WO2012100522A1 - 基于图像亮度特征的ptz视频能见度检测方法 - Google Patents

基于图像亮度特征的ptz视频能见度检测方法 Download PDF

Info

Publication number
WO2012100522A1
WO2012100522A1 PCT/CN2011/078247 CN2011078247W WO2012100522A1 WO 2012100522 A1 WO2012100522 A1 WO 2012100522A1 CN 2011078247 W CN2011078247 W CN 2011078247W WO 2012100522 A1 WO2012100522 A1 WO 2012100522A1
Authority
WO
WIPO (PCT)
Prior art keywords
brightness
image
road surface
visibility
camera
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2011/078247
Other languages
English (en)
French (fr)
Inventor
李勃
郁健
张潇
董蓉
江登表
陈钊正
陈启美
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University
Nanjing Tech University
Original Assignee
Nanjing University
Nanjing Tech University
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 Nanjing University, Nanjing Tech University filed Critical Nanjing University
Priority to US13/981,163 priority Critical patent/US9225943B2/en
Publication of WO2012100522A1 publication Critical patent/WO2012100522A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/47Scattering, i.e. diffuse reflection
    • G01N21/49Scattering, i.e. diffuse reflection within a body or fluid
    • G01N21/53Scattering, i.e. diffuse reflection within a body or fluid within a flowing fluid, e.g. smoke
    • G01N21/538Scattering, i.e. diffuse reflection within a body or fluid within a flowing fluid, e.g. smoke for determining atmospheric attenuation and visibility
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00Television systems
    • H04N7/18Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10016Video; Image sequence
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30248Vehicle exterior or interior
    • G06T2207/30252Vehicle exterior; Vicinity of vehicle

Definitions

  • the invention belongs to the technical field of image processing, and relates to a camera calibration technology, an adaptive Nagao filtering technology, a noise suppression technology, a visibility solving technology, etc., and on the basis of ensuring the height and illuminance consistency of selected pixels, extracting and reflecting the change of road surface brightness. Contrast curve, find the characteristic point of the brightness curve, calculate the farthest pixel that can be distinguished by the human eye in the image by the extinction coefficient; convert the maximum visible distance with the camera calibration to determine the visibility value, specifically a PTZ video visibility detection based on the image brightness feature method. Background technique
  • the expressway is an emerging industry involving the national economy and the people's death.
  • the total mileage plan of China will exceed 80,000 kilometers.
  • Jiangsu is an economically developed region in China and a leader in the Intelligent Transportation System (ITS).
  • ITS Intelligent Transportation System
  • the density of expressways is 3.46 kilometers / 100 square kilometers.
  • the total mileage will reach 5,000 kilometers.
  • the meteorological visibility detection equipment used in expressways is mainly based on traditional laser visibility instruments.
  • atmospheric transmissometers, scatterometers, etc. can be used.
  • These two types of equipment in low-visibility weather such as rain and fog will be caused by complicated conditions such as water vapor absorption. Large errors, difficult to observe normally, and it is difficult to accurately detect the fog that occurs in a local area.
  • the cost is high, and the maintenance cost is high. It is too costly to construct a separate meteorological inspection station in the entire highway. Wide. Taking the Shanghai-Nanjing Expressway as an example, the cost of building a meteorological inspection system with more than 10 observation points is nearly 10 million.
  • the video visibility detection method combines image analysis and artificial intelligence technology with traditional atmospheric optical analysis. Through the analysis and processing of video images, the relationship between the image and the real scene is established, and the meteorological visibility is calculated by measuring the changes of the image features. Value. Compared with the traditional method, the detection principle is close to the human eye viewing mode, which has the advantages of low cost, easy operation, and wide coverage of compatible cameras along the road, but is not mature enough and needs improvement.
  • the University of Minnesota proposed a video visibility detection method based on fixed-distance target objects [1] . It requires a preset number of video detection targets, which is costly, cumbersome to operate, and susceptible to terrain environment and other factors.
  • MIT proposes a logo-based image.
  • the method for calculating the relative visibility [2] compares and analyzes the detected scene image with the pre-stored known meteorological visibility image to obtain relative visibility. The method does not require an auxiliary facility and is convenient to use, but is difficult to use for the PTZ camera and is also susceptible to the scene.
  • a visibility detection algorithm based on road condition markers is proposed.
  • the matching algorithm is used to segment the preset target from the image, and then the features are fitted to obtain the visibility value.
  • This method requires additional markings, which is costly.
  • the detection range and accuracy are limited by the distance and number of objects that can be selected in the field of view. It is difficult to be compatible with the PTZ cameras that have been deployed.
  • the literature [5] proposes the detection based on video image contrast. The method is to analyze the contrast between each pixel and its four neighbors. If the selected maximum value is greater than a given threshold, it is a human eye distinguishable pixel, and then combined with the camera calibration to convert the visibility value. Due to the threshold division, it is susceptible to noise interference, including lane segmentation. Line area noise, CCD imaging current noise, especially quantization error noise can cause the detection result to jump, and the algorithm is not stable enough.
  • the problems to be solved by the present invention are as follows:
  • the prior art has insufficient monitoring of visibility, such as large detection limit, unable to meet the requirements of large-area road condition monitoring, insufficient real-time monitoring, high monitoring cost, and susceptible to monitoring accuracy;
  • the technical scheme of the present invention is: a PTZ video visibility detection method based on image brightness characteristics, using a PTZ video camera to acquire a road condition video image, extracting a region of interest ROI of the road surface, obtaining a high degree of consistency of selected pixels; using a region growth based on Nagao filtering
  • the algorithm obtains the precise road surface, removes the influence of the roadbed and the vehicle, and ensures the illumination of the selected pixels in the world coordinates.
  • extracts the contrast curve reflecting the change of the road surface brightness finds the brightness curve feature points, and calculates the image through the extinction coefficient.
  • the human eye can distinguish the farthest pixel; the maximum visible distance is converted by the camera calibration, and the visibility value is determined, including the following steps:
  • the area between the lane dividing lines is the ROI of the current image, and the subsequent processing is limited. Ensure that the image pixels are highly consistent within the ROI;
  • the road surface mask region is accurately extracted, and all subsequent processing will be limited to the mask region to reduce the calculation amount and ensure the selected pixel in the world coordinates.
  • the illuminance in the same is the same: Calculate the gray value of the bottom row of the ROI, counted as median ⁇ ), select the pixel with the median(P g ) as the seed point, and follow the principle of bottom-up and left-to-right.
  • the mask area is scanned, and the target pixel of the scan is sequentially determined according to the following growth criteria to determine whether it belongs to the road surface area: 41) Brightness balance
  • Equation (6) it is a constant less than 1, representing ⁇ , the number of rows separated from the initial seed point, G m .
  • x refers to the brightness difference threshold between a pixel and its upper 3-neighbor, the upper 3-neighbor, ie, the upper left, upper, and upper right three pixel points of the pixel, the upper left luminance difference threshold ⁇ , the upper luminance difference threshold G x , the upper right luminance difference threshold ax , where -
  • the adaptive window width Nagao median filter is used to filter out the noise interference in the image, and the energy of the non-diffused noise point is not satisfied.
  • the pixel points satisfying the brightness balance are filtered by the Nagao median of the adaptive window width to obtain the pixel gray scale.
  • Luminance feature extraction Using the road surface region with the same initial illumination and high uniformity obtained by the foregoing steps, analyzing the brightness change trend of the road surface pixel due to atmospheric extinction, and finding out the change feature point, that is, the second derivative zero point of the brightness curve;
  • Visibility calculation Determine the atmospheric extinction coefficient by using the vanishing point coordinate and camera parameters obtained by the camera calibration algorithm and the second derivative zero coordinate obtained by the brightness curve, and then using the Cosimed's theorem to derive the atmospheric extinction coefficient and The relationship between visibility is then solved to determine the visibility value.
  • the camera in step 2) is calibrated as:
  • the PTZ video camera imaging mapping model includes three coordinate systems: the road world coordinate system ⁇ , ; ⁇ ⁇ , the camera coordinate system ( c , ; r c , z c ) and the image plane coordinate system ⁇ , Z c and
  • the angle between the road surface is ⁇
  • the distance from the road surface is H
  • / is the effective focal length of the lens
  • the relationship between the road coordinate system and the camera coordinate system, the camera coordinate system and the imaging plane is established as follows:
  • step 6 The relationship between atmospheric visibility and extinction coefficient in step 6) is - according to Cosimed's theorem, set the apparent extinction coefficient, the apparent brightness or radiance L of an object of a fixed brightness at a distance from the human eye, and the object itself.
  • the relationship between brightness and background brightness is as follows:
  • Equation (9) shows that the apparent brightness of an object consists of two parts: the inherent brightness of the object is weakened, and the background brightness is gradually strengthened in the form, the contrast changes.
  • the relationship between the atmospheric extinction coefficient and the distance ⁇ is: / (10)
  • the target receives the brightness contrast
  • c e is the intrinsic brightness contrast, when the scattering coefficient is independent of the azimuth and along the observer, target
  • the relation (10) is established.
  • ⁇ ei be the farthest distance that can be observed by the human eye, that is, the pixel with a contrast of 0.05, which has:
  • Relationship (11) shows the relationship between atmospheric visibility and extinction coefficient
  • the image brightness L is determined by the second derivative of the vertical coordinate V of the image plane, and the distance ⁇ is substituted by the relation (5).
  • 1 ⁇ 4 is the second-order inflection point position of the brightness curve, that is, the second-order derivative zero point, and 1 ⁇ 4 is the horizon, that is, the vanishing point position.
  • the atmospheric visibility distance is:
  • V j U (14) k 2(v,. - v A )
  • Len(Pix(j)) min(Lengh, lengh(j)) ( 15) represents the set of the first row of points that make up the measurement band, fe ⁇ 3 ⁇ 4c ⁇ is its length, which is a set constant threshold; Obtain the median brightness of each line in the measurement band, obtain the brightness-distance curve and find the second derivative of S, determine the change feature point, that is, the second derivative zero point, and interpolate and filter the curve S before finding the second derivative zero point. Eliminate the confusion of the second-order zero to reduce the error and get the most accurate measurement possible.
  • the detection method of the invention has high stability, can better adapt to the image under various weather conditions, can accurately determine the degree of smog, and the detection is accurate within 10 meters;
  • the Nagao median filter with adaptive window width replaces the traditional mean filtering, which can ensure the image resolution and edge effect at the same time, and still has better robustness when the noise density changes widely.
  • the invention realizes the seamless detection of the full visibility of the road, including the local area fog, the whole process visibility distribution, etc.
  • the invention provides accurate data support, improves road utilization, and has important economic benefits.
  • Figure 1 is a flow chart of the method of the present invention.
  • FIG. 2 is a model diagram of a camera calibration model of the present invention.
  • FIG. 3 is a flow chart of a region growing method for ensuring brightness uniformity according to the present invention.
  • Figure 4 shows the region-growth algorithm using a 3-neighbor map on the seed store.
  • Figure 5 is an image grayscale map corresponding to the visibility calculation in step 6).
  • Figure 6 is a graphical representation of the actual brightness composition of the target measured by the camera of the present invention.
  • Figure 7 is a flow chart for solving the visibility based on the luminance feature points.
  • Figure 8 is a schematic view of a road surface measurement belt obtained after pretreatment of the present invention.
  • FIG. 9 is a first derivative and a filter diagram of a luminance variation curve obtained after calculation according to the present invention.
  • Figure 10 is a schematic diagram showing the actual detection results of the present invention.
  • FIG. 11 is a schematic diagram showing the actual detection result of the present invention compared with the contrast algorithm and the human eye observation result. detailed description
  • the present invention uses a digital camera to simulate the perceptual characteristics of the human eye, and by studying the trend of pixel contrast and brightness in the video image, the characteristic parameter variation of the image is converted into the perceived intensity of the human, thereby solving the visibility.
  • the present invention proposes a unified video processing method, which adopts the most extensive surveillance camera with the most comprehensive coverage on the existing highway for processing. Build low-cost, wide coverage, intuitive information detection, detect current road visibility, achieve low false positive rate, low miss detection rate, and high-precision detection system to achieve high-density, low-cost, easy-to-maintain real-time traffic information collection.
  • the present invention uses a PTZ video camera to acquire a road condition video image, extracts a region of interest R0I of the road surface, and obtains a high degree of consistency of the selected pixels.
  • the region growth algorithm based on the Nagao filter is used to obtain an accurate road surface region, and the influence of the roadbed and the vehicle is removed. , to ensure that the selected pixels have the same illumination in the world coordinates; in the road surface region, extract the contrast curve reflecting the change of the road surface brightness, find the brightness curve feature points, calculate the farthest pixel in the image by the human eye through the extinction coefficient; The maximum visible distance is converted to determine the visibility value.
  • PTZ video cameras ie Pan, Tilt, Zoom, cameras that can change the angle of view and zoom horizontally and vertically, for monitoring systems.
  • the invention comprises the following steps:
  • the area between the lane dividing lines is the ROI of the current image, and the subsequent processing is limited. Ensure that the image pixels are highly consistent within the ROI;
  • the pixel scans the mask area line by line according to the principle of bottom-up and left-to-right. For the target pixel of the scan, according to the following growth criteria, it is sequentially determined whether it belongs to the road surface area: the brightness continuity and consistency are satisfied. Pixels can be added to the pavement area until the mask area is scanned, and the precise road surface area is available;
  • Luminance feature extraction Using the road surface region with the same initial illumination and high uniformity obtained by the foregoing steps, analyzing the brightness change trend of the road surface pixel due to atmospheric extinction, and finding out the change feature point, that is, the second derivative zero point of the brightness curve;
  • the brightness of the image gradually changes with the distance, but since the brightness value is a discrete integer value between 0 and 255, the same brightness value between adjacent lines often occurs, or due to noise points. The influence results in a lot of confusing second-order derivative zeros. Therefore, in order to avoid false detection, the luminance curve is interpolated and filtered before the mutation point is found, and the second-order zero point is eliminated.
  • step 2) the calculation method of the imaging mapping calibration module of the PTZ camera is as follows:
  • the imaging model of the road condition camera is described in detail in [8], including three coordinate systems: the road coordinate system ( , ; ⁇ , ), the camera coordinate system ( c , ; r c , z c ) and the image plane coordinate system ( M , . Z c and the road surface angle O is the road surface height, where / is the effective focal length of the lens.
  • the distance of the road surface represented by the pixels in the image from the optical center of the camera can be expressed as:
  • step 3) and step 4 video image consistency is ensured from two aspects: high consistency and illumination consistency.
  • R0I detection guarantees high consistency.
  • the present invention uses the Kluge model to fit the projection of the lane dividing line in the video image, and solves the projection of the lane dividing line in the image by solving the unknown parameters in the model by randomized Hough transform (RTT).
  • RTT randomized Hough transform
  • the area between the lines is the ROI of the current image.
  • the subsequent processing is limited to the ROI to ensure the contour of the image pixels, and the complexity of subsequent operations is greatly reduced.
  • the bottom line of the image ROI must For the area where the road surface is located, for this purpose, calculate the median gray value of the bottom row of the ROI, calculated as medianCP g ), and select the pixel with the brightness of medianCP g ) as the seed point, according to the principle of bottom-up and left-to-right. Line scan mask area.
  • the flow chart of the area growth is as shown in FIG. 3. For judging the target pixel point, it is determined in turn whether or not it belongs to the road surface area according to the following growth criterion.
  • Equation (6) it is a constant less than 1, representing ⁇ , the number of rows separated from the initial seed point, G m .
  • x refers to the brightness difference threshold between a pixel and its upper 3-neighbor, and the upper 3-neighbor is the upper left, upper, and upper right of the pixel.
  • the three pixel points, the upper left luminance difference thresholdRIC1, the upper luminance difference threshold G x , and the upper right luminance difference threshold ax as shown in FIG. 4, generally include:
  • G m—a 1 x G m 1 ax ⁇ G m 0 ax (7)
  • Brightness balance ensures that the brightness of the pixel is prevented from drifting. It is assumed that the gray value range of the image is 0 ⁇ 255, and the threshold between adjacent lines is 8, If only adjacent lines are restricted without this restriction, it is possible that after 32 lines, the black point (luminance value is 0) and the white point (luminance value is 255) are simultaneously present in the road surface area, and the gray value is relative to the seed point gray. A large drift has occurred.
  • the Nagao median filter here is described in detail in [9][10] and will not be described in detail.
  • the pixel points satisfying the brightness balance are filtered by the Nagao median of the adaptive window width to obtain the pixel gray value.
  • the idea of the Nagao algorithm is: First, rotate a long strip template around the center pixel, select the template position with the smallest variance, replace the gray value of the center pixel with its gray mean value, and iterate to change the number of pixels to 0.
  • the adaptive window width Nagao filter uses a large-scale template for the homogenous region, which makes the angular resolution more fine; For the edge and texture regions, small-scale templates are used to avoid blurring edges and textures. Since the traditional Nagao filter uses mean filtering, its effect is not ideal.
  • the template uses median filtering instead of mean filtering, which is more resistant to noise. Even if there are multiple noise points in the template, it can be clearly filtered.
  • the adaptive window width Nagao median filter effectively removes road noise points caused by subgrades, green belts, shadows, etc., while preserving the edge and texture characteristics of the mask area. Pixels that meet the continuity and consistency of the brightness can be added to the pavement area until the mask area is scanned and the precise road area is available.
  • step 5 the brightness feature extraction steps are as follows:
  • the midpoint coordinates of each line are used as the center of the road surface measurement belt, and under the condition of equation (15), a road surface measurement belt is formed.
  • Len(Pix(j)) min(Lengh, lengh(j)) ( 15) represents the set of j-th row points that make up the measurement band, fe ⁇ 3 ⁇ 4c ⁇ is its length, Lengh is a constant threshold, generally According to the image resolution, take 5% ⁇ 10% of the horizontal resolution of the image, in this system, due to image resolution Is 704 * 576, the horizontal resolution is 704, the threshold is taken as 50, the measuring tape and the midpoint thereof are shown in Figure 8; the brightness of each row in the measurement band is obtained, as follows -
  • the horizontal axis represents the coordinate of the image coordinate system along the vertical direction, that is, the distance between the target point and the camera in the image coordinate system
  • the blue line represents the first derivative value of the brightness of each line
  • the red line represents the first-order derivative filtering effect picture.
  • the maximum value of the first derivative is the second-order luminance feature point, from which the visibility is calculated.
  • step 6 the visibility is as follows:
  • Atmospheric visibility is an indicator of atmospheric transparency. It is generally defined as the maximum distance that can be seen and recognized by a black target placed at an appropriate scale near the ground, relative to the background of scattered light in the sky. This definition depends on human vision. There is a difference in the ability of humans and computers to perceive images. As shown in Figure 5, a 16-bit grayscale, the computer can accurately distinguish the difference between any two levels, and the human eye can only Differentiate the level of brightness difference.
  • the human eye can distinguish between the target and the pixel with a background contrast greater than 0.05. Only by determining the difference between a person's and a computer's ability to perceive an image can a computer have the ability to measure visibility.
  • Koschmieder proposes that there is attenuation in the sky background when the light passes through the atmosphere. It is assumed that the atmospheric extinction coefficient, the brightness or radiance of an object of fixed brightness at a distance of ⁇ from the human eye, and the brightness of the object itself and the background brightness The relationship is as follows:
  • Equation (9) shows that the apparent brightness of an object consists of two parts: the inherent brightness of the object is weakened and the background brightness is gradually strengthened in the form of ⁇ , as shown in Figure 6. According to this, the contrast change and the atmospheric extinction coefficient can be derived.
  • the relationship between the distance ⁇ is: L f (10)
  • the target receives the brightness contrast
  • c e is the intrinsic brightness contrast.
  • the relation (10) holds when the scattering coefficient is independent of the azimuth angle and the illuminance along the entire path between the observer, the target, and the horizon sky.
  • the image brightness is determined by the second derivative of the image plane vertical direction coordinate V, which is: dv (v ⁇ v h ) v - v h ( 12 )
  • V image plane vertical direction coordinate
  • the image pixel brightness and its derivative change with distance. A change has occurred.
  • 1 ⁇ 4 is the second-order inflection point position of the brightness curve, and 1 ⁇ 4 is the horizon or the vanishing point position.
  • the atmospheric visibility distance is:
  • the visibility detection test hardware platform is P4/2.8GHZ single CPU, 1G memory, SUSE Linux operating system, video capture from Jiangsu Ninglian Expressway video surveillance image, MPEG-2, resolution 704 X 576.
  • Figure 10 is a set of four images taken from the same scene with a gradual change in visibility. Each frame is intercepted every 30 minutes.
  • the "+" in the figure represents the critical visibility obtained by the contrast algorithm.
  • the horizontal line position is obtained by the method of the present invention. Visibility, which is the visibility value observed by the human eye.
  • the method of the present invention is compared with the human eye and the contrast based method.
  • the algorithm is consistent with the human eye observation effect, the accuracy rate is over 96%, and the detection error is within 10 meters.
  • the operation is simple, anti-interference and high accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Pathology (AREA)
  • Immunology (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Chemical & Material Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)

Abstract

基于图像亮度特征的PTZ视频能见度检测方法,采用PTZ视频摄像机获取路况视频图像,提取路面的兴趣域ROI,获得所选像素的高度一致性;利用基于Nagao滤波的区域增长算法得到精确路面区域,确保所选像素在世界坐标中的照度一致;在路面区域中,提取反映路面亮度变化的对比度曲线,找寻亮度曲线特征点,通过消光系数计算图像中人眼可分辨最远像素;结合摄像机标定换算出最大能见距离,确定能见度值。本发明无需设置任何人工标志物,充分利用已有的PTZ摄像机进行路况摄像,获取图像,能够实时监测,且监测成本低;满足大面积路况监测的要求,监测稳定不受外界环境干扰,是一种简便易实现,精度高效果良好的能见度监测方法。

Description

说明书 基于图像亮度特征的 PTZ视频能见度检测方法
技术领域
本发明属于图像处理技术领域, 涉及摄像机定标技术、 自适应 Nagao滤波技术、 噪 声抑制技术、 能见度求解技术等, 在保证所选像素的高度、 照度一致性的基础上, 提取 反映路面亮度变化的对比度曲线, 找寻亮度曲线特征点, 通过消光系数计算图像中人眼 可分辨最远像素; 结合摄像机标定换算出最大能见距离, 确定能见度值, 具体为一种基 于图像亮度特征的 PTZ视频能见度检测方法。 背景技术
高速公路是涉及国计民生的新兴产业, 2010年我国总里程计划数将超过 8万公里。 江苏是我国经济发达地区, 是智能交通系统 ITS(Intelligent Transportation System)的领头 羊。 目前已有高速公路 3558公里, 高速公路密度为 3.46公里 /百平方公里, 2010年总里 程将达 5000公里。
同时我国境内山水平原错落, 河流湖泊纵横, 特别是中、 西部地区, 是地势复杂的 丘陵地带, 雾、 靈、 霾等自然灾害性天气多有发生, 给交通运输带来了极大的隐患, 特 别是在不定时间和地点突然生成的团雾, 对车辆安全的危害尤大。 1975 年,美国加利福 利亚至纽约的高速公路, 因大雾致使 300多辆车相撞, 死伤 1000多人, 造成了世界最 大的交通事故。 法国在 1986年, 有 1200 起事故 (不含市内)是由于雾引起的, 造成 182 人死, 175人受伤, 1352人轻伤, 虽然高速公路上因雾产生的事故率仅为该年度的 4%, 但死亡率却高达 7%〜8%。 在管理手段上, 沪宁高速公路构建的仅 10余观测点的气象 检测系统耗资近千万, 仍很难准确检测出发生在局部地区的团雾。
针对能见度降低, 我国公路管理部门动辄封路, 以降低交通事故的发生。 然而管理 部门实施交通管制的程序主观性较强, 实施管制的条件无量化指标、 不够科学、 不够规 范, 通过关闭来求安全可能适得其反。 为此, 对路况气象, 特别是低能见度的及时检测 发布, 是加强应对灾害天气能力, 减少损失, 提高高速公路管理水平的切实要求。
目前高速公路使用的气象能见度检测设备主要以传统激光能见度仪为主,一般可用 大气透射仪、 散射仪等, 这两类设备在雨、 雾等低能见度天气, 会因水汽吸收等复杂条 件造成较大误差, 难以正常观测, 也很难准确检测出发生在局部地区的团雾。 同时造价 昂贵, 维护成本高, 在高速公路全程密集构建单独的气象检测站成本过高, 难以普及推 广。 以沪宁高速公路为例, 构建 10余观测点的气象检测系统的耗资近千万。
视频能见度检测方法将图像分析和人工智能技术与传统的大气光学分析结合起来, 通过对视频图像的分析处理, 建立图像与真实场景之间的关系, 通过测量图像特征的变 化情况, 计算出气象能见度的值。 与传统手段相比, 在检测原理上接近人眼观看方式, 具有低成本、 易操作, 及兼容沿路摄像机的广覆盖的优势, 但不够成熟, 尚需改进。
目前, 国外对此研究较少, 仍处于理论与实验阶段。 美国明尼苏达州大学提出基于 固定距离目标物的视频能见度检测方法 [1], 需人为预置多个视频检测目标, 成本较高, 操作繁琐, 容易受到地形环境等因素的制约; MIT提出基于标志图像计算相对能见度的 方法 [2], 将检测场景图像与预存的已知气象能见度图像比较分析, 获得相对能见度, 该 方法无需辅助设施, 使用方便, 但难以用于云台摄像机, 也易受到场景中运动物体遮挡 的影响; 瑞典国家道路管理中心提出了基于神经网络和红外视频的能见度检测方法 [3], 提取不同能见度图像边缘, 采用神经网络算法分类, 并转换为能见度相应等级, 红外摄 像机摄像噪声相对低, 但其价格昂贵, 维护复杂, 难以沿路密集布设。
文献 [4]提出一种基于路况标记物的能见度检测算法采用匹配算法从图像上分割出 预置的目标物, 进而对其特征进行拟合, 得出能见度值。 该方法需要设置额外标记, 成 本高, 其探测范围和精度受限于视野范围内可选用的目标物的距离和数量, 难以兼容已 布设的 PTZ摄像机; 文献 [5]提出基于视频图像对比度的检测方法, 解析各像素与其四 邻的对比度, 所选取的最大值若大于给定阈值, 即为人眼可分辨像素, 再结合摄像机标 定来换算出能见度值, 由于阈值划分, 易受噪声干扰, 包括车道分割线区域噪声、 CCD 成像电流噪声, 特别是量化误差噪声会引起检测结果跳变, 算法不够稳定。
参考文献
[1] T. M. Kwon. Atmospheric visibility measurements using video cameras: Relative visibility.
Technical report[R], America: University of Minnesota Duluth, July 2004.
[2] Robert G. Hallowell, Michael P. Matthews, and Paul A. Pisano, Automated Extraction of
Weather Variables from Camera Imagery [C]. Proceedings of the 2005 Mid-Continent
Transportation Research Symposium, Ames, IA, 2005.8
[3] Final Report on Signal and Image Processing for Road Condition Classfication[R],
AerotechTelub and Dalarma University under the7 Swedish National Road Agency.
Report #2002-02-06. 2002.6.
[4] 陈钊正, 周庆逵, 陈启美. 基于小波变换的视频能见度检测算法研究与实现 [J] . 仪 器仪表学报 2010, 31 (1) : 92 98
[5] 李 勃, 董 蓉, 陈启美.无需人工标记的视频对比度道路能见度检测 [J] . 《计算机 辅助设计与图形学学报》, 2009, 11 (21): 1575-1982
[6] Li Bo, Chen Qimei. Vehicle activity analysis from freeway traffic video [J]. Chinese Journal of Scientific Instrument, 2006,27(Suppl.3): 2118-2120 (in Chinese)
(李勃, 陈启美.基于监控视频的运动车辆行为分析算法 [J] .仪器仪表学报, 2006, 27 (增刊 3) : 2118-2120)
[7] 于莉娜,胡正平,练秋生. 基于改进随机 Hough变换的混合圆 /椭圆快速检测方法 [J] . 电子测量与仪器学报, 2004, 18 (2) : 92〜97
[8] 蒋耳关源 . Fast detection of multi-circle with randomized Hough transform[J].
Optoelectronics Letters VOL5, 2009
[9] 张煜东, 吴乐南. 基于 SPC N和 Nagao滤波的图像去噪. 中国科学 F辑: 信息科学,
2009, 39(6): 598-607
[10] Nagao M,Takashi M. Edge preserving smoothing. Comput Graph Image Process, 1979, (9): 394-407 发明内容
本发明要解决的问题是: 现有技术对能见度的监测存在不足, 如检测局限性较大、 不能满足大面积路况监测的要求, 监测实时性不够, 监测成本高昂, 监测精度易受影响 等; 需要一种简便易实现, 精度高效果良好的能见度监测方法。
本发明的技术方案为: 基于图像亮度特征的 PTZ视频能见度检测方法, 采用 PTZ 视频摄像机获取路况视频图像, 提取路面的兴趣域 ROI, 获得所选像素的高度一致性; 利用基于 Nagao滤波的区域增长算法得到精确路面区域, 去除路基、 车辆的影响, 确保 所选像素在世界坐标中的照度一致; 在路面区域中, 提取反映路面亮度变化的对比度曲 线, 找寻亮度曲线特征点, 通过消光系数计算图像中人眼可分辨最远像素; 结合摄像机 标定换算出最大能见距离, 确定能见度值, 包括以下步骤:
1)通过设置的 PTZ视频摄像机, 实时获取路况视频图像信息;
2)从视频摄像机获取的图像信息, 通过摄像机定标技术完成视频图像与路面世界坐 标的转换关系, 计算视频图像中路面区域与摄像机的距离;
3)采用 Kluge模型拟合车道分割线在视频图像中的投影, 通过随机霍夫变化求解模 型中的未知参数, 车道分割线间的区域即为当前图像的兴趣域 ROI, 将后续的处理过程 限制于 ROI内, 确保图像像素的高度一致性;
4)采用区域增长算法, 结合 ROI区域内亮度判断准则以及自适应 Nagao滤波方法, 精确提取路面 mask区域, 所有的后续处理将限制在 mask区域中, 以减少计算量, 保证 所选像素在世界坐标中的照度一致: 计算 ROI 中最下面一行的灰度中值, 计为 median^), 选取亮度为 median(Pg)的像素作为种子点, 依照自下而上、 从左至右原则 逐行扫描 mask区域, 对扫描的目标像素点 根据如下增长准则, 依次判断是否属 于路面区域: 41)亮度均衡性
P(i, :)与 median(Pg)之间满足
P(i,j) -median(Pg) < pnr minG^
(6)
(yt— 1,0,1)
式 (6)中, 是小于 1的常数, 代表 Υ, 和初始种子点 相隔行数, Gmx则指 某像素点与其上 3-邻域内亮度差阈值, 上 3-邻域即像素点的左上方、上方、 以及右上方 的三个像素点,左上方亮度差阈值 ^,上方亮度差阈值 G x,右上方亮度差阈值 ax, 其中-
(? ma ιx― (? mιax < (7 m°ax (7)
42)照度一致性
采用自适应窗宽的 Nagao中值滤波器, 滤除图像中的噪声干扰, 同时不扩散噪声点 的能量, 满足亮度均衡性的像素点经过自适应窗宽的 Nagao中值滤波, 得到像素灰度值 Q(i, 满足:
3m e { - 1, i, i + 1}
Q(i, j) - Q(m, j + \) < Gm' :x (8) 满足亮度连续性以及一致性的像素点即可加入到路面区域中, 直至 mask区域扫描 完毕, 精确的路面区域即可得到;
5)亮度特征提取: 利用前述步骤获得的初始照度一致、 高度一致的路面区域, 分析 路面像素因大气消光导致的亮度变化趋势, 找出其中的变化特征点, 即亮度曲线的二阶 导数零点;
6)能见度计算: 通过由摄像机标定算法求得的灭点坐标以及摄像机参数, 以及亮度 曲线求得的二阶导数零点坐标, 确定大气消光系数, 再利用科西米德定理推导出大气消 光系数与能见度之间的关系, 进而求解确定能见度值。
步骤 2)中的摄像机定标为:
PTZ视频摄像机成像映射模型包括 3个坐标系: 路面世界坐标系^,;^ ^^, 摄像 机坐标系 ( c,;rc,zc)以及视频图像的像平面坐标系^, , Zc与路面夹角为 Θ , O距离路面 高度为 H, /为镜头有效焦距, 建立路面坐标系与摄像机坐标系间、 摄像机坐标系与成 像平面间的变换关系为:
Figure imgf000007_0001
由 yw=+∞可得, 地平线 VA, 也就是灭点在像平面中的投影为:
=-/tan^
Figure imgf000007_0002
代入式 (2)中可以得到:
Figure imgf000007_0003
据此, 图像中像素所
Figure imgf000007_0004
步骤 6)中大气能见度与消光系数的关系为- 根据科西米德定理, 设 表示大气消光系数, 某一固定亮度的物体在距离人眼距离 为 ί处的视亮度或辐射率 L与物体本身亮度 及背景亮度 关系如下式:
L = L0e M+Lf{\-e M) (9) 式 (9)表明物体的视亮度由两部分组成: 物体固有亮度 以 消弱, 以及背景亮度 以 形式逐渐加强, 则对比度变化与大气消光系数 , 距离 ί之间的关系为: / (10) 式 (10)中, 是目标物接收亮度对比度, ce是固有亮度对比度, 当散射系数与方 位角无关且沿观测者、 目标物和地平天空之间的整个路径上的照度均勾时, 关系式 (10) 成立,
设^ ei为人眼所能观测到的最远距离, 即对比度为 0.05处的像素点, 有:
Figure imgf000007_0005
关系式 (11)表明了大气能见度与消光系数之间的关系; 则基于亮度特征点的能见度求解为:
将图像亮度 L对像平面竖直方向坐标 V求二阶导数, 并将距离 ί用关系式 (5)代入,
^ 。 — (12) dv (v - vh ) v - vh
在消光系数 的作用下, 图像像素亮度 及其导数随距离的变化而发生变化, 随着 雾的增大, 在天空背景下, 目标物变得更加模糊, 并且求得的极值点也在减小, 令二阶 导数为 0, 舍弃无意义的 ^=0的解, 得到:
k = ^^ (13) λ
¼即亮度曲线二阶拐点位置, 也就是二阶导数零点, ¼为地平线即灭点位置, 由此, 大气能见距离为:
V j = U (14) k 2(v,. - vA )
当 趋近于 ¼的时候, ^处于临界状态, 认为此时刚好看见雾的出现; 当 大于 ¼的时候, 雾是可以检测得到的; 相反, 小于 ¼的时候, 认为没有雾的出现。
步骤 5)具体为, 在步骤 4)得到的路面区域中, 依次找出第 行最大长度连续像素点 集 Pix(i, j),起始于 (, , 终止于 其长度 /e"g/?(/;j=(¾-a+ _, 中点 middle b+a)/2,j), 各行的中点坐标 middle作为路面测量带的中心, 在式(15)的条件下, 组 成一条路面测量带:
len(Pix(j)) = min(Lengh, lengh(j)) ( 15) 式中 代表组成测量带的第 '行点集, fe< ¾c^^即为其长度, 为一设置的 常数阈值; 获取测量带内每行中值亮度, 得到亮度-距离变化曲线 对 S求二阶导数, 确定变化特征点 即二阶导数零点,并且在寻找二阶导数零点前,对曲线 S进行插值、 滤波, 消除混淆二阶零点, 以减小误差, 得到尽可能精确测量结果。
本发明具有以下优点:
1. 无需设置任何人工标志物, 充分利用已有的 PTZ摄像机进行路况摄像, 获取图像, 能够实时监测, 且监测成本低;
2. 本发明检测方法稳定性高, 能较好的适应各天气状况下的图像, 能准确判断出雾霾 程度, 检测精确在 10米内;
3. 采用自适应窗宽的 Nagao中值滤波器取代传统的均值滤波, 能够同时保证图像分辨 率以及边缘的效果, 且在噪声密度大范围变化时仍然具有较好的鲁棒性;
4. 本发明实现了道路全程能见度无缝检测, 包括局部地区团雾、 全程能见度分布等, 为在雾霾天气有针对性的给出路段车辆限速, 路由迂回等信息, 而非盲目封路提供 了准确的数据支撑, 提高了道路利用率, 具有重要的经济效益。 附图说明
图 1为本发明方法的流程图。
图 2为本发明摄像机标定模模型图。
图 3为本发明为确保亮度一致性采用的区域增长方法流程图。
图 4为区域增长算法采用种子店上 3-邻域示意图。
图 5为图像灰度等级图,对应步骤 6)中的能见度计算。
图 6为本发明摄像机所测得目标物实际亮度组成图示。
图 7为基于亮度特征点的能见度求解流程图。
图 8为本发明预处理后得到的路面测量带示意图。
图 9为本发明计算后得到的亮度变化曲线一阶导数及滤波图。
图 10为本发明实际检测结果示意图。
图 11为本发明实际检测结果与对比度算法及人眼观测结果相比较示意。 具体实施方式
本发明使用数字摄像机模拟人眼的感知特性, 通过研究视频图像中像素对比度、 亮 度变化趋势, 将图像的特征参量变化转换为人的感知强度, 进而求解能见度。 相对于传 统的人力观看视频, 采用纷繁复杂的外场设备进行交通参数采集, 本发明提出了统一的 视频处理方法, 采用现有高速公路上覆盖面最广、 信息反映最直观的监控摄像机进行处 理, 研究构建低成本、 覆盖面广、 信息反映直观的检测手段, 检测当前路况能见度, 实 现低误报率、 低漏检率、 高精度的检测系统, 以实现高密度、 低成本、 易维护的实时路 况信息采集。
如图 1, 本发明采用 PTZ视频摄像机获取路况视频图像, 提取路面的兴趣域 R0I, 获得所选像素的高度一致性; 利用基于 Nagao滤波的区域增长算法得到精确路面区域, 去除路基、 车辆的影响, 确保所选像素在世界坐标中的照度一致; 在路面区域中, 提取 反映路面亮度变化的对比度曲线, 找寻亮度曲线特征点, 通过消光系数计算图像中人眼 可分辨最远像素; 结合摄像机标定换算出最大能见距离, 确定能见度值。 PTZ视频摄像 机, 即 Pan (平移), Tilt (倾斜), Zoom (变焦), 可以水平、 垂直改变视角及变焦的摄像 机, 用于监控系统。
本发明包括以下步骤:
1)通过设置的 PTZ视频摄像机, 实时获取路况视频图像信息;
2)从视频摄像机获取的图像信息, 通过摄像机定标技术完成视频图像与路面世界坐 标的转换关系, 计算视频图像中路面区域与摄像机的距离;
3)采用 Kluge模型拟合车道分割线在视频图像中的投影, 通过随机霍夫变化求解模 型中的未知参数, 车道分割线间的区域即为当前图像的兴趣域 ROI, 将后续的处理过程 限制于 ROI内, 确保图像像素的高度一致性;
4)大气光散射的作用, 使图像中路面像素亮度随着距离呈现一定的变化关系。 路面 上路基、 车道分割线等亮度的跳变, 可能导致路面亮度提取过程中误差较大。 采用区域 增长算法, 结合 ROI区域内亮度判断准则以及自适应 Nagao滤波方法, 精确提取路面 mask区域, 所有的后续处理将限制在 mask区域中, 以减少计算量, 保证所选像素在世 界坐标中的照度一致: 根据摄像机投影成像原理, 图像 ROI中最下面几行必为路面所在 区域,为此,计算 ROI中最下面一行的灰度中值,计为 median( ¾,选取亮度为 median( ¾ 的像素作为种子点, 依照自下而上、 从左至右原则逐行扫描 mask区域, 对扫描的目标 像素点 Υ, 根据如下增长准则, 依次判断是否属于路面区域: 满足亮度连续性以及一 致性的像素点即可加入到路面区域中, 直至 mask区域扫描完毕, 精确的路面区域即可 得到;
5)亮度特征提取: 利用前述步骤获得的初始照度一致、 高度一致的路面区域, 分析 路面像素因大气消光导致的亮度变化趋势, 找出其中的变化特征点, 即亮度曲线的二阶 导数零点;
6)能见度计算: 如图 7, 通过由摄像机标定算法求得的灭点坐标以及摄像机参数, 以及亮度曲线求得的二阶导数零点坐标, 确定大气消光系数, 再利用科西米德定理推导 出大气消光系数与能见度之间的关系, 进而求解确定能见度值。
进一步的, 本发明中, 图像亮度随着距离逐渐变化, 但是由于亮度值是 0~255之间 的离散的整数值, 经常出现相邻行之间亮度值相同的现象, 或是由于噪声点的影响, 导 致存在很多混淆的二阶导数零点, 因此, 为了避免误检, 在寻找突变点前, 对亮度曲线 进行插值、 滤波, 消除混淆二阶零点。
步骤 2)中, PTZ摄像机的成像映射标定模块的计算方法如下:
由路况摄像机成像映射模型, 如图 2所示, 参考文献 [8]中有详细说明, 包括 3个坐 标系: 路面坐标系( ,;^, ), 摄像机坐标系 ( c,;rc,zc)以及像平面坐标系 (M, 。 Zc与路 面夹角为 O距离路面高度为 , 其中/为镜头有效焦距。 建立路面坐标系与摄像机 坐标系间、 摄像机坐标系与成像平面间的变换关系为:
,
Figure imgf000010_0001
v = ~f 由 ; Tw=+∞可得, 地平线 VA, 也即灭点在像平面中的投影为:
/ tan 6»
(3) 代入 (2)中可以得到:
^ +
f (4)
Yc = L - d si O Zc二 d cos Θ
据此, 图像中像素所表征的路面区域距摄像机光心的距离 可表示为:
Figure imgf000011_0001
步骤 3)和步骤 4)中, 分别从高度一致性和照度一致性两个方面来保证视频图像一 致性。
3) R0I检测保证高度一致性。在路面成像过程中,不可避免的会丢失物体高度信息, 如路侧树木成像可能在地平线之上。这将导致提取到的图像特征点难以通过摄像机标定 算法转换为具体的能见度值。 为此, 本发明采用 Kluge模型拟合车道分割线在视频图像 中的投影, 通过随机霍夫变化 (randomized Hough transform, RHT)求解模型中的未知参 数,检测车道分割线在图像中的投影,分割线间的区域即为当前图像的 ROI,关于 Kluge 模型及其参数的求解具体可参见参考文件 [6]的记载。 将后续的处理过程限制于 ROI内, 以确保图像像素的等高性, 同时也大大降低后续运算的复杂度。
4)设定 mask区域的最底端为种子区域, 保证每个种子点灰度值与此行所有像素中 值灰度值相差不大,根据摄像机投影成像原理, 图像 ROI中最下面几行必为路面所在区 域,为此,计算 ROI中最下面一行的灰度中值,计为 medianCPg),选取亮度为 medianCPg) 的像素作为种子点, 依照自下而上、 从左至右原则逐行扫描 mask区域。 区域增长流程 图如图 3所示, 对判断目标像素点 Υ, 根据如下增长准则, 依次判断是否属于路面区 域。
41)亮度均衡性
P(i, 与 median(Pg)之间满足
P(i,j) -median(Pg) < pnr minG^
(6)
(yt— 1, 0,1)
式 (6)中, 是小于 1的常数, 代表 Υ, 和初始种子点 相隔行数, Gmx则指 某像素点与其上 3-邻域内亮度差阈值, 上 3-邻域即像素点的左上方、上方、 以及右上方 的三个像素点,左上方亮度差阈值 „1,上方亮度差阈值 G x,右上方亮度差阈值 ax, 如图 4所示, 一般有:
G m—a 1x = G m1 ax < G m0ax (7) 亮度均衡性保证防止像素点亮度发生漂移, 假设图像灰度值范围为 0~255, 相邻两 行间阈值为 8, 如果只限制相邻行而不加此限制, 有可能在 32行后黑点 (亮度值为 0)和白 点 (亮度值为 255)同时存在在路面区域内, 灰度值相对于种子点灰度发生了较大的漂移。
42)基于自适应窗宽的 Nagao中值滤波器的照度一致性
这里的 Nagao中值滤波器在文件 [9][10]中已有详细介绍, 不再详述。 满足亮度均衡 性的像素点经过自适应窗宽的 Nagao中值滤波, 得到像素灰度值 满足:
Q(i, j) - Q(m, j + \) < Gm' x (8) 这一条件在去除图像噪声的基础上有效防止了区域内灰度值发生跳变。
Nagao算法的思想是: 首先围绕中心像素将一个长条形模板旋转一周, 选择方差最 小的模板位置, 以其灰度均值来代替中心像素的灰度值, 迭代至变化的像素数为 0。
考虑到角分辨率、边缘的保留以及计算准确率,选择了自适应窗宽的 Nagao滤波器: 自适应窗宽 Nagao滤波器对于同质区, 采用大尺度模板, 使得角分辨率更为精细; 而对于边缘区和纹理区, 采用小尺度模板, 避免了模糊边缘与纹理。 由于传统 Nagao滤 波器采用均值滤波, 本身效果并不理想。 而对模板采用中值滤波替代均值滤波, 抵御噪 声的能力更强, 即使模板内存在多个噪声点, 也能清晰滤波。 采用自适应窗宽 Nagao中 值滤波器可以有效的滤除由于路基、 绿化带、 阴影等所导致的路面噪声点, 同时保留 mask 区域的边缘与纹理特性。 满足亮度连续性以及一致性的像素点即可加入到路面区 域中, 直至 mask区域扫描完毕, 精确的路面区域即可得到。
步骤 5)中, 亮度特征提取步骤如下所示:
在步骤 4)得到的路面区域中, 由于车辆、 绿化带的影响, 每行像素点存在间断, 直 接取其中值亮度可能受到干扰物的影响。 从中依次找出第 '行最大长度连续像素点集 Pix(i, j), 起始于 (a,j), 终止于 (b, j) , 实长度 kngh(j)=(b-a+l)—, ¾ ·, 中点 middle为 ((b+a)/2,j) , 当 为奇数时, 按本领域常规进行取整操作, 也就是在位于中间的两个 像素点中任选一个作为中点。各行的中点坐标 middle作为路面测量带的中心,在式(15) 的条件下, 组成一条路面测量带。
len(Pix(j)) = min(Lengh, lengh(j)) ( 15) 式中 代表组成测量带的第 j行点集, fe< ¾c^^即为其长度, Lengh为一常数阈 值,一般依据图像分辨率, 取图像水平分辨率的 5%〜10%, 在本系统中, 由于图像分辨率 为 704*576,水平分辨率为 704, 该阈值取为 50, 测量带以及其中点示意如图 8所示; 获 取测量带内每行中值亮度, 如下式-
Bj = median{Pix{j)) (16) 得到亮度-距离变化曲线 s。 对 s求二阶导数, 确定突变点 结合摄像机标定计 算消光系数, 继而得到最大能见距离。 亮度 s 随着距离逐渐变化, 但是由于亮度值是
0~255之间的离散的整数值, 经常出现相邻行之间亮度值相同的现象, 或是由于噪声点 的影响, 导致存在很多混淆的二阶导数零点。 因此, 为了避免误检, 在寻找突变点前, 对 S插值、 滤波, 消除混淆二阶零点, 寻找 S—阶导数极大值点, 也即亮度函数二阶突 变点, 如图 9所示, 横轴代表图像坐标系沿竖直方向坐标, 也即图像坐标系中目标点距 摄像机距离, 蓝色线条表示对各行亮度求导的一阶导数值, 红色线条代表一阶导数滤波 效果图, 一阶导数最大值处即为二阶亮度特征点, 据此计算能见度。
步骤 6)中, 能见度的推导方式如下:
61)大气能见度与消光系数的关系
大气能见度属反映大气透明度的指标。 一般定义为相对于天空散射光背景下, 观测 安置于地面附近、 适当尺度的黑色目标物, 能看到且能辨认出的最大距离。 该定义依赖 于人的视觉, 人与计算机对图像的感知能力存在差异, 如图 5所示, 一个 16位的灰度 等级, 计算机能够准确分辨出任两等级之间的差异, 而人眼只能区分亮度差大的等级。
根据 CIE的定义, 目标物相对于背景对比度大于 0.05的像素点, 人眼才能够区分 出来。 只有确定人和计算机对图像感知能力之间的差异, 计算机才能具有测量能见度的 能力。
Koschmieder提出, 在天空背景下光线透过大气层时存在衰减, 假设 表示大气消 光系数,某一固定亮度的物体在距离人眼距离为 ί处的视亮度或辐射率 £与物体本身亮 度 ^及背景亮度 关系如下式:
τ τ ―] id τ —kd \
= L0e + Lf (l - e ) (9) 式 (9)表明物体的视亮度由两部分组成: 物体固有亮度 以 消弱以及背景亮度 以 ^形式逐渐加强, 如图 6所示, 据此可推导出对比度变化与大气消光系数 , 距离 ί之间的关系为: Lf (10) 式 (10)中, 是目标物接收亮度对比度, ce是固有亮度对比度。 当散射系数与方 位角无关且沿观测者、 目标物和地平天空之间的整个路径上的照度均勾时, 关系式 (10) 成立。
设^ ei为人眼所能观测到的最远距离, 即对比度为 0.05处的像素点, 有: ^ = -}ln(¾ = -}ln(0.05) » ^
k C0 k k (U) 关系式 (11)表明了大气能见度与消光系数之间的关系。
62)基于亮度特征点的能见度求解
将图像亮度对像平面竖直方向坐标 V求二阶导数, 有: dv (v ~ vh) v - vh (12) 在消光系数 的作用下, 图像像素亮度 及其导数随距离的变化而发生变化。 随着 雾的增大, 在天空背景下, 目标物变得更加模糊, 并且求得的极值点也在减小。 令二阶 导数为 0, 舍弃无意义的 ^=0的解, 得到:
h = 2(vi - vh)
λ (13)
¼即亮度曲线二阶拐点位置, ¼为地平线即灭点位置。 由此, 大气能见距离为:
3 3Λ
V , » .
k 2(vi - vh ) (14) 当 ^趋近于 A¾的时候, ^^处于临界状态, 认为此时刚好看见雾的出现; 当 ^大于 ¼的时候, 雾是可以检测得到的; 相反, 小于 ¼的时候, 认为没有雾的出现。
下面以路况能见度检测来具体说明本发明的实施。
能见度检测测试硬件平台为 P4/2.8GHZ单 CPU、 1G内存、 SUSE Linux操作系统, 视频采集来自江苏省宁连高速公路视频监控图像, MPEG-2, 分辨率为 704 X 576。
图 10分别取自同一场景的能见度逐步变化的一组 4幅图像, 每隔约 30min截取一 帧图像, 图中 "+ "代表由对比度算法得到的临界能见度, 横线位置即为本发明方法得 到的能见度, 其中 为人眼观测得到的能见度值。
图 11 所示场景能见度变化过程中, 本发明方法与人眼以及基于对比度方法的检测 结果比较。 该算法与人眼观测效果一致, 准确率达到 96%以上, 检测误差在 10米以内, 与能见度的对比度算法相比较, 操作简单、 抗干扰强、 准确率高。
由图 10~11可以看出, 随着能见度的增加, 检测到的最大能见距离与图像的顶端越 近; 在大多数情况下, 本发明方法与人眼观测值相一致, 也与基于对比度计算法结果符 合。 在误差允许范围内存在微小波动, 而在图像噪声较大或者团雾环境下, 基于对比度 算法存在跳变, 误差较大, 本发明方法可给出较好的能见度值。

Claims

权利要求书
1、基于图像亮度特征的 PTZ视频能见度检测方法, 其特征是采用 ΡΤΖ视频摄像机 获取路况视频图像, 提取路面的兴趣域 ROI, 获得所选像素的高度一致性; 利用基于 Nagao滤波的区域增长算法得到精确路面区域, 去除路基、 车辆的影响, 确保所选像素 在世界坐标中的照度一致; 在路面区域中, 提取反映路面亮度变化的对比度曲线, 找寻 亮度曲线特征点, 通过消光系数计算图像中人眼可分辨最远像素; 结合摄像机标定换算 出最大能见距离, 确定能见度值, 包括以下步骤:
1)通过设置的 PTZ视频摄像机, 实时获取路况视频图像信息;
2)从视频摄像机获取的图像信息, 通过摄像机定标技术完成视频图像与路面世界坐 标的转换关系, 计算视频图像中路面区域与摄像机的距离;
3)采用 Kluge模型拟合车道分割线在视频图像中的投影, 通过随机霍夫变化求解模 型中的未知参数, 车道分割线间的区域即为当前图像的兴趣域 ROI, 将后续的处理过程 限制于 ROI内, 确保图像像素的高度一致性;
4)采用区域增长算法, 结合 ROI区域内亮度判断准则以及自适应 Nagao滤波方法, 精确提取路面 mask区域, 将所有的后续处理限制在 mask区域中, 以减少计算量, 保证 所选像素在世界坐标中的照度一致; 计算 ROI 中最下面一行的灰度中值, 计为 median^), 选取亮度为 median(Pg)的像素作为种子点, 依照自下而上、 从左至右原则 逐行扫描 mask区域, 对扫描的目标像素点 根据如下增长准则, 依次判断是否属 于路面区域:
41)亮度均衡性
P(i, 与 median(Pg)之间满足
P(i,j) -median(Pg) < pnr minG^
(6)
(yt— 1, 0,1)
式 (6)中, ^是小于 1的常数, 代表 Υ, 和初始种子点 ^相隔行数, Gmx则指 某像素点与其上 3-邻域内亮度差阈值, 上 3-邻域即像素点的左上方、上方、 以及右上方 的三个像素点,左上方亮度差阈值 ^,上方亮度差阈值 G x,右上方亮度差阈值 ax, 其中-
G m—a 1x = G m1 ax < G m0ax (7)
42)照度一致性
采用自适应窗宽的 Nagao中值滤波器, 滤除图像中的噪声干扰, 同时不扩散噪声点 的能量, 满足亮度均衡性的像素点经过自适应窗宽的 Nagao中值滤波, 得到像素灰度值 Q(i, 满足:
3m e { - 1, i, i + 1}
Q(i,j)-Q(m,j + \)<Gm':x (8) 满足亮度连续性以及一致性的像素点即可加入到路面区域中, 直至 mask区域扫描 完毕, 精确的路面区域即可得到;
5)亮度特征提取: 利用前述步骤获得的初始照度一致、 高度一致的路面区域, 分析 路面像素因大气消光导致的亮度变化趋势, 找出其中的变化特征点, 即亮度曲线的二阶 导数零点;
6)能见度计算: 通过由摄像机标定算法求得的灭点坐标以及摄像机参数, 以及亮度 曲线求得的二阶导数零点坐标, 确定大气消光系数, 再利用科西米德定理推导出大气消 光系数与能见度之间的关系, 进而求解确定能见度值。
2、 根据权利要求 1所述的基于图像亮度特征的 PTZ视频能见度检测方法, 其特征 是步骤 2)中的摄像机定标为:
PTZ视频摄像机成像映射模型包括 3个坐标系: 路面世界坐标系 ( w, ,ZW), 摄像 机坐标系 ( c,;rc,zc)以及视频图像的像平面坐标系^, , Zc与路面夹角为 Θ , 摄像机光心 O距离路面高度为 , /为镜头有效焦距, 建立路面坐标系与摄像机坐标系间、 摄像机 坐标系与成像平面间的变换关系为:
Figure imgf000016_0001
由 yw=+∞可得, 地平线 vk, 也就是灭点在像平面中的投影为:
Figure imgf000016_0002
代入式 (2)中可以得到:
Y„ - L-ά νίθ Zr -d cos Θ
据此, 图像中像素所表征的路面区域距摄像机光心的距离 4可表示为: (5
Figure imgf000017_0001
3、 根据权利要求 1或 2所述的基于图像亮度特征的 PTZ视频能见度检测方法, 其 特征是步骤 6)中大气能见度与消光系数的关系为- 根据科西米德定理, 设 表示大气消光系数, 某一固定亮度的物体在距离人眼距离 为 ί处的视亮度或辐射率 L与物体本身亮度 及背景亮度 关系如下式:
L = L0e M + Lf {\ - e M) (9) 式 (9)表明物体的视亮度由两部分组成: 物体固有亮度 以 消弱, 以及背景亮度 以 形式逐渐加强, 则对比度变化与大气消光系数 , 距离 ί之间的关系为:
L -Lr Ln -L(
C ■■ Cne~
L L
f (10) 式 (10)中, 是目标物接收亮度对比度, ce是固有亮度对比度, 当散射系数与方 位角无关且沿观测者、 目标物和地平天空之间的整个路径上的照度均匀时, 关系式 (10) 成立,
设^ ei为人眼所能观测到的最远距离, 即对比度为 0.05处的像素点, 有:
^ = -}ln(¾ = -}ln(0.05)
k Cn k
(i i) 关系式 (11)表明了大气能见度与消光系数之间的关系;
则基于亮度特征点的能见度求解为:
将图像亮度 L对像平面竖直方向坐标 V求二阶导数, 并将距离 ί用关系式 (5)代入,
Figure imgf000017_0002
在消光系数 的作用下, 图像像素亮度 及其导数随距离的变化而发生变化, 随着 雾的增大, 在天空背景下, 目标物变得更加模糊, 并且求得的极值点也在减小, 令二阶 导数为 0, 舍弃无意义的 ^=0的解, 得到:
k 2( ,- - (13) λ
¼即亮度曲线二阶拐点位置, 也就是二阶导数零点, ¼为地平线即灭点位置, 由此, 大气能见距离为:
V— (14) k 2(v,. -vA) 当 趋近于 ¼的时候, !^^处于临界状态, 认为此时刚好看见雾的出现; 当 大于 W的时候, 雾是可以检测得到的; 相反, 小于 ¼的时候, 认为没有雾的出现。
4、 根据权利要求 1或 2所述的基于图像亮度特征的 ΡΤΖ视频能见度检测方法, 其 特征是步骤 5)具体为, 在步骤 4)得到的路面区域中, 依次找出第 行最大长度连续像素 集 Pix(i,j),起始于 (α, ;,终止于 (¾,;,其长度 /£^/?(/;)吖 ό-α+^_, ¾ :)中点 middle 为 ((b+a)/2,j), 各行的中点坐标 middle作为路面测量带的中心, 在式(15)的条件下, 组 成一条路面测量带:
len(Pix(j)) = min(Lengh, lengh(j)) (15) 式中 代表组成测量带的第 '行点集, fe< ¾c^^即为其长度, 为一设置的 常数阈值; 获取测量带内每行中值亮度, 得到亮度-距离变化曲线 对 S求二阶导数, 确定变化特征点 即二阶导数零点,并且在寻找二阶导数零点前,对曲线 S进行插值、 滤波, 消除混淆二阶零点, 以减小误差, 得到尽可能精确测量结果。
5、 根据权利要求 3所述的基于图像亮度特征的 PTZ视频能见度检测方法, 其特征 是步骤 5)具体为, 在步骤 4)得到的路面区域中, 依次找出第 行最大长度连续像素点集 Pix(i, j), 起始于 (α, , 终止于 其长度 /£^^(/;)= -0+^_, ¾ ', 中点 middle为 ((b+a)/2,j), 各行的中点坐标 middle作为路面测量带的中心, 在式(15)的条件下, 组成 一条路面测量带:
len(Pix(j)) = min(Lengh, lengh(j)) (15) 式中 代表组成测量带的第 '行点集, fe< ¾c^^即为其长度, 为一设置的 常数阈值; 获取测量带内每行中值亮度, 得到亮度-距离变化曲线 对 S求二阶导数, 确定变化特征点 即二阶导数零点,并且在寻找二阶导数零点前,对曲线 S进行插值、 滤波, 消除混淆二阶零点, 以减小误差, 得到尽可能精确测量结果。
PCT/CN2011/078247 2011-01-26 2011-08-11 基于图像亮度特征的ptz视频能见度检测方法 Ceased WO2012100522A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US13/981,163 US9225943B2 (en) 2011-01-26 2011-08-11 PTZ video visibility detection method based on luminance characteristic

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201110028103.5 2011-01-26
CN2011100281035A CN102175613B (zh) 2011-01-26 2011-01-26 基于图像亮度特征的ptz视频能见度检测方法

Publications (1)

Publication Number Publication Date
WO2012100522A1 true WO2012100522A1 (zh) 2012-08-02

Family

ID=44518820

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2011/078247 Ceased WO2012100522A1 (zh) 2011-01-26 2011-08-11 基于图像亮度特征的ptz视频能见度检测方法

Country Status (3)

Country Link
US (1) US9225943B2 (zh)
CN (1) CN102175613B (zh)
WO (1) WO2012100522A1 (zh)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104634740A (zh) * 2013-11-12 2015-05-20 中国电信股份有限公司 雾霾能见度监测方法和装置
CN106404720A (zh) * 2016-08-24 2017-02-15 中国气象科学研究院 一种能见度观测方法
CN108961391A (zh) * 2018-06-12 2018-12-07 温州大学激光与光电智能制造研究院 一种基于曲率滤波的表面重构方法
CN109919889A (zh) * 2019-02-28 2019-06-21 温州大学 一种基于双目视差的能见度检测算法
CN110378865A (zh) * 2019-04-28 2019-10-25 湖南师范大学 一种复杂背景下的雾天能见度智能分级识别方法及系统
CN111899309A (zh) * 2020-07-31 2020-11-06 上海眼控科技股份有限公司 上坡雾检测方法、装置、计算机设备和可读存储介质
CN112017243A (zh) * 2020-08-26 2020-12-01 大连信维科技有限公司 一种介质能见度识别方法
CN112184725A (zh) * 2020-09-21 2021-01-05 东南大学 一种沥青路面图像的结构光光条中心提取方法
US20210358162A1 (en) * 2014-04-28 2021-11-18 Canon Kabushiki Kaisha Image processing method and image capturing apparatus
CN113959987A (zh) * 2021-09-27 2022-01-21 湖南国天电子科技有限公司 一种机场能见度圆周运动测量方法及装置
CN114998790A (zh) * 2022-05-26 2022-09-02 武汉大学 一种基于毫米波雷达和机器视觉的路侧绿化监控方法
CN115629046A (zh) * 2022-10-14 2023-01-20 合肥中科光博量子科技有限公司 一种带自标定的大范围团雾快速监测系统

Families Citing this family (69)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103631786B (zh) * 2012-08-22 2017-04-12 腾讯科技(深圳)有限公司 一种视频文件的聚类方法和装置
JP6099948B2 (ja) * 2012-11-28 2017-03-22 京セラ株式会社 電子機器及び制御プログラム並びに表示制御方法
CN104634784B (zh) * 2013-11-08 2018-01-05 中国电信股份有限公司 大气能见度监测方法和装置
CN104648397B (zh) * 2013-11-19 2017-05-17 沙漠科技股份有限公司 车道偏移警示系统及方法
KR101498114B1 (ko) * 2013-11-28 2015-03-05 현대모비스 주식회사 보행자를 검출하는 영상 처리 장치 및 그 방법
CN103630496B (zh) * 2013-12-12 2015-09-30 南京大学 基于路面视亮度和最小二乘法的交通视频能见度检测方法
DE102014208271A1 (de) * 2014-05-02 2015-11-05 Robert Bosch Gmbh Verfahren und Vorrichtung zur bildbasierten Sichtweitenschätzung
CN103983239B (zh) * 2014-05-21 2016-02-10 南京航空航天大学 基于车道等宽线的测距方法
CN104794903B (zh) * 2014-06-11 2017-05-10 合肥齐飞信息技术有限公司 基于沿线插值原理的公路能见度优化算法
CN104251768B (zh) * 2014-09-18 2017-03-08 常熟实盈光学科技有限公司 判定摄像头模组中镜头阴影效应的方法
CN104537659B (zh) * 2014-12-23 2017-10-27 金鹏电子信息机器有限公司 双摄像机的自动标定方法和系统
CN104766270B (zh) * 2015-03-20 2017-10-03 北京理工大学 一种基于鱼眼镜头的虚实光照融合方法
CN105046703A (zh) * 2015-07-10 2015-11-11 南京信息工程大学 一种图像解析大气能见度方法
CN108391051B (zh) * 2015-07-14 2020-08-14 Oppo广东移动通信有限公司 一种对焦异常处理方法及移动终端
US10832426B2 (en) 2015-09-24 2020-11-10 Apple Inc. Systems and methods for surface monitoring
WO2017053407A1 (en) 2015-09-24 2017-03-30 Quovard Management Llc Systems and methods for localization using surface imaging
CN105261018B (zh) * 2015-10-14 2017-11-28 山东交通学院 基于光学模型和暗原色先验理论的能见度检测方法
CN107635119B (zh) * 2016-01-26 2019-05-21 北京智谷睿拓技术服务有限公司 投射方法和设备
CN105675623B (zh) * 2016-01-29 2018-05-25 重庆扬讯软件技术股份有限公司 一种基于污水口视频的污水颜色与流量检测的实时分析方法
CN105868495B (zh) * 2016-04-19 2019-02-12 江南大学 一种非旋转对称照明系统模型的快速建立方法
KR102597435B1 (ko) * 2016-04-20 2023-11-03 엘지이노텍 주식회사 영상 취득 장치 및 그 방법
EP3276529A1 (en) * 2016-07-28 2018-01-31 Araani BVBA Efficient smoke detection based on video data processing
CN106713872B (zh) * 2017-02-08 2023-10-17 广西交通科学研究院有限公司 一种基于图像的隧道洞内外亮度检测系统的检测方法
US10748012B2 (en) * 2018-02-13 2020-08-18 Ford Global Technologies, Llc Methods and apparatus to facilitate environmental visibility determination
JP6723492B2 (ja) * 2018-03-12 2020-07-15 三菱電機株式会社 霧特定装置、霧特定方法及び霧特定プログラム
CN108693087B (zh) * 2018-04-13 2020-09-01 中国科学院城市环境研究所 一种基于图像理解的空气质量监测方法
CN108830880B (zh) * 2018-04-25 2021-09-21 安徽超远信息技术有限公司 一种适用于高速公路的视频能见度检测预警方法及其系统
CN108765453B (zh) * 2018-05-18 2022-02-18 武汉倍特威视系统有限公司 基于视频流数据的高速公路团雾识别方法
CN108956397B (zh) * 2018-05-23 2019-10-29 南京大学 一种基于迹范数的道路能见度检测方法
IT201800005817A1 (it) * 2018-05-29 2019-11-29 Dispositivo e metodo per misurazioni di visibilita'
CN109034104A (zh) * 2018-08-15 2018-12-18 罗普特(厦门)科技集团有限公司 一种场景标签定位方法以及装置
CN109285187A (zh) * 2018-09-11 2019-01-29 东南大学 一种基于道路监控视频图像的最远可见点检测方法
TWI692969B (zh) * 2019-01-15 2020-05-01 沅聖科技股份有限公司 攝像頭自動調焦檢測方法及裝置
CN109752089B (zh) * 2019-01-28 2021-01-15 大连海事大学 一种基于中间视觉特性和光源透雾性的亮度计算方法
US10944900B1 (en) * 2019-02-13 2021-03-09 Intelligent Security Systems Corporation Systems, devices, and methods for enabling camera adjustments
CN109978842A (zh) * 2019-03-14 2019-07-05 藏龙信息科技(苏州)有限公司 一种基于摄像头图像的能见度解析方法
CN110097762B (zh) * 2019-03-25 2020-10-09 南京微达电子科技有限公司 一种道路视频图像低能见度刻度估算方法及系统
DE102019207404A1 (de) * 2019-05-21 2020-11-26 OSRAM Opto Semiconductors Gesellschaft mit beschränkter Haftung Optoelektronische Messvorrichtung zur Messung einer Intensität einer elektromagnetischen Strahlung
CN110926515B (zh) * 2019-10-20 2021-09-10 中国航空工业集团公司洛阳电光设备研究所 一种不同大气能见度下光电传感器等效作用距离折算方法
CN110853024B (zh) * 2019-11-14 2020-12-22 推想医疗科技股份有限公司 医疗图像处理方法、装置、存储介质及电子设备
CN110888441B (zh) * 2019-11-29 2023-04-07 广州乐比计算机有限公司 一种基于陀螺仪的轮椅车控制系统
CN110889876B (zh) * 2019-12-10 2022-05-03 兰州交通大学 一种基于ca-spcnn算法的彩色图像量化方法
CN111275698B (zh) * 2020-02-11 2023-05-09 西安汇智信息科技有限公司 基于单峰偏移最大熵阈值分割雾天道路能见度检测方法
CN111627222A (zh) * 2020-05-12 2020-09-04 浙江工贸职业技术学院 一种高速公路二次交通事故预警装置及方法
CN111738314B (zh) * 2020-06-09 2021-11-02 南通大学 基于浅层融合的多模态图像能见度检测模型的深度学习方法
CN111736237A (zh) * 2020-07-31 2020-10-02 上海眼控科技股份有限公司 辐射雾检测方法、装置、计算机设备和可读存储介质
CN111785094B (zh) * 2020-07-31 2021-12-07 上海眼控科技股份有限公司 平流雾检测方法、装置、计算机设备和可读存储介质
CN112330553B (zh) * 2020-10-30 2022-07-01 武汉理工大学 一种裂缝图像去噪方法、设备及存储介质
CN112598655A (zh) * 2020-12-25 2021-04-02 上海眼控科技股份有限公司 一种能见度检测方法及设备
CN113762057B (zh) * 2021-05-21 2024-11-08 南京御达电信息技术有限公司 一种基于毫米波雷达与多传感器信息融合的能见度测量方法
US11783453B2 (en) 2021-06-10 2023-10-10 Bank Of America Corporation Adapting image noise removal model based on device capabilities
CN113566723B (zh) * 2021-08-09 2023-03-21 中国商用飞机有限责任公司北京民用飞机技术研究中心 一种零件间隙与干涉检查方法及其系统
US12462575B2 (en) 2021-08-19 2025-11-04 Tesla, Inc. Vision-based machine learning model for autonomous driving with adjustable virtual camera
CN113850768A (zh) * 2021-09-15 2021-12-28 上海眼控科技股份有限公司 基于图像的能见度计算方法与设备
CN114119545A (zh) * 2021-11-25 2022-03-01 安徽百诚慧通科技有限公司 一种高速公路能见度估算方法、系统、设备及存储介质
CN114267279B (zh) * 2021-12-21 2024-04-19 昆山国显光电有限公司 补偿灰阶确定方法、装置及设备
CN114527046B (zh) * 2022-01-05 2024-04-09 上海应用技术大学 一种多因素影响下视敏度测试装置
CN114399924B (zh) * 2022-02-15 2023-07-11 青岛海信网络科技股份有限公司 一种车辆、边缘计算设备、服务器及信息传输方法
CN114550118B (zh) * 2022-02-23 2023-07-11 烟台大学 一种基于视频图像驱动的高速公路全自动智能划线方法
CN116797517A (zh) * 2022-03-16 2023-09-22 小米汽车科技有限公司 对象检测方法、装置及车辆
CN114720425B (zh) * 2022-04-24 2023-02-21 安徽气象信息有限公司 一种基于图像识别的能见度监测系统及方法
US20230406356A1 (en) * 2022-05-20 2023-12-21 Tesla, Inc. Fail-safe corrective actions based on vision information for autonomous vehicles
CN117291864A (zh) * 2022-06-17 2023-12-26 纬哲纽咨信息咨询有限公司 能见度估计装置及方法、以及记录介质
CN115115617A (zh) * 2022-07-26 2022-09-27 安徽气象信息有限公司 一种应用于气象能见度检测仪的能见度检测系统
US11715305B1 (en) * 2022-11-30 2023-08-01 Amitha Nandini Mandava Traffic detection system using machine vision
CN115684637B (zh) * 2022-12-30 2023-03-17 南京理工大学 基于路侧单目相机标定的高速公路车辆测速方法及设备
CN116930261A (zh) * 2023-07-26 2023-10-24 南京信息工程大学 一种基于图像识别的气象检测方法及用于检测的冷镜式露点仪
CN118781559B (zh) * 2024-09-06 2024-12-06 四川智慧高速科技有限公司 基于图像处理的低能见度气象灾害处置方法、系统及设备
CN119147097B (zh) * 2024-11-11 2025-02-14 杭州虹谱光色科技有限公司 一种成像照度分布的测量分析方法

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3501239A (en) * 1964-10-14 1970-03-17 Acec Measuring device for the maximum visual distance in the fog
CN101382497A (zh) * 2008-10-06 2009-03-11 南京大学 基于路况监控视频的能见度检测方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2632724B1 (fr) * 1988-06-14 1990-09-14 Univ Lille Flandres Artois Procede et dispositif de mesure de la visibilite a travers un milieu d'opacite variable
CN101118648A (zh) * 2007-05-22 2008-02-06 南京大学 交通监视环境下的路况摄像机标定方法

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3501239A (en) * 1964-10-14 1970-03-17 Acec Measuring device for the maximum visual distance in the fog
CN101382497A (zh) * 2008-10-06 2009-03-11 南京大学 基于路况监控视频的能见度检测方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
AN, MINGWEI ET AL.: "Visibility detection method and system design based on traffic video", CHINESE JOURNAL OF SCIENTIFIC INSTRUMENT, vol. 31, no. 5, May 2010 (2010-05-01), CHINESE, pages 1148 - 1153 *
CHEN, ZHAOZHENG ET AL.: "Video Contrast Visibility Detection Algorithm and Its Implementation Based on Camera Self-calibration", JOURNAL OF ELECTRONICS & INFORMATION TECHNOLOGY, vol. 32, no. 12, December 2010 (2010-12-01), pages 2907 - 2912 *
I, BO ET AL.: "Visibility Detection Based on Traffic Video Contrast Analysis without Artificial arkers", JOURNAL OF COMPUTER-AIDED DESIGN & COMPUTER GRAPHICS, vol. 21, no. 11, pages 1575 - 1582 *

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104634740A (zh) * 2013-11-12 2015-05-20 中国电信股份有限公司 雾霾能见度监测方法和装置
US20210358162A1 (en) * 2014-04-28 2021-11-18 Canon Kabushiki Kaisha Image processing method and image capturing apparatus
CN106404720A (zh) * 2016-08-24 2017-02-15 中国气象科学研究院 一种能见度观测方法
CN106404720B (zh) * 2016-08-24 2019-02-26 中国气象科学研究院 一种能见度观测方法
CN108961391A (zh) * 2018-06-12 2018-12-07 温州大学激光与光电智能制造研究院 一种基于曲率滤波的表面重构方法
CN109919889B (zh) * 2019-02-28 2022-12-09 温州大学 一种基于双目视差的能见度检测算法
CN109919889A (zh) * 2019-02-28 2019-06-21 温州大学 一种基于双目视差的能见度检测算法
CN110378865A (zh) * 2019-04-28 2019-10-25 湖南师范大学 一种复杂背景下的雾天能见度智能分级识别方法及系统
CN111899309A (zh) * 2020-07-31 2020-11-06 上海眼控科技股份有限公司 上坡雾检测方法、装置、计算机设备和可读存储介质
CN112017243A (zh) * 2020-08-26 2020-12-01 大连信维科技有限公司 一种介质能见度识别方法
CN112017243B (zh) * 2020-08-26 2024-05-03 大连信维科技有限公司 一种介质能见度识别方法
CN112184725A (zh) * 2020-09-21 2021-01-05 东南大学 一种沥青路面图像的结构光光条中心提取方法
CN112184725B (zh) * 2020-09-21 2024-02-02 东南大学 一种沥青路面图像的结构光光条中心提取方法
CN113959987A (zh) * 2021-09-27 2022-01-21 湖南国天电子科技有限公司 一种机场能见度圆周运动测量方法及装置
CN113959987B (zh) * 2021-09-27 2023-08-29 湖南国天电子科技有限公司 一种机场能见度圆周运动测量方法及装置
CN114998790A (zh) * 2022-05-26 2022-09-02 武汉大学 一种基于毫米波雷达和机器视觉的路侧绿化监控方法
CN115629046A (zh) * 2022-10-14 2023-01-20 合肥中科光博量子科技有限公司 一种带自标定的大范围团雾快速监测系统

Also Published As

Publication number Publication date
CN102175613A (zh) 2011-09-07
US9225943B2 (en) 2015-12-29
US20130342692A1 (en) 2013-12-26
CN102175613B (zh) 2012-11-14

Similar Documents

Publication Publication Date Title
CN102175613B (zh) 基于图像亮度特征的ptz视频能见度检测方法
CN104809707B (zh) 一种单幅雾天图像能见度估计方法
KR101364727B1 (ko) 촬영된 영상의 처리를 이용한 안개 감지 방법 및 장치
CN101957309B (zh) 一种能见度的全天候视频测量方法
CN101281142B (zh) 一种测量大气能见度的方法
CN103630496B (zh) 基于路面视亮度和最小二乘法的交通视频能见度检测方法
JP6646527B2 (ja) 物体検知評価システム及び物体検知評価方法
CN105261018B (zh) 基于光学模型和暗原色先验理论的能见度检测方法
CN109801302A (zh) 一种基于双目视觉的高压输电线路异物检测方法
CN103458156B (zh) 一种恶劣天气条件下交通事件检测视频信号预处理方法
CN102592454A (zh) 一种基于车辆侧面与路面交线检测的交叉口车辆运动参数测量方法
CN110321855A (zh) 一种雾天检测预警装置
Awrangjeb et al. Improved building detection using texture information
Bugarić et al. Adaptive estimation of visual smoke detection parameters based on spatial data and fire risk index
CN120449111A (zh) 基于多模态数据融合的街道清洁度实时评估方法
CN115909285A (zh) 一种雷达和视频信号融合的车辆跟踪方法
KR101874968B1 (ko) 이미지 정보 기반의 시정거리 측정 시스템 및 이를 이용한 시정거리 측정 방법
CN112907739B (zh) 一种井盖高程差信息获取方法、装置及系统
CN118052764A (zh) 雾天图像能见度检测方法、装置、计算机设备及存储介质
CN108956397B (zh) 一种基于迹范数的道路能见度检测方法
Laureshyn et al. Automated video analysis as a tool for analysing road user behaviour
CN105469604A (zh) 一种基于监控图像的隧道内车辆检测方法
Chen PTZ visibility detection based on image luminance changing tendency
CN119618151A (zh) 输电线路杆塔姿态及树线距离实时监测及预警方法
Zhao-Zheng et al. Real-time video detection of road visibility conditions

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11856853

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 13981163

Country of ref document: US

122 Ep: pct application non-entry in european phase

Ref document number: 11856853

Country of ref document: EP

Kind code of ref document: A1