed to be constant in the region characterized ro window, and therefore to give

[0084] dX / w(ggT) w dX = / ffhg w dX (11) [0084] dX / w (ggT) w dX = / ffhg w dX (11)

[0085] 上式是特征跟踪计算中的基本计算步骤,对于特征窗口内的所有像素都可以计算出其沿X和Y方向的梯度,因此可以得到实对称的交叉梯度矩阵G,同时对于特征窗口内的所有像素都能计算出两帧之间灰度差并得到向量e。 [0085] The above formula is a basic calculation step wherein tracing calculation, all pixels for the feature of the window can be calculated along the X and Y direction gradient can be obtained real symmetric cross gradient matrix G, while for the feature window all the pixels in the gray scale can be calculated and a difference between the obtained two vector e. 这样就能计算出运动參数d的值。 This will calculate the value of the motion parameter d. 得到d以后,移动特征窗ロ,再重复以上过程,直到d小于某个阈值,这表明相邻两帧的特征窗ロ已经匹配成功,将前面重复过程中每ー轮得到的d都加起来就得到最終的平移运动參数。 After getting d, wherein moving the window ro, then repeat the process until d is less than a certain threshold, indicating that two adjacent window frames characterized ro successfully paired, the foregoing process is repeated for each wheel obtained ー d add up to to get the final translational motion parameters.

[0086] 第二步,为了将特征点聚类,以准确将运动目标分离出来,需要采用有效的聚类算法。 [0086] The second step, in order to cluster the feature point, to accurately separate the moving object, need efficient clustering algorithm. 传统的聚类算法应用较多的是k-means算法。 Traditional clustering algorithm is more k-means algorithm.

[0087] k-means算法的工作过程说明如下:首先从n个数据对象任意选择k个对象作为初始聚类中心;而对于所剩下其它对象,则根据它们与这些聚类中心的相似度(距离),分别将它们分配给与其最相似的(聚类中心所代表的)聚类;然后再计算每个所获新聚类的聚类中心(该聚类中所有对象的均值);不断重复这一过程直到标准测度函数开始收敛为止。 [0087] The processes and k-means algorithm is described as follows: First, from the n selected data objects as objects any k initial cluster centers; for other objects are left, the center of the clusters based on their similarity with ( distance), respectively assigned to them most similar () represents the center of the cluster of the cluster; then calculate the new cluster center of each cluster obtained (average of all the objects in the cluster); repeated this process began until the standard measure function converges. 一般都采用均方差作为标准测度函数.k个聚类具有以下特点:各聚类本身尽可能的紧凑,而各聚类之间尽可能的分开。 Are generally used as the standard variance measure function .k clusters has the following characteristics: each cluster itself as compact as possible, as far as possible among the separated clusters.

[0088] k-means有其缺点:产生类的大小相差不会很大,对于脏数据很敏感。 [0088] k-means has its disadvantages: class size have no great difference, it is sensitive to the dirty data. 基于这一点,提出了一种改进的算法:k-medoids方法。 Based on this, an improved algorithm: k-medoids method.

[0089] K-medoids算法选取一个对象叫做medoid来代替上面的中心的作用,这样的一个medoid就标识了这个类。 [0089] K-medoids select an object called medoid algorithm instead of the central role of the above, such a medoid identifies the class. 步骤: step:

[0090] (I)从特征点中任意选取K个对象作为medoids (O1, O2,...01...0k)。 [0090] (I) from the feature point selected as the objects of any K medoids (O1, O2, ... 01 ... 0k).

[0091] 以下是循环的: [0091] The following is cyclic:

[0092] (2)将余下的对象分到各个类中去(根据与medoid最相近的原则); [0092] (2) the remaining objects assigned to each class (medoid principles and according to the most similar);

[0093] (3)对于每个类(Oi)中,顺序选取一个0r,计算用Or代替Oi后的消耗-E(Or)。 [0093] (3) for each class (Oi), select a sequence 0r, instead of consumption calculation Or -E (Or) after Oi. 选择E最小的那个O,来代替O”这样K个medoids就改变了,下面就再转到(2)。 Selecting the minimum E O, which uses instead of the O "so that the K medoids changed, and then go to the following (2).

[0094] (4)这样循环直到K个medoids固定下来。 [0094] (4) the cycle until the K medoids fixed.

[0095] 这种算法对于脏数据和异常数据不敏感,但计算量显然要比K均值要大,一般只适合小数据量。 [0095] This algorithm is not sensitive to dirty data and abnormal data, but the amount of calculation is clearly better than K-means to a large, generally only suitable for a small amount of data.

[0096] 基于密度的k-medoids算法是在聚类过程中,不需要输入聚类的个数,而是根据数据之间的距离将相邻类进行合并。 [0096] k-medoids density clustering algorithm process is not necessary to enter the number of clusters, but the distance between the merged data adjacent classes.

[0097] 第三步,道路交通参数的提取。 [0097] The third step is to extract road traffic parameters.

[0098] I)运动方向的提取。 [0098] I) the extracted moving direction. 由前边得到的目标运动矢量以及背景点相对于飞机的运动矢量,根据(I)式可以得出运动方向。 Obtained from the front and the background motion vector of a target point with respect to the motion vector of the aircraft, according to formula (I) direction of motion can be derived.

[0099] 2)车辆速度值的提取。 Extraction [0099] 2) the vehicle speed value. 由运动矢量得出的相对速度以及从航拍中获得的飞机速度可以根据式⑵和式⑶得出车辆速度。 Derived by the motion vector and the aircraft speed relative speed obtained from the aerial vehicle speed can be derived according to Formula and Formula ⑵ ⑶.

[0100] 3)车型提取。 [0100] 3) extraction models. 根据相应点的坐标以及运动矢量由式⑷可以获得车辆长度,并进行分类提取出车型。 The coordinates of the corresponding points and the motion vector can be obtained by the formula ⑷ vehicle length, and extract the classified models.

[0101] 对于静止车辆,需要得到车辆目标的位置信息。 [0101] For a stationary vehicle, the vehicle needs to be the target position information.

[0102] 如图3所示,首先进行道路区域选定,一般来说,路面颜色分布相对比较固定,根据颜色直方图分布,可以将道路区域标识出来,结合前边检测出的道路线信息共同限定出道路区域。 [0102] As shown in FIG. 3, first, a road area is selected, in general, the color distribution of the road surface is relatively fixed, according to the color distribution histogram, out of the road area can be identified, in conjunction with the front information detected by the road line together define out of the way area.

[0103] 其次,采用斑块分析的方法确定静止目标位置(坐标)。 [0103] Secondly, the analysis method for determining plaque stationary target position (coordinates).

[0104] 对于斑块分析方法,共需两次扫描,第一次扫描从下往上、从左往右扫描,扫描到象素值为O的点,不做任何处理。 [0104] For the analysis of plaque, totaling two scan, the first scan from bottom to top, from left to right scan, the scanning point O of the pixel values, without any treatment. 当扫描到第一个象素值为255的点,把它标记为“1”,并记录下来。 When scanning a pixel value 255 to the first point, it is marked as "1", and recorded. 然后扫描该点的四个邻居,如果有象素值为255的点,把它标记为与该点相同的标记“1”,表示它们属于同一个斑块“I”。 Then scan the four neighbor points, if the pixel value of 255 points, it is marked as the same point flag "1" to indicate that they belong to the same patch "I". 最后记录该点的坐标值,以及记录“I号斑块目前有I个象素”。 Recording of the coordinate value of the last point, and recording "I No. I currently plaques pixels." 然后继续扫描,重复上面的步骤。 Then continue scanning, repeat the above steps.

[0105] 第二次扫描主要完成的是标记修正。 [0105] The second main scanning completion flag is corrected.

[0106] 在合并象素的同时,这次扫描也完成了信息的统计,包括斑块的面积,最上、下、左、右的坐标,中心点坐标,RGB彳目息等。 [0106] In the combined pixels at the same time, the scans also completed statistical information, including plaque area, the top, bottom, left and right coordinates, center point coordinates, RGB left foot head interest rates and so on.

[0107] 所有检测结束后,根据前述方法提取出所需各交通参数,完成系统功能。 [0107] After all the detectors, the desired method of extraction according to various traffic parameters, complete the system function. [0108] 本系统能够有效地获得某一路段上的交通流量、车速、车型等交通參数和交通状态信息,是ー种综合性的系统,包括视频输入设备,视频分析工具(软件)以及结果显示设备(计算机),通过系统分析处理将结果显示出来,给相关人员或部门提供所需信息。 [0108] The present system can effectively obtain traffic, speed, and other vehicle parameters and traffic information on the traffic status of a road, a kind ー integrated system, comprising a video input devices, video analysis tools (software) and the results display device (computer), through system analysis results are displayed, provide the required information to the appropriate person or department.

[0109] 上述的对实施例的描述是为便于该技术领域的普通技术人员能理解和应用本发明。 [0109] Description of embodiments described above for ease of ordinary skill in the art to understand and apply the invention. 熟悉本领域技术的人员显然可以容易地对这些实施例做出各种修改,并把在此说明的一般原理应用到其他实施例中而不必经过创造性的劳动。 A person skilled in art may readily apparent that various modifications to these embodiments, and the generic principles described herein apply to other embodiments without going through creative work. 因此,本发明不限于这里的实施例,本领域技术人员根据本发明的掲示,对于本发明做出的改进和修改都应该在本发明的保护范围之内。 Accordingly, the present invention is not limited to the embodiments herein, those skilled in the art in accordance with the present invention shown kei, improvements and modifications to the present invention should be made within the scope of the present invention.

Claims (4)

1.一种基于航拍图像的交通信息获取方法,其特征在于:根据航拍图像进行分析,检测动态目标和静态目标,提取道路交通参数;所述提取道路交通参数包括提取车辆行驶方向、车辆速度和车身长度; 其中,动态目标检测方法为采用KLT算法获得若干特征点以及动态目标的运动参数,通过道路区域方向投影确定运动方向矢量,采用k-medoids算法将若干特征点聚类,从而通过特征点将动态目标分离出来; 静态目标检测方法为采用道路区域限制和斑块分析方法限定出道路区域和静态目标; 所述KLT算法是在航拍图像相邻帧之间的特征区域灰度相似的条件下,在灰度图像序列中选取大量的特征点,进行二维特征跟踪从而得到特征点位置,据此求解出二维特征运动参数的方法;同时,根据运动矢量的投影方向,提取出道路方向,并利用霍夫变换检测出道路线; 所 An aerial image based on the traffic information acquisition method, wherein: according aerial image analysis, detect the dynamic and static target targets, extract road traffic parameters; extracting the parameters comprises extracting road traffic traveling direction of the vehicle, and the vehicle speed the length of the body; wherein the dynamic object detection method using KLT algorithm to obtain a plurality of feature points and the motion parameters of moving targets, the projected direction of motion is determined by the vector direction of the road area, using k-medoids several features clustering algorithm, so that by the feature point separating out the dynamic object; static target detection method employed analysis of plaque area limitation and road area defining a static target road; KLT algorithm is similar to the feature region gradation between adjacent frames aerial images under conditions select a large number of feature points in the gray image sequence, to thereby obtain a two-dimensional feature tracking characteristic points, a method for solving two-dimensional feature whereby motion parameters; Meanwhile, the projection direction of the motion vector extracted road direction, detection and Hough transform debut line; the 述k-medoids算法,包括以下步骤: (1)从特征点中任意选取K个对象作为medoids (O1, O2,...<V..0k); (2)将余下的对象根据与medoid最相近的原则分到各个类中去; (3)对于每个类(Oi)中,顺序选取一个Op计算用Or代替Oi后的消耗E (0J,选择E最小的那个A来代替0i; (4)返回步骤(2)循环计算,直到K个medoids固定下来; 其中,所述道路区域限制是采用颜色直方图分布的方式将道路区域标识处理,结合霍夫变化检测出的道路线共同限定出道路区域; 所述斑块分析是具体获取斑块的中心点坐标、斑块面积、斑块的RGB信息和斑块数目信息,斑块的中心点坐标反映静态目标的位置;斑块面积用于去除噪声;斑块的RGB信息是在将航拍得到的图像进行二值化处理后,在此图像中确定斑块的位置,然后到原航拍图像中找到相应的斑块位置,统计RGB信息以区分出目标物体和周围环境;斑块数目用 Said k-medoids algorithm, comprising the steps of: (1) select from the feature point objects as arbitrary K medoids (O1, O2, ... <V..0k); (2) the remaining objects according to the most medoid similar principles assigned to each class; (3) for each class (Oi) in the order of the selected Op calculating a consumption Or E replaced by Oi (0J, selecting the minimum E a instead 0i; (4 ) returns to step (2) calculation cycle until the K medoids fixed; wherein said area limiting road is a road marking treatment region by way of a color histogram distribution, binding Hough variation detected road line together define road region; plaque analysis of the specific center point coordinates acquired plaque, plaque area, the center point coordinates and RGB information plaque patch number information plaques reflect the location of the static target; for removing plaque area noise; after plaque RGB information is binarization processing in the aerial image obtained, determining the position of the plaque in this image, and then to find the corresponding original aerial images of the position of plaque, to distinguish the RGB information statistics a target object and the surroundings; patch number with 于对整个航拍图像中的斑块统一处理。 To the unified treatment of the entire aerial images of plaques.
2.如权利要求1所述的基于航拍图像的交通信息获取方法,其特征在于:所述车辆行驶方向,是根据车辆运动矢量与背景点运动矢量的大小关系来确定车辆的行驶方向,所述车辆运动矢量是指动态目标相对于飞机的运动矢量,所述背景点运动矢量是指静态目标相对于飞机的运动矢量,如果车辆运动矢量大于背景点运动矢量;那么车辆与飞机逆向行驶;如果车辆运动矢量小于背景点运动矢量,那么车辆与飞机同向行驶。 1, 2. The traffic information based on the aerial image acquisition method, as claimed in claim wherein: the vehicle traveling direction, the traveling direction of the vehicle is determined according to the magnitude relation between the vehicle and the motion vector the motion vector of the background points, the vehicle dynamic motion vector refers to a motion vector with respect to the target aircraft, the background motion vector refers to a static point target with respect to aircraft motion vectors, the motion vector is greater than the background if the vehicle motion vector points; then reverse-way traveling vehicle and aircraft; if the vehicle motion vector is less than the background motion vector points, then the vehicle with the aircraft traveling the same direction.
3.如权利要求1所述的基于航拍图像的交通信息获取方法,其特征在于:所述车辆速度根据下式计算, 车辆与飞机逆向行驶时,V车=V相对-V飞机; 车辆与飞机同向行驶时,V车=V相对+V飞机; 其中,^¥是指动态目标的速度,乂„是指静态目标的速度,是指飞机的速度。 3. The aerial images based on the traffic information acquisition method according to claim 1, wherein: the vehicle speed is calculated according to the formula, when the reverse driving of the vehicle with the aircraft, V = V relative vehicle -V aircraft; plane of the vehicle and when traveling in the same direction, V = V vehicle relative + V aircraft; which, ^ ¥ is the speed of moving targets, qe "refers to the speed of static target, is the speed of the aircraft.
4.如权利要求1所述的基于航拍图像的交通信息获取方法,其特征在于:所述车身长度根据length= (y.-yj^co s θ * (Vplane*0.04/vectorbackground)计算,其中 为车辆在飞机飞行方向上的位移,Θ为飞机飞行方向与车辆行驶方向的夹角,Vpl■是飞机的飞行速度,VeCt0rbac;kg_d为图像上背景点的相对运动速度,0.04为每帧之间相隔时间。 4. The aerial images based on the traffic information acquisition method according to claim 1, wherein: the length of the vehicle body calculated length = (y.-yj ^ co s θ * (Vplane * 0.04 / vectorbackground), which is displacement of the vehicle in the direction of flight of the aircraft, the flight direction [Theta] is the angle between the plane of the vehicle traveling direction, Vpl ■ the aircraft flying speed, VeCt0rbac; kg_d relative moving speed of the image background point is spaced between 0.04 per frame time.
CN2010105888800A 2010-12-15 2010-12-15 Method for acquiring traffic information based on aerial images CN102073846B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010105888800A CN102073846B (en) 2010-12-15 2010-12-15 Method for acquiring traffic information based on aerial images

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010105888800A CN102073846B (en) 2010-12-15 2010-12-15 Method for acquiring traffic information based on aerial images

Publications (2)

Publication Number Publication Date
CN102073846A CN102073846A (en) 2011-05-25
CN102073846B true CN102073846B (en) 2013-06-05

Family

ID=44032380

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010105888800A CN102073846B (en) 2010-12-15 2010-12-15 Method for acquiring traffic information based on aerial images

Country Status (1)

Country Link
CN (1) CN102073846B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015134453A1 (en) * 2014-03-03 2015-09-11 Inrix Inc. Estimating traveler volume by evaluating aerial images

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103366555B (en) * 2013-07-01 2015-06-17 中国人民解放军第三军医大学第三附属医院 Aerial image-based traffic accident scene image rapid generation method and system
CN103413444B (en) * 2013-08-26 2015-08-19 深圳市川大智胜科技发展有限公司 Traffic Flow method based on the investigation and handling of high-definition video UAV
US9175966B2 (en) * 2013-10-15 2015-11-03 Ford Global Technologies, Llc Remote vehicle monitoring
BR102013033041A2 (en) * 2013-12-20 2017-02-14 Perkons S/A system and method of monitoring and enforcing traffic and recording traffic violations and corresponding unmanned aerial vehicle
CN104035985B (en) * 2014-05-30 2017-07-07 同济大学 A kind of method for digging towards Fundamental Geographic Information System abnormal data
CN105046952B (en) * 2015-06-17 2017-12-08 天津职业技术师范大学 The sparse air-ground associated detecting method of road traffic accident
CN108009590A (en) * 2017-12-13 2018-05-08 广东机场白云信息科技有限公司 Flight information display system and its terminal equipment failure display methods

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101368828A (en) 2008-10-15 2009-02-18 同济大学 Blind man navigation method and system based on computer vision
CN101770581A (en) 2010-01-08 2010-07-07 西安电子科技大学 Semi-automatic detecting method for road centerline in high-resolution city remote sensing image

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101368828A (en) 2008-10-15 2009-02-18 同济大学 Blind man navigation method and system based on computer vision
CN101770581A (en) 2010-01-08 2010-07-07 西安电子科技大学 Semi-automatic detecting method for road centerline in high-resolution city remote sensing image

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2015134453A1 (en) * 2014-03-03 2015-09-11 Inrix Inc. Estimating traveler volume by evaluating aerial images
EP3114663A4 (en) * 2014-03-03 2018-03-14 Inrix, Inc. Estimating traveler volume by evaluating aerial images

Also Published As

Publication number Publication date
CN102073846A (en) 2011-05-25

Similar Documents

Publication Publication Date Title
JP5325899B2 (en) Intrusion alarm video processor
EP2798611B1 (en) Camera calibration using feature identification
DE112016005059T5 (en) Subcategory aware folding neural networks for object detection
TWI452540B (en) Image based detecting system and method for traffic parameters and computer program product thereof
US20120062736A1 (en) Hand and indicating-point positioning method and hand gesture determining method used in human-computer interaction system
Zhou et al. Efficient road detection and tracking for unmanned aerial vehicle
Marin et al. Learning appearance in virtual scenarios for pedestrian detection
US8340420B2 (en) Method for recognizing objects in images
US9846946B2 (en) Objection recognition in a 3D scene
DE102013206009A1 (en) Robust cutting of license plate images
CN102663377B (en) Character recognition method based on template matching
CN103149939A (en) Dynamic target tracking and positioning method of unmanned plane based on vision
TW201324383A (en) Method and apparatus for video analytics based object counting
CN101789080B (en) Detection method for vehicle license plate real-time positioning character segmentation
WO2012139228A1 (en) Video-based detection of multiple object types under varying poses
US20110025834A1 (en) Method and apparatus of identifying human body posture
CN1984236A (en) Method for collecting characteristics in telecommunication flow information video detection
CN1897015A (en) Method and system for inspecting and tracting vehicle based on machine vision
CN103049751A (en) Improved weighting region matching high-altitude video pedestrian recognizing method
US8902053B2 (en) Method and system for lane departure warning
CN103366156A (en) Road structure detection and tracking
CN105354568A (en) Convolutional neural network based vehicle logo identification method
Chen et al. Vehicle detection in high-resolution aerial images via sparse representation and superpixels
Hadi et al. Vehicle detection and tracking techniques: a concise review
CN103455984A (en) Method and device for acquiring Kinect depth image

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C14 Grant of patent or utility model