WO2010069167A1 - 识别图像中障碍物的方法和装置 - Google Patents

识别图像中障碍物的方法和装置 Download PDF

Info

Publication number
WO2010069167A1
WO2010069167A1 PCT/CN2009/071577 CN2009071577W WO2010069167A1 WO 2010069167 A1 WO2010069167 A1 WO 2010069167A1 CN 2009071577 W CN2009071577 W CN 2009071577W WO 2010069167 A1 WO2010069167 A1 WO 2010069167A1
Authority
WO
WIPO (PCT)
Prior art keywords
confidence
obstacle
current frame
motion
image
Prior art date
Application number
PCT/CN2009/071577
Other languages
English (en)
French (fr)
Inventor
段勃勃
刘威
袁淮
Original Assignee
东软集团股份有限公司
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 东软集团股份有限公司 filed Critical 东软集团股份有限公司
Priority to JP2011539877A priority Critical patent/JP5276721B2/ja
Priority to US13/133,546 priority patent/US8406474B2/en
Priority to DE112009003144.7T priority patent/DE112009003144B4/de
Publication of WO2010069167A1 publication Critical patent/WO2010069167A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/50Context or environment of the image
    • G06V20/56Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads

Definitions

  • the present invention relates to the field of obstacle recognition technology, and more particularly to a method and apparatus for identifying obstacles in an image.
  • an obstacle generally refers to a three-dimensional object that is higher than the ground.
  • the motion compensation based method is one of the commonly used methods for obstacle detection based on monocular vision. Its principle is that the pixel value of the corresponding image point in the image frame of the adjacent time is unchanged at any point of the road plane under the condition that the road is flat and the illumination condition is unchanged in a short time. If it is assumed that all the points in the image at the previous moment are the corresponding imaging points of the points on the road surface, then by the camera motion parameters and the imaging principle, it can be calculated that all the points in the image from the previous moment will be moved after the camera is moved at the next moment.
  • the imaginary image composed, the difference between the hypothetical image and the image actually captured at the current time is caused by those points that are not on the road plane.
  • the image pixels corresponding to these differences may be obstacles that protrude from the ground.
  • the current method is mainly based on the result of motion compensation to judge that there is an obstacle in a certain area of the image. This recognition strategy will result in the result of motion compensation being unsatisfactory under the influence of low accuracy of motion parameters or image noise. The effect is that some non-obstructions are misidentified as obstacles.
  • the embodiment of the present invention provides a method and a device for identifying an obstacle in an image, which can reduce the erroneous detection.
  • the method for identifying an obstacle in an image provided by the embodiment of the present invention includes:
  • An obstacle in the image is determined based on each block area.
  • the step of determining whether each block region in the current frame image is an obstacle includes: if the current frame and the same block region corresponding to the previous N frame corresponding to the current frame have a value of the confidence of the obstacle of motion equal to 1 is greater than a first threshold number, the current frame image overall confidence in the block of the 1 _1 ( ⁇ 1 1, otherwise the overall confidence region C_Total block is 0;
  • each block region in the current frame image sequentially according to the current frame and the motion obstacle confidence of each block region of the previous N frames closest to the current frame, and the feature obstacle confidence of each block region Whether it is an obstacle; An obstacle in the image is determined based on each block area.
  • the characteristic obstacle confidence includes a characteristic obstacle confidence based on a vertical characteristic or a characteristic obstacle confidence based on the texture characteristic.
  • the step of determining whether a region of the current frame has a vertical attribute includes:
  • aOl calculating the intensity of the vertical direction of a block region / v ;
  • the gray value of the row and column pixels is an integer, i' je R , ? is the image block area, N is the image width, and M is the image height.
  • step aO1 If the intensity I v of the vertical direction in step aO1 is greater than the intensity threshold, the block area has a vertical attribute, otherwise the block area does not have a vertical attribute.
  • the step of determining whether each block region in the current frame image is an obstacle includes: if the current frame and the same block region corresponding to the previous N frame corresponding to the current frame have a value of the confidence of the obstacle of motion equal to The number of 1 is greater than the first number threshold, and the current frame is closest to the current frame The value of the characteristic obstacle confidence of the certain same block region corresponding to the previous N frame is equal to the number of 1 is greater than the second number threshold, and the total confidence of the block region in the current frame image. - ⁇ is 1, otherwise the total confidence C_Total of the block area is 0;
  • An image segmentation unit configured to acquire an image of a current frame and a previous N frame that is closest to the current frame, and divide each acquired image in the same manner, and obtain a plurality of divided block regions for each frame image;
  • a motion obstacle confidence calculation unit configured to calculate a motion obstacle confidence level of each block region corresponding to a current frame and a previous N frame closest to the current frame
  • a first block obstacle recognition unit configured to sequentially determine, according to the current frame and the motion obstacle confidence of each block region of the previous N frames closest to the current frame, whether each block region in the current frame image is sequentially determined As an obstacle;
  • An obstacle determining unit configured to determine an obstacle in the image according to each block region.
  • the dyskine obstacle confidence calculation unit includes:
  • a first motion confidence calculation unit configured to calculate a degree of similarity between the current frame and the imaginary image for a block region, to obtain a first motion confidence C_M_A1;
  • a second motion confidence calculation unit configured to calculate a similarity degree between the current frame and the image of the n-k time before the current frame for one block region, to obtain a second motion confidence C_M_A2;
  • a dyskine obstacle confidence determining unit configured to: at a value of the first motion confidence C_M_A1 being greater than a first motion threshold, and a ratio of the first motion confidence C_M_A1 to a second motion confidence C_M_A2 being greater than a second motion threshold When it is determined, the motion obstacle confidence C_M of the block region of the current frame is 1, otherwise it is determined that the motion obstacle confidence C_M of the block region of the current frame is 0.
  • the first block area obstacle recognition unit includes:
  • a first total confidence determining unit configured to: in the current frame and the same block area corresponding to the previous N frame corresponding to the current frame, the value of the motion obstacle confidence value equal to 1 is greater than the first quantity threshold The value, the total confidence C_Total of the block area in the current frame image is determined to be 1, otherwise the total confidence C_Total of the block area is determined to be 0;
  • the first identifying unit is configured to determine that the block area is an obstacle when the total confidence C_Total of the certain block area in the current frame image is 1, and otherwise determine that the block area is a non-obstacle.
  • An image segmentation unit configured to acquire an image of a current frame and a previous N frame that is closest to the current frame, and divide each acquired image in the same manner, and obtain a plurality of divided block regions for each frame image;
  • a motion obstacle confidence calculation unit configured to calculate a motion obstacle confidence level of each block region corresponding to a current frame and a previous N frame closest to the current frame
  • a feature obstacle confidence calculation unit configured to calculate a feature obstacle confidence of each block region corresponding to the current frame and the previous N frame closest to the current frame;
  • a second block obstacle recognition unit configured to determine a moving obstacle confidence level of each of the block regions according to the current frame and the previous N frames closest to the current frame, and a feature obstacle confidence level of each block region , sequentially determining whether each block area in the current frame image is an obstacle;
  • An obstacle determining unit configured to determine an obstacle in the image according to each block region.
  • the dyskine obstacle confidence calculation unit includes:
  • a first motion confidence calculation unit configured to calculate a degree of similarity between the current frame and the imaginary image for a block region, to obtain a first motion confidence C_M_A1;
  • a second motion confidence calculation unit configured to calculate a similarity degree between the current frame and the image of the n-k time before the current frame for one block region, to obtain a second motion confidence C_M_A2;
  • a dyskine obstacle confidence determining unit configured to: at a value of the first motion confidence C_M_A1 being greater than a first motion threshold, and a ratio of the first motion confidence C_M_A1 to a second motion confidence C_M_A2 being greater than a second motion threshold When it is determined, the motion obstacle confidence C_M of the block region of the current frame is 1, otherwise it is determined that the motion obstacle confidence C_M of the block region of the current frame is 0.
  • the characteristic obstacle confidence includes a feature obstacle confidence based on a vertical characteristic, a characteristic obstacle confidence based on a vertical edge characteristic, or a characteristic obstacle confidence based on a texture characteristic Degree.
  • the characteristic obstacle confidence calculation unit includes:
  • a vertical attribute determining unit configured to determine whether a block area of the current frame has a vertical attribute, and notify the feature obstacle confidence determining unit of the determination result;
  • the feature obstacle confidence determining unit is configured to determine that the feature obstacle of the current frame has a confidence level of 1 when the current frame has a vertical attribute, and otherwise determine the feature obstacle confidence of the block area of the current frame.
  • C_F is 0.
  • the second block obstacle recognition unit includes:
  • a second total confidence determining unit configured to: in the current frame and the same block area corresponding to the previous N frame, the value of the motion obstacle confidence value equal to 1 is greater than the first quantity threshold, and Determining the total of the block area in the current frame image when the value of the feature obstacle confidence of the same frame region corresponding to the previous N frame corresponding to the current frame is equal to 1 is greater than the second number threshold Confidence C_Total is 1, otherwise the total confidence of the block area is determined (_1 1 ( ⁇ 1 is 0; the second identification unit is used to know the total confidence of the certain block area in the current frame image)
  • C_Total it is determined that the block area is an obstacle, otherwise it is determined that the block area is a non-obstacle.
  • a method and a device for identifying an obstacle in an image according to an embodiment of the present invention are compared with a conventional motion compensation method based on the absolute degree of motion compensation, and a relative motion compensation based The strategy reduces the false detection rate and accurately determines the obstacles in the image.
  • Another method and apparatus for identifying an obstacle in an image is not only based on a motion-based absolute degree strategy but also a motion compensation-based relative method, compared with a conventional motion compensation-only method. Based on this strategy, the strategy of feature analysis is further integrated, which further reduces the false detection rate and accurately determines the obstacles in the image.
  • FIG. 1 is a flow chart of a method of identifying obstacles in an image, in accordance with an embodiment of the present invention
  • FIG. 2 is a flow chart of a method for calculating a motion obstacle confidence level for each block region of a current frame, in accordance with an embodiment of the present invention
  • FIG. 3 is a schematic diagram showing the relationship between a set world coordinate system and a world coordinate system of a camera of a previous time image according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of a virtual image of a current moment according to an embodiment of the present invention.
  • FIG. 5 is a structural view of a device for identifying an obstacle in an image according to an embodiment of the present invention
  • FIG. 6 is a structural view of another device for identifying an obstacle in an image according to an embodiment of the present invention.
  • a method for identifying an obstacle in an image includes: acquiring an image of a current frame and a previous N frame that is closest to the current frame, and dividing each acquired image in the same manner, for each frame image. Obtaining a plurality of divided block regions; calculating a motion obstacle confidence of each block region corresponding to the current frame and the previous N frames closest to the current frame; according to the current frame and the previous N frames closest to the current frame. The motion obstacle confidence level of each of the block regions sequentially determines whether each block region in the current frame image is an obstacle; and determines an obstacle in the image according to each block region.
  • the invention compared with the traditional motion compensation only method, not only the strategy based on the absolute degree of motion compensation but also the relative strategy based on motion compensation is adopted, thereby reducing the false detection rate and accurately determining the image. Obstacle in the middle.
  • Another method for identifying an obstacle in an image includes: acquiring an image of a current frame and a previous N frame that is closest to the current frame, and dividing each acquired image in the same manner for each frame. Obtaining a plurality of divided block regions in the image; calculating a motion obstacle confidence level of each block region corresponding to the current frame and the previous N frames closest to the current frame, and a characteristic obstacle confidence level of each block region; The current frame and each of the previous N frames closest to the current frame The trajectory confidence of the block area, and the characteristic obstacle confidence of each block area, sequentially determines whether each block area in the current frame image is an obstacle; and determines an obstacle in the image according to each block area.
  • FIG. 1 there is shown a flow chart of a method of identifying obstacles in an image in accordance with an embodiment of the present invention.
  • the strategy based on the absolute degree of motion compensation is adopted, and the relative strategy based on motion compensation is also adopted.
  • the strategy of feature analysis is further integrated. Specifically include:
  • Step 101 Divide an image block area, specifically, obtain an image of a current frame and a previous N frame that is closest to the current frame, and divide each acquired image in the same manner, and obtain a plurality of divisions for each frame image. Block area.
  • an image of the current frame ie, an image representing the time n
  • each of the above-mentioned images is divided into a plurality of block regions, and the shape of the block region may be a rectangle, or may be any shape that can completely divide the image, such as a triangle, which is divided into: ⁇ ⁇ ⁇ pixels. Rectangular block areas that do not overlap each other, such that the same number of divided rectangular block areas are obtained for each frame of image.
  • the specific segmentation method is not limited herein, as long as it can be divided into a plurality of rectangular block regions.
  • Step 102 Calculate the motion obstacle confidence of each block region corresponding to the current frame and the previous frame closest to the current frame.
  • the so-called skeptical confidence refers to whether the block area after the motion compensation is an obstacle. Confidence, the motion compensation here is slightly different from the existing motion compensation.
  • the existing motion compensation usually only includes the absolute degree based on motion compensation, and the motion compensation in the present invention includes the absolute degree and relative degree based on the motion compensation.
  • Step 201 Acquire camera motion parameters.
  • the camera motion parameters obtained in the embodiments of the present invention may be obtained by a sensor, such as a speed sensor and a gyro sensor, or may be obtained by a sequence image, such as an optical flow method and a motion recovery structure (Structure). From Motion, SFM) method based on feature check, direct method, etc.
  • SFM motion recovery structure
  • the method estimates the motion parameters, 2 "Transforming camera geometry to a virtual downward-looking camera: robust ego-motion estimation and ground-layer detection” using an improved direct method to obtain the motion parameters.
  • the embodiment of the present invention uses the direct estimation method of Document 1 to acquire camera motion parameters.
  • the world coordinate system [ ⁇ ; , H] and the origin coordinate G of the camera coordinate system [ ⁇ ' ; ⁇ , , ⁇ ] of the image of the previous moment are set, and the rotation angle may be set between the coordinate axes.
  • the axis of the world coordinate system is parallel to the road plane and perpendicular to the road plane. As shown in Figure 3.
  • a pixel point c) in the image of the previous moment is r
  • c is the row coordinate and column coordinate of the point in the image respectively
  • P is the point of the point P 0 (X w , ⁇ , Z w ) on the road plane.
  • the point p can be calculated from the camera imaging formula (also known as the camera imaging formula). World sitting
  • is the rotation angle of the camera coordinate system around the world coordinate system U, Z axis, For camera external parameters, obtained when the camera is installed;
  • the coordinates of the camera coordinate system origin in the world coordinate system after movement are 7 ⁇ ⁇ ' ⁇ + ⁇ , and the rotation angle of the camera coordinate system around the world coordinate system ⁇ becomes ⁇ . , ⁇ + ⁇ , ⁇ .
  • the rotation matrix and translation matrix of the camera at this time and the calculated ⁇ .
  • the right end of the equation is known, and the left end unknown number, that is, the point p can be obtained.
  • the imaging coordinates in the imaginary image Specifically, based on the above transformation, all the points on the previous image F ⁇ can be moved according to the motion parameters, and a new hypothetical image F demo is generated.
  • the plane area in the image is exactly the same as the plane area of the image F Dock at the current time, but the plane area is different.
  • the following principle is used to determine the confidence that the area is a three-dimensional object.
  • Step 203 Calculate, for a block region, a similarity degree between the current frame and the virtual image of the previous n-k time, and use the obtained similarity degree as the first motion confidence C_M_A1 based on the motion compensation.
  • the degree of similarity between the current frame and the imaginary image of the previous nk time is referred to as the first motion confidence C_M_A1, and the first motion confidence is actually based on the absolute degree of motion compensation;
  • the degree of similarity of the nk time image is called the second motion confidence C_M_A2, and the second motion confidence is actually not motion compensated; and the ratio of the first motion confidence C_M_A1 to the second motion confidence C_M_A2 is actually based on motion compensation. Relative degree.
  • the step 203 includes: searching for pixels with the same coordinates in the imaginary image for all the pixels in a certain block region in the image, and then calculating the current time image of the block region and the imaginary image of the current time based on the pixel values of the corresponding points.
  • the size of the block area is N*M, which is the gray value of the pixel ( ) in the image
  • X ' ( , j) is the gray value of the pixel ( , j) in the imaginary image F n of the image.
  • Step 204 Calculate, for a block region, a similarity degree between the current frame and the previous nk time image, and use the obtained similarity degree as the second motion confidence C_M_A2 that does not perform motion compensation; specifically, for a block region in the image. All the pixels, find the pixels with the same coordinates in F ⁇ , and then calculate the similarity degree between the current time image of the block region and the previous nk time image based on the pixel values of the corresponding point pair, thereby obtaining the block region as the second obstacle
  • the motion confidence C_M_A2 wherein the similarity NC can be calculated using a normalized correlation value (NC):
  • N*M which is the gray value of the pixel ( ) in the image.
  • X "(i, j is the pixel (i, j) gray value in the image F: at the previous n-k time.
  • Step 205 Determine a motion obstacle confidence C_M.
  • the block of the current frame The occlusion obstacle confidence C_M of the region is 1, otherwise the kinetic obstacle confidence C_M of the block region of the current frame is 0.
  • the above-mentioned dyskine obstacle confidence C_M is based on the absolute degree of motion compensation (ie, the case where the value of C_M_A1 is greater than the first motion threshold), and the relative degree based on the motion compensation (ie, the ratio of C_M_A1 and C_M_A2 is greater than the second motion threshold) Situation) obtained.
  • the ratio of the first motion confidence C_M_A1 to the second motion confidence C_M_A2 is a generalized concept, and the ratio may be the value of C_M_A1/C_M_A2 or (C_M_A1-C_M_A2)/(1- The value of C_M_A2) can also be the value of other representations.
  • step 206 the above steps 203 ⁇ 205 are repeatedly performed until the motion obstacle confidence of each block region is calculated.
  • Step 103 Calculate a feature obstacle confidence C_F of each block region corresponding to the current frame and the previous N frames closest to the current frame.
  • the feature obstacle confidence includes a feature obstacle confidence based on a vertical characteristic, a feature obstacle confidence based on a vertical edge characteristic, or a feature obstacle confidence based on a texture characteristic.
  • the vertical characteristic is taken as an example to illustrate how to calculate the characteristic obstacle confidence C_F.
  • Step a determining whether a region of the current frame has a vertical attribute, if yes, the feature obstacle confidence C_F of the block region of the current frame is 1, otherwise the feature obstacle confidence C_F of the block region of the current frame is 0.
  • the specific step of determining whether a region of the current frame has a vertical attribute includes:
  • aOl calculating the intensity of the vertical direction of a block region / v ;
  • the gray value of the column of pixels, k is an integer, i' je R , is the image block area, N is the image block area width, and M is the image block area height.
  • step b step a is repeated until the characteristic obstacle confidence of each block region is calculated.
  • Step 104 sequentially determine, according to the current frame and the motion obstacle confidence of each block region of the previous N frames closest to the current frame, and the feature obstacle confidence of each block region, sequentially determine each of the current frame images. Whether the block area is an obstacle. Specifically, the confidence level of the motion obstacle corresponding to the latest N frame is taken as a region, and is set to C_M t , C_M t _ 1 , . . . , C_M t _ N _ 1 , and the value of the statistical obstacle of the motion obstacle is equal to 1.
  • Sum_M taking the block region corresponding to the feature confidence of the last N frames, set to C_F t , C_F M ,..., CV N 4 , the statistical feature obstacle confidence value is equal to the number of Sum_F; if Sum_M is greater than the first The quantity threshold S_M_Limit, and Sum_F is greater than the second number threshold C_F_Limit, then the total confidence C_Total is 1, otherwise 0. If the total confidence of a certain block area in the current frame image ( _1 1 ( ⁇ 1 is 1 , the block area is Obstacle, otherwise the block area is non-obstacle.
  • the total confidence C_Total of the block area in the current frame image is 1, otherwise The total confidence C_Total of the block area is 0; the total confidence of a block area in the current frame image.
  • Step 105 Determine an obstacle in the image according to each block area.
  • FIG. 1 shows a preferred embodiment. In practical applications, there may be another case where only motion compensation (including absolute degree and relative degree) is performed, and compensation is no longer performed based on feature analysis. The same can be achieved to reduce the false detection rate. In this embodiment, the steps are substantially the same as those shown in FIG.
  • step 103 is no longer performed, and the corresponding step 104 becomes: according to the current frame and each of the previous N frames closest to the current frame.
  • the occlusion obstacle confidence of the block area sequentially determines whether each block area in the current frame image is an obstacle.
  • the step of determining whether each block region in the current frame image is an obstacle includes: specifically, taking a region to correspond to the confidence of the motion obstacle of the latest N frame, and setting C_M t , C_M M , ..., C_M t _ N4 , the value of the statistical obstacle of the dynamometer is equal to the number of Sum_M; Sum_M is greater than the first number threshold S_M_Limit, then the total confidence C_Total is 1, otherwise 0.
  • the embodiment of the invention further provides a device for identifying an obstacle in an image, see FIG. 5, and a figure thereof.
  • an image segmentation unit 501 there are: an image segmentation unit 501, a motion obstacle confidence calculation unit 502, a feature obstacle confidence calculation unit 503, a second block obstacle recognition unit 504, and an obstacle determination unit 505. among them,
  • the image segmentation unit 501 is configured to acquire an image of a current frame and a previous N frame that is closest to the current frame, divide the acquired image of each frame in the same manner, and obtain a plurality of divided block regions for each frame of the image;
  • the dyskine obstacle confidence calculation unit 502 is configured to calculate a motion obstacle confidence level of each block region corresponding to the current frame and the previous N frames closest to the current frame,
  • the feature obstacle confidence calculation unit 503 is configured to calculate a feature obstacle confidence level of each block region corresponding to the current frame and the previous N frames closest to the current frame; the feature obstacle confidence may include a feature based on vertical characteristics Obstacle confidence, characteristic obstacle confidence based on vertical edge characteristics, or characteristic obstacle confidence based on texture characteristics.
  • a second block obstacle recognition unit 504 configured to perform a motion obstacle confidence level of each of the block regions according to the current frame and the previous N frames closest to the current frame, and a characteristic obstacle of each block region Degree, sequentially determining whether each block area in the current frame image is an obstacle;
  • An obstacle determining unit 505 is configured to determine an obstacle in the image according to each block region.
  • the kinetic obstacle confidence calculation unit 502 may further include: a hypothetical image generation unit, a first motion confidence calculation unit, a second motion confidence calculation unit, and a motion obstacle confidence determination Unit. among them,
  • a first motion confidence calculation unit configured to calculate a degree of similarity between the current frame and the imaginary image for a block region, to obtain a first motion confidence C_M_A1;
  • a second motion confidence calculation unit configured to calculate a similarity degree between the current frame and the image of the n-k time before the current frame for one block region, to obtain a second motion confidence C_M_A2;
  • a dyskine obstacle confidence determining unit configured to: at a value of the first motion confidence C_M_A1 being greater than a first motion threshold, and a ratio of the first motion confidence C_M_A1 to a second motion confidence C_M_A2 being greater than a second motion threshold Determining, the motion obstacle confidence C_M of the block area of the current frame is 1, otherwise determining that the motion obstacle confidence C_M of the block area of the current frame is 0; when the above characteristic obstacle confidence is based on the vertical characteristic,
  • the feature obstacle confidence calculation unit 503 may include: a vertical attribute determination unit and a feature obstacle confidence determination unit.
  • the vertical attribute determining unit is configured to determine whether a certain block area of the current frame has a vertical attribute, and notify the feature obstacle confidence determining unit of the determination result;
  • the feature obstacle confidence determining unit is configured to determine that the feature obstacle of the current frame has a confidence level of 1 when the current frame has a vertical attribute, and otherwise determine the feature obstacle confidence of the block area of the current frame.
  • C_F is 0.
  • the second block obstacle recognition unit 504 may include: a second total confidence determining unit and a second identifying unit, where
  • a second total confidence determining unit configured to: in the current frame and the same block area corresponding to the previous N frame, the value of the motion obstacle confidence value equal to 1 is greater than the first quantity threshold, and Determining the total of the block area in the current frame image when the value of the feature obstacle confidence of the same frame region corresponding to the previous N frame corresponding to the current frame is equal to 1 is greater than the second number threshold
  • the confidence C_Total is 1, otherwise the total confidence of the block area is determined ( _1 1 ( ⁇ 1 is 0; the second identifying unit is configured to know that the total confidence C_Total of the certain block area in the current frame image is 1) Determining that the block area is an obstacle, otherwise determining that the block area is a non-obstacle.
  • Embodiments of the present invention also provide a device for identifying an obstacle in an image, see FIG. 6, and
  • the difference between the embodiment shown in FIG. 5 is that the feature obstacle confidence calculation unit 503 is not included, that is, the apparatus shown in FIG. 6 includes: an image segmentation unit 601, a motion obstacle confidence calculation unit 602, and a first region obstacle.
  • the image segmentation unit 601 is configured to acquire an image of a current frame and a previous N frame that is closest to the current frame, and divide each acquired image in the same manner, and obtain a plurality of divided block regions for each frame image;
  • the dyskine obstacle confidence calculation unit 602 is configured to calculate a motion obstacle confidence level of each block region corresponding to the current frame and the previous N frames closest to the current frame,
  • the first block obstacle recognition unit 603 is configured to sequentially determine each block region in the current frame image according to the current frame and the motion obstacle confidence of each block region of the previous N frames closest to the current frame. Whether it is an obstacle;
  • the obstacle determining unit 604 is configured to determine an obstacle in the image according to each block region.
  • the kinetic obstacle confidence calculation unit 602 is the same as the kinic obstacle confidence calculation unit 502 in FIG. 5, and includes: a hypothetical image generation unit, a first motion confidence calculation unit, a second motion confidence calculation unit, and a motion obstacle The reliability determining unit is not described in detail.
  • the first block area obstacle recognition unit 603 may include:
  • a first total confidence determining unit configured to determine, when the value of the motion obstacle confidence value of the same block region corresponding to the current frame and the previous N frame corresponding to the current frame is equal to 1 and greater than the first quantity threshold The total confidence C_Total of the block area in the current frame image is 1, otherwise it is determined that the total confidence C_Total of the block area is 0;
  • the first identifying unit is configured to determine that the block area is an obstacle when the total confidence C_Total of the certain block area in the current frame image is 1, and determine that the block area is a non-obstacle.
  • the apparatus for identifying obstacles in an image provided by the present invention not only adopts a strategy based on the absolute degree of motion compensation but also adopts a relative strategy based on motion compensation, compared with a conventional logic device based only on motion compensation, and On this basis, the strategy of feature analysis is further integrated, which improves the accuracy of detection while reducing the false detection rate, and can accurately determine the obstacles in the image.

Description

识别图像中障碍物的方法和装置 本申请要求于 2008 年 12 月 15 日提交中国专利局、 申请号为 200810185833.4、 发明名称为"识别图像中障碍物的方法和装置"的中国专利申 请的优先权, 其全部内容通过引用结合在本申请中。
技术领域
本发明涉及障碍物识别技术领域,特别涉及识别图像中障碍物的方法和装 置。
背景技术
在本领域, 障碍物通常是指高于地面的立体物。
目前, 基于运动补偿的方法是基于单目视觉的障碍物检测常用方法之一。 它的原理是在道路平坦、短时间内光照条件不变前提下, 道路平面的任一点在 相邻时刻图像帧内所成对应像点的像素值不变。如果假设前一时刻图像中所有 的点都是路面上的点的对应的成像点, 则由相机运动参数和成像原理, 能够计 算出由前一时刻图像中所有的点在下一时刻相机发生运动后所组成的假想图 像,则该假设图像与当前时刻实际拍摄到的图像的差异都是由于那些不是道路 平面上的点所引起的。 这些差异所对应的图像像素即可能是突出地面的障碍 物。
目前的方法主要是单纯基于运动补偿的结果来判断图像中某个区域是存 在障碍物, 这种识别策略在运动参数精度较低或图像噪声的影响下,会导致运 动补偿的结果达不到理想的效果, 从而对于某些非障碍物被误识别为障碍物。
发明内容
本发明实施例提供一种识别图像中障碍物的方法和装置,能够降低误检测 本发明实施例提供的一种识别图像中障碍物的方法, 包括:
获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像 按相同方式进行划分, 对于每帧图像均获得若干个划分后的块区域;
计算当前帧以及与当前帧最近的之前 N帧所对应的每个块区域的运动障 碍物置信度;
根据所述当前帧以及与当前帧最近的之前 N帧的所述每个块区域的运动 障碍物置信度, 依次确定当前帧图像中每个块区域是否为障碍物;
根据每个块区域确定图像中的障碍物。
其中, 所述计算当前帧的每个块区域的运动障碍物置信度的步骤包括: i )获取运动参数,生成当前帧对应之前 n-k时刻的假想图像; n >=2, k>=l; ii )对于一个块区域, 计算该块的当前帧与所述假想图像的相似程度, 获 得第一运动置信度 C_M_A1;
iii )对于一个块区域, 计算该块的当前帧与当前帧之前 n-k时刻图像的相 似程度, 获得第二运动置信度 C_M_A2;
iv )对于一个块区域,若所述第一运动置信度 C_M_A1的值大于第一运动 阈值, 且所述第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值大 于第二运动阈值, 则当前帧的该块区域的运动障碍物置信度 C_M为 1 , 否则 当前帧的该块区域的运动障碍物置信度 C_M为 0;
¥ )重复执行步骤10至^ ,直到计算出每个块区域的运动障碍物置信度。 其中, 所述确定当前帧图像中每个块区域是否为障碍物的步骤包括: 若当前帧及与当前帧最近的之前 N帧所对应的某个相同块区域的运动障 碍物置信度的值等于 1的数量大于第一数量阈值,则当前帧图像中该块区域的 总置信度 _11(^1为 1 , 否则该块区域的总置信度 C_Total为 0;
若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。
本发明实施例提供的另一种识别图像中障碍物的方法, 包括:
获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像 按相同方式进行划分, 对于每帧图像均获得若干个划分后的块区域;
计算当前帧以及与当前帧最近的之前 N帧所对应的每个块区域的运动障 碍物置信度, 和每个块区域的特征障碍物置信度;
根据所述当前帧以及与当前帧最近的之前 N帧的所述每个块区域的运动 障碍物置信度, 和每个块区域的特征障碍物置信度,依次确定当前帧图像中每 个块区域是否为障碍物; 根据每个块区域确定图像中的障碍物。
其中, 所述计算当前帧的每个块区域的运动障碍物置信度的步骤包括: i )获取运动参数,生成当前帧对应之前 n-k时刻的假想图像; n >=2, k>=l; ii )对于一个块区域, 计算该块的当前帧与所述假想图像的相似程度, 获 得第一运动置信度 C_M_A1;
iii )对于一个块区域, 计算该块的当前帧与当前帧之前 n-k时刻图像的相 似程度, 获得第二运动置信度 C_M_A2;
iv )对于一个块区域,若所述第一运动置信度 C_M_A1的值大于第一运动 阈值, 且所述第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值大 于第二运动阈值, 则当前帧的该块区域的运动障碍物置信度 C_M为 1 , 否则 当前帧的该块区域的运动障碍物置信度 C_M为 0;
V )重复执行步骤 ii )至 iv ),直到计算出每个块区域的运动障碍物置信度。 其中, 所述特征障碍物置信度包括基于垂直特性的特征障碍物置信度、或 者基于纹理特性的特征障碍物置信度。
其中,所述根据垂直特性计算当前帧的每个块区域的特征障碍物置信度的 步骤包括:
a )判断当前帧一块区域是否具有垂直属性, 若有则当前帧该块区域的特 征障碍物置信度为 1 , 否则当前帧的该块区域的特征障碍物置信度 C_F为 0; b )重复执行步骤 a ), 直到计算出每个块区域的特征障碍物置信度。
其中, 所述判断当前帧一块区域是否具有垂直属性的步骤包括:
aOl )计算一个块区域的垂直方向的强度 /v; 其
Figure imgf000005_0001
行,第 列像素的灰度值, 为整数, i' j e R , ?为该图像块区域, N为图像宽度, M为图像高度。
a02 )若步骤 aOl )所述垂直方向的强度 Iv大于强度阈值, 则所述块区域 具有垂直属性, 否则所述块区域不具有垂直属性。
其中, 所述确定当前帧图像中每个块区域是否为障碍物的步骤包括: 若当前帧及与当前帧最近的之前 N帧所对应的某个相同块区域的运动障 碍物置信度的值等于 1的数量大于第一数量阈值, 且, 当前帧及与当前帧最近 的之前 N帧所对应的所述某个相同块区域的特征障碍物置信度的值等于 1的 数量大于第二数量阈值, 则当前帧图像中该块区域的总置信度。—^^^ 为 1 , 否则该块区域的总置信度 C_Total为 0;
若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。
本发明实施例提供的一种识别图像中障碍物的装置, 包括:
图像分割单元, 用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划分,对于每帧图像均获得若干个划分后 的块区域;
运动障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的运动障碍物置信度,
第一块区域障碍物识别单元,用于根据所述当前帧以及与当前帧最近的之 前 N帧的所述每个块区域的运动障碍物置信度, 依次确定当前帧图像中每个 块区域是否为障碍物;
障碍物确定单元, 用于根据每个块区域确定图像中的障碍物。
其中, 所述运动障碍物置信度计算单元包括:
假想图像生成单元, 用于获取运动参数, 生成当前帧之前 n-k时刻的假想 图像; n >=2, k>=l;
第一运动置信度计算单元,用于对于一个块区域计算当前帧与所述假想图 像的相似程度, 获得第一运动置信度 C_M_A1 ;
第二运动置信度计算单元,用于对于一个块区域计算当前帧与当前帧之前 n-k时刻图像的相似程度, 获得第二运动置信度 C_M_A2;
运动障碍物置信度确定单元, 用于在所述第一运动置信度 C_M_A1 的值 大于第一运动阈值, 且所述第一运动置信度 C_M_A1 与第二运动置信度 C_M_A2的比值大于第二运动阈值时,确定当前帧的该块区域的运动障碍物置 信度 C_M为 1 , 否则确定当前帧的该块区域的运动障碍物置信度 C_M为 0。
其中, 所述第一块区域障碍物识别单元包括:
第一总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值时, 确定当前帧图像中该块区域的总置信度 C_Total为 1 , 否则确定该块区 域的总置信度 C_Total为 0;
第一识别单元, 用于获知当前帧图像中所述某个块区域的总置信度 C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。
本发明实施例提供的另一种识别图像中障碍物的装置, 包括:
图像分割单元, 用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划分,对于每帧图像均获得若干个划分后 的块区域;
运动障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的运动障碍物置信度,
特征障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的特征障碍物置信度;
第二块区域障碍物识别单元,用于根据所述当前帧以及与当前帧最近的之 前 N帧的所述每个块区域的运动障碍物置信度, 和每个块区域的特征障碍物 置信度, 依次确定当前帧图像中每个块区域是否为障碍物;
障碍物确定单元, 用于根据每个块区域确定图像中的障碍物。
其中, 所述运动障碍物置信度计算单元包括:
假想图像生成单元, 用于获取运动参数, 生成当前帧之前 n-k时刻的假想 图像; n >=2, k>=l;
第一运动置信度计算单元,用于对于一个块区域计算当前帧与所述假想图 像的相似程度, 获得第一运动置信度 C_M_A1;
第二运动置信度计算单元,用于对于一个块区域计算当前帧与当前帧之前 n-k时刻图像的相似程度, 获得第二运动置信度 C_M_A2;
运动障碍物置信度确定单元, 用于在所述第一运动置信度 C_M_A1 的值 大于第一运动阈值, 且所述第一运动置信度 C_M_A1 与第二运动置信度 C_M_A2的比值大于第二运动阈值时,确定当前帧的该块区域的运动障碍物置 信度 C_M为 1 , 否则确定当前帧的该块区域的运动障碍物置信度 C_M为 0。
其中, 所述特征障碍物置信度包括基于垂直特性的特征障碍物置信度、基 于垂直边缘特性的特征障碍物置信度, 或者基于纹理特性的特征障碍物置信 度。
其中,在所述特征障碍物置信度基于垂直特性时, 所述特征障碍物置信度 计算单元包括:
垂直属性判断单元, 用于判断当前帧的某一块区域是否具有垂直属性, 将 判断结果通知给特征障碍物置信度确定单元;
特征障碍物置信度确定单元, 用于获知当前帧某块区域具有垂直属性时, 确定当前帧该块区域的特征障碍物置信度为 1 , 否则确定当前帧的该块区域的 特征障碍物置信度 C_F为 0。
其中, 所述第二块区域障碍物识别单元包括:
第二总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值, 且, 当前帧及与当前帧最近的之前 N帧所对应的所述某个相同块区域的 特征障碍物置信度的值等于 1的数量大于第二数量阈值时,确定当前帧图像中 该块区域的总置信度 C_Total为 1 ,否则确定该块区域的总置信度( _11(^1为 0; 第二识别单元, 用于获知当前帧图像中所述某个块区域的总置信度
C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。
应用本发明实施例提供的一种识别图像中障碍物的方法和装置,与传统仅 基于运动补偿方法相比, 不但采用了基于运动补偿的绝对程度的策略,还采用 了基于运动补偿的相对的策略,从而降低了误检测率, 能够准确的判断出图像 中的障碍物。
应用本发明实施例提供的另一种识别图像中障碍物的方法和装置,与传统 仅基于运动补偿方法相比, 不但采用了基于运动补偿的绝对程度的策略, 以及 采用了基于运动补偿的相对的策略,并在此基础上进一步融合了特征分析的策 略, 从而更进一步地降低了误检测率, 能够准确的判断出图像中的障碍物。 附图说明
为了更清楚地说明本发明实施例中的技术方案,下面对实施例中所需要使 用的附图作筒单地介绍, 显而易见地, 下面描述中的附图仅仅是本发明的一些 实 施 例 , 对 于 本 领 域 普 通 技 术 人员来讲, 在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附 图。
图 1是根据本发明实施例的识别图像中障碍物的方法流程图;
图 2 是根据本发明实施例的计算当前帧的每个块区域的运动障碍物置信 度的方法流程图;
图 3 是根据本发明实施例的所设定的世界坐标系和前一时刻图像的摄相 机的世界坐标系的关系示意图;
图 4是根据本发明实施例的当前时刻假想图像的示意图;
图 5是根据本发明实施例的一种识别图像中障碍物的装置结构图; 图 6是根据本发明实施例的另一种识别图像中障碍物的装置结构图。 具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清 楚、 完整地描述, 显然, 所描述的实施例仅仅是本发明一部分实施例, 而不是 全部的实施例。基于本发明中的实施例, 本领域普通技术人员在没有作出创造 性劳动前提下所获得的所有其他实施例, 都属于本发明保护的范围。
本发明实施例提供的一种识别图像中障碍物的方法, 包括: 获取当前帧以 及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划 分,对于每帧图像均获得若干个划分后的块区域; 计算当前帧以及与当前帧最 近的之前 N帧所对应的每个块区域的运动障碍物置信度; 根据所述当前帧以 及与当前帧最近的之前 N帧的所述每个块区域的运动障碍物置信度, 依次确 定当前帧图像中每个块区域是否为障碍物;根据每个块区域确定图像中的障碍 物。 应用本发明, 与传统仅基于运动补偿方法相比, 不但采用了基于运动补偿 的绝对程度的策略,还采用了基于运动补偿的相对的策略,从而降低了误检测 率, 能够准确的判断出图像中的障碍物。
本发明实施例提供的另一种识别图像中障碍物的方法, 包括: 获取当前帧 以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行 划分,对于每帧图像均获得若干个划分后的块区域; 计算当前帧以及与当前帧 最近的之前 N帧所对应的每个块区域的运动障碍物置信度, 和每个块区域的 特征障碍物置信度; 根据所述当前帧以及与当前帧最近的之前 N帧的所述每 个块区域的运动障碍物置信度, 和每个块区域的特征障碍物置信度,依次确定 当前帧图像中每个块区域是否为障碍物; 根据每个块区域确定图像中的障碍 物。 应用本发明, 与传统仅基于运动补偿方法相比, 不但采用了基于运动补偿 的绝对程度的策略,还采用了基于运动补偿的相对的策略, 并在此基础上进一 步融合了特征分析的策略, 从而在降低了误检测率的同时提高了检测的准确 率, 能够准确的判断出图像中的障碍物。
参见图 1 , 其是根据本发明实施例的识别图像中障碍物的方法流程图。 本 例中既采用了基于运动补偿的绝对程度的策略,还采用了基于运动补偿的相对 的策略, 并在此基础上进一步融合了特征分析的策略。 具体包括:
步骤 101 , 划分图像块区域, 具体的, 获取当前帧以及与当前帧最近的之 前 N帧的图像, 对所获取的每帧图像按照相同的方式进行划分, 对于每帧图 像均获得若干个划分后的块区域。
例如, 首先从图像序列中取出当前帧的图像 (即 表示 n时刻的图像) 和之前与当前帧最接近的 N帧图,其中, n>=2;图像序列的帧间隔可以设为 Δ 。 其次, 将上述的每帧图像都分成多个块区域, 块区域的形状可以是矩形, 也可以是三角形等能够将图像完全划分的任何形状, 本实施例为: 都划分为 Ν χ Μ像素的互不重叠的矩形块区域, 这样,对于每帧图像均获得了相同的若干 个划分的矩形块区域。 至于具体的分割方法本文未做限定, 只要能够分割成多 个矩形块区域即可。 上述与当前帧最近的之前 Ν帧图像中的 Ν, 具体数目可根据实际需要确 定, 例如, 可以根据物体移动的速度, 在相机中从出现到消失这段时间内所获 得的帧数作为前述 Ν的具体数目。 步骤 102,计算当前帧以及与当前帧最近的之前 Ν帧所对应的每个块区域 的运动障碍物置信度。 所谓运动障碍物置信度,是指基于运动补偿后的该块区域是否为障碍物的 置信度, 这里的运动补偿与现有的运动补偿稍有不同,现有的运动补偿通常只 包括基于运动补偿的绝对程度,而本发明中的运动补偿包括基于运动补偿的绝 对程度和相对程度。 可以理解, 由于与当前帧最近的之前 N帧所对应的每个块区域的运动障 碍物置信度的计算方法,和当前帧的每个块区域的运动障碍物置信度的计算方 法相同,下面仅对当前帧的每个块区域的运动障碍物置信度的计算方法进行说 明, 对于与当前帧最近的之前 N帧的不再重复说明。 参见图 2, 其是根据本发明实施例的计算当前帧的每个块区域的运动障碍 物置信度的方法流程图。 具体包括: 步骤 201 , 获取相机运动参数。 本发明实施例中所获取的相机运动参数可以由传感器获得,如可以通过速 度传感器和陀螺仪传感器等获取,也可以由序列图像用计算的方法获得,如采 用光流方法、 运动恢复结构 (Structure from Motion, SFM)的方法或基于特征点 检则、直接法等方法估计得到 , ^口文献 1" A robust method for computing vehicle ego-motion (一种计算自车运动的直接方法 ) "采用直接的方法估计运动参数, 文献 2"Transforming camera geometry to a virtual downward-looking camera: robust ego-motion estimation and ground-layer detection"采用改进的直接方法菝 得运动参数。 本发明实施例采用文献 1的直接估计方法获取相机运动参数。 在本发明实施例中设定世界坐标系 [< ; ,H]和前一时刻图像的相机坐 标系 [< '; Χ, ,Ζ ]的原点 G, 重合,坐标轴之间可以有旋转角,世界坐标系的 轴平行于道路平面, 垂直于道路平面。 如图 3所示。
根据道路平面假设, 相机运动参数 m表示为 = ,^^ 即在 方向 有一个平移, 在1 ^方向有一个旋转。 步骤 202,生成当前帧对应之前 n-k时刻的假想图像;其中, n >=2, k>=l ; 所谓的假想图像是指假设前某一时刻(本例中即 n-k时刻)图像中所有像 素点都是道路平面上的点所成的像(其中 相机高度), 相机运动后, 这些 地面上的点在新的相机位置下所成的像, 即为假想图像。 如图 4所示。 下面以 前一时刻图像上某点 P为例, 说明如何生成新的相机位置下的假想图像。 假设前一时刻图像上点尸是道路平面上点 P。所成的像(实际上是立体物上 的点 ρι所成的像),相机运动后, P。所成的像点尸 '即为当前时刻假想图像上与尸 对应的点, 即将点尸的灰度值赋给点尸'。 基于这一原理, 可以得到当前时刻假 想图像上其它像素点与前一时刻图像上点的对应关系,从而可以得到新的相机 位置下的当前时刻的假想图像。 下面说明如何求得假想图像中尸'点的坐标。
以前一时刻图像中某一像素点 c)为例, r, c分别为该点在图像中的行坐 标和列坐标, P是道路平面上点 P0 (Xw , ^ , Zw )的所成的像点。
根据相机成像公式(也被称为摄像机成像公式)可以计算出点 p。的世界坐
^ (Χ Υ , Ζ )。:
Figure imgf000012_0001
Figure imgf000012_0002
其中 ax, , 。,v。是相机内部参数, 可由相机标定获得;
其中 = [Γχ,7 ,Γ/为平移向量, 7 ,7 ,7为相机坐标系原点在世界坐标系下的 置, 为相机外部参数, 在安装相机时获得;
cos γ cos β cos γ sin β α - sin γ cos a cos γ sin β cos + sin ^ sin a 其中 sin γ cos β cos a cos ^ + sin ^ sin β sin a sin γ sin β cos a - cos γ sin a
- sin β cos β sin a cos β cos a 为旋转矩阵, ",Α^分别是相机坐标系绕世界坐标系 U,Z轴的旋转角, 为相机外部参数, 在安装相机时获得;
已知 =相机高度, /^,(;)为像素的坐标;
求解上述方程(1 ), 可得 ^ ,2^和 , Zc是 P的在相机坐标系中 Z轴的坐 标。
利用获得的运动参数 m = ,^^}可得运动后相机坐标系原点在世界坐标 系下的坐标为7 ^ ^'^ +^ , 相机坐标系绕世界坐标系 ^轴的旋转角变为 α,β + ω ,γ。 根据此时相机的旋转矩阵和平移矩阵, 以及已经计算出的 ρ。的世
Figure imgf000013_0001
再由上述相机成像公式(1 ), 方程右端为已知, 可求得左 端未知数, 即点 p。在假象图像中的成像 坐标。 具体的, 基于上述变换, 可将 之前某图像 F ^上所有的点按运动参数运动, 产生了一幅新的假想图像 F„。 如 果不考虑噪声的影响,在平面运动的假设前提下,假想图像中的平面区域与当 前时刻的图像 F„的平面区域是完全一致的, 而非平面区域则不同, 以下将利用 这个原理, 确定区域为立体物的置信度。
步骤 203 , 对于一个块区域计算当前帧与之前 n-k时刻的假想图像的相似 程度, 将所述获得的相似程度作为基于运动补偿的第一运动置信度 C_M_A1。
需要说明的是,本文将当前帧与之前 n-k时刻的假想图像的相似程度称为 第一运动置信度 C_M_A1 , 该第一运动置信度实际是基于运动补偿绝对程度 的; 将当前帧与之前实际的 n-k 时刻图像的相似程度称为第二运动置信度 C_M_A2 , 该第二运动置信度实际是未经运动补偿的; 而第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值实际是基于运动补偿相对程度的。
具体的, 步骤 203包括: 针对图像 中某个块区域中的所有像素, 在假想 图像中寻找坐标相同的像素,然后基于对应点的像素值计算该块区域的当前时 刻图像和当前时刻的假想图像的相似程度,从而获得该块区域为障碍物的第一 运动置信度 C_M_A1 , 其中可以采用归一化相关值(NC )计算所述相似程度 NC:
Figure imgf000013_0002
其中, 块区域的大小为 N*M, 为图像 中的像素( )的灰度值,
X ' ( , j)为图像 的假想图像 Fn中的像素 ( , j)灰度值。
步骤 204, 对于一个块区域计算当前帧与之前 n-k时刻图像的相似程度, 将所述获得的相似程度作为不进行运动补偿的第二运动置信度 C_M_A2; 具体的, 针对图像 中某个块区域中的所有像素, 在 F ^中寻找坐标相同 的像素, 然后基于对应点对的像素值计算该块区域的当前时刻图像和之前 n-k 时刻图像的相似程度, 从而得到该块区域为障碍物的第二运动置信度 C_M_A2 , 其中可以采用归一化相关值( NC )计算所述相似程度 NC:
Figure imgf000014_0001
其中, 块区域的大小为 N*M, 为图像 中的像素( )的灰度值,
X "(i, j、为之前 n-k时刻图像 F:中的像素 (i, j)灰度值。
步骤 205 , 确定运动障碍物置信度 C_M。
具体的, 若所述第一运动置信度 C_M_A1 的值大于第一运动阈值, 且所 述第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值大于第二运动 阈值, 则当前帧的该块区域的运动障碍物置信度 C_M为 1 , 否则当前帧的该 块区域的运动障碍物置信度 C_M为 0。
也就是说, 上述运动障碍物置信度 C_M 是基于运动补偿绝对程度(即 C_M_A1 的值大于第一运动阈值的情况), 与基于运动补偿相对程度 (即 C_M_A1与 C_M_A2的比值大于第二运动阈值的情况)获得的。
其中需要说明的是,第一运动置信度 C_M_A1与第二运动置信度 C_M_A2 的比值是一个广义的概念, 该比值既可以是 C_M_A1/ C_M_A2的值, 也可以 是 (C_M_A1- C_M_A2)/(1- C_M_A2)的值, 还可以是其他表现形式的值。
步骤 206, 重复执行上述步骤 203~205 , 直到计算出每个块区域的运动障 碍物置信度。
步骤 103 ,计算当前帧以及与当前帧最近的之前 N帧所对应的每个块区域 的特征障碍物置信度 C_F。 其中, 所述特征障碍物置信度包括基于垂直特性的特征障碍物置信度、基 于垂直边缘特性的特征障碍物置信度, 或者基于纹理特性的特征障碍物置信 度。 下面以垂直特性为例, 具体说明如何计算特征障碍物置信度 C_F。 步骤 a, 判断当前帧一块区域是否具有垂直属性, 若有则当前帧该块区域 的特征障碍物置信度 C_F为 1 , 否则当前帧的该块区域的特征障碍物置信度 C_F为 0。 具体的判断当前帧一块区域是否具有垂直属性的步骤包括:
aOl )计算一个块区域的垂直方向的强度 /v; 其
Figure imgf000015_0001
第 列像素的灰度值, k为整数, i' j e R , 为该图像块区域, N为图像块区域宽度, M为图像块区域高度。
a02 )若步骤 aOl )所述垂直方向的强度 /v大于强度阈值 T, 即 /ν > Γ , 则 所述块区域具有垂直属性, 否则所述块区域不具有垂直属性。
步骤 b, 重复执行步骤 a, 直到计算出每个块区域的特征障碍物置信度。 步骤 104,根据所述当前帧以及与当前帧最近的之前 N帧的所述每个块区 域的运动障碍物置信度, 和每个块区域的特征障碍物置信度,依次确定当前帧 图像中每个块区域是否为障碍物。 具体的, 取一块区域对应最近 N 帧的运动障碍物置信度, 设为 C_Mt,C_Mt_1,...,C_Mt_N_1, 统计运动障碍物置信度的值等于 1的数量 Sum_M; 取该块区域对应最近 N帧的特征置信度, 设为 C_Ft,C_FM,...,C VN4 , 统 计特征障碍物置信度的值等于 1的数量 Sum_F; 若 Sum_M大于第一数量阈值 S_M_Limit, 且 Sum_F大于第二数量阈值 C_F_Limit , 则总置信度 C_Total为 1 , 否则为 0。 若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。 也就是说, 若当前帧及与当前帧最近的之前 N帧所对应的某个相同块区 域的运动障碍物置信度的值等于 1的数量大于第一数量阈值,且, 当前帧及与 当前帧最近的之前 N帧所对应的所述某个相同块区域的特征障碍物置信度的 值等于 1 的数量大于第二数量阈值, 则当前帧图像中该块区域的总置信度 C_Total为 1 , 否则该块区域的总置信度 C_Total为 0; 在当前帧图像中某个块 区域的总置信度。—^^^ 为 1时, 则该块区域为障碍物, 否则该块区域为非障 碍物。
步骤 105, 根据每个块区域确定图像中的障碍物。
具体的, 当判断出每个块区域是否为障碍物后, 综合所有结果, 即可得到 图像中的障碍物。
可见, 图 1所提供的识别图像中障碍物的方法, 与传统仅基于运动补偿方 法相比, 不但采用了基于运动补偿的绝对程度的策略,还采用了基于运动补偿 的相对的策略, 并在此基础上进一步融合了特征分析的策略,从而在降低了误 检测率的同时提高了检测的准确率, 能够准确的判断出图像中的障碍物。 图 1所示为一较佳实施例, 在实际应用中, 还可以有另一种情况是, 即, 只进行运动补偿(包括绝对程度和相对程度), 而不再基于特征分析进行补偿, 这同样可以达到降低了误检测率的目的。这种实施方式下, 与图 1所示步骤基 本相同, 其区别是, 不再执行步骤 103 , 相应的步骤 104变为: 根据所述当前 帧以及与当前帧最近的之前 N帧的所述每个块区域的运动障碍物置信度, 依 次确定当前帧图像中每个块区域是否为障碍物。其中,确定当前帧图像中每个 块区域是否为障碍物的步骤包括: 具体的, 取一块区域对应最近 N 帧的运动障碍物置信度, 设为 C_Mt,C_MM,...,C_Mt_N4, 统计运动障碍物置信度的值等于 1的数量 Sum_M; 若 Sum_M大于第一数量阈值 S_M_Limit, 则总置信度 C_Total为 1 , 否则为 0。 若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。 也就是说, 若当前帧及与当前帧最近的之前 Ν帧所对应的某个相同块区 域的运动障碍物置信度的值等于 1的数量大于第一数量阈值,则当前帧图像中 该块区域的总置信度 C_Total为 1 , 否则该块区域的总置信度 C_Total为 0; 在 当前帧图像中某个块区域的总置信度 _11(^1为 1时, 则该块区域为障碍物, 否则该块区域为非障碍物。 本发明实施例还提供了一种识别图像中障碍物的装置, 参见图 5, 其与图
1所示实施例对应,包括:图像分割单元 501、运动障碍物置信度计算单元 502、 特征障碍物置信度计算单元 503、 第二块区域障碍物识别单元 504和障碍物确 定单元 505。 其中,
图像分割单元 501 ,用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同的方式进行划分,对于每帧图像均获得若干个划分 后的块区域;
运动障碍物置信度计算单元 502, 用于计算当前帧以及与当前帧最近的之 前 N帧所对应的每个块区域的运动障碍物置信度,
特征障碍物置信度计算单元 503, 用于计算当前帧以及与当前帧最近的之 前 N帧所对应的每个块区域的特征障碍物置信度; 该特征障碍物置信度可以 包括基于垂直特性的特征障碍物置信度、基于垂直边缘特性的特征障碍物置信 度, 或者基于纹理特性的特征障碍物置信度。
第二块区域障碍物识别单元 504, 用于根据所述当前帧以及与当前帧最近 的之前 N帧的所述每个块区域的运动障碍物置信度, 和每个块区域的特征障 碍物置信度, 依次确定当前帧图像中每个块区域是否为障碍物;
障碍物确定单元 505 , 用于根据每个块区域确定图像中的障碍物。
上述运动障碍物置信度计算单元 502还可以包括: 假想图像生成单元、第 一运动置信度计算单元、第二运动置信度计算单元以及运动障碍物置信度确定 单元。 其中,
假想图像生成单元, 用于获取运动参数, 生成当前帧之前 n-k时刻的假想 图像; n >=2, k>=l;
第一运动置信度计算单元,用于对于一个块区域计算当前帧与所述假想图 像的相似程度, 获得第一运动置信度 C_M_A1 ;
第二运动置信度计算单元,用于对于一个块区域计算当前帧与当前帧之前 n-k时刻图像的相似程度, 获得第二运动置信度 C_M_A2;
运动障碍物置信度确定单元, 用于在所述第一运动置信度 C_M_A1 的值 大于第一运动阈值, 且所述第一运动置信度 C_M_A1 与第二运动置信度 C_M_A2的比值大于第二运动阈值时,确定当前帧的该块区域的运动障碍物置 信度 C_M为 1 , 否则确定当前帧的该块区域的运动障碍物置信度 C_M为 0; 当上述特征障碍物置信度基于垂直特性时,所述特征障碍物置信度计算单 元 503可以包括: 垂直属性判断单元和特征障碍物置信度确定单元。 其中, 垂直属性判断单元, 用于判断当前帧的某一块区域是否具有垂直属性, 将 判断结果通知给特征障碍物置信度确定单元;
特征障碍物置信度确定单元, 用于获知当前帧某块区域具有垂直属性时, 确定当前帧该块区域的特征障碍物置信度为 1 , 否则确定当前帧的该块区域的 特征障碍物置信度 C_F为 0。
上述第二块区域障碍物识别单元 504可以包括:第二总置信度确定单元和 第二识别单元, 其中,
第二总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值, 且, 当前帧及与当前帧最近的之前 N帧所对应的所述某个相同块区域的 特征障碍物置信度的值等于 1的数量大于第二数量阈值时,确定当前帧图像中 该块区域的总置信度 C_Total为 1 ,否则确定该块区域的总置信度( _11(^1为 0; 第二识别单元, 用于获知当前帧图像中所述某个块区域的总置信度 C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。 本发明实施例还提供了一种识别图像中障碍物的装置, 参见图 6, 其与图 5所示实施例的区别是, 不包括特征障碍物置信度计算单元 503, 也即, 图 6 所示装置包括: 图像分割单元 601、 运动障碍物置信度计算单元 602、 第一块 区域障碍物识别单元 603和障碍物确定单元 604。 其中,
图像分割单元 601 ,用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划分,对于每帧图像均获得若干个划分后 的块区域;
运动障碍物置信度计算单元 602, 用于计算当前帧以及与当前帧最近的之 前 N帧所对应的每个块区域的运动障碍物置信度,
第一块区域障碍物识别单元 603, 用于根据所述当前帧以及与当前帧最近 的之前 N帧的所述每个块区域的运动障碍物置信度, 依次确定当前帧图像中 每个块区域是否为障碍物;
障碍物确定单元 604, 用于根据每个块区域确定图像中的障碍物。
上述运动障碍物置信度计算单元 602与图 5中的运动障碍物置信度计算单 元 502相同, 包括: 假想图像生成单元、 第一运动置信度计算单元、 第二运动 置信度计算单元以及运动障碍物置信度确定单元, 具体不再赘述。
上述第一块区域障碍物识别单元 603可以包括:
第一总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值时, 确定当前帧图像中该块区域的总置信度 C_Total为 1 , 否则确定该块区 域的总置信度 C_Total为 0;
第一识别单元, 用于获知当前帧图像中所述某个块区域的总置信度 C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。 应用本发明所提供的识别图像中障碍物的装置,与传统仅基于运动补偿的 逻辑装置相比, 不但采用了基于运动补偿的绝对程度的策略,还采用了基于运 动补偿的相对的策略, 并在此基础上进一步融合了特征分析的策略,从而在降 低了误检测率的同时提高了检测的准确率, 能够准确的判断出图像中的障碍 物。 本领域普通技术人员可以理解实现上述方法实施方式中的全部或部分步 骤是可以通过程序来指令相关的硬件来完成,所述的程序可以存储于计算机可 读取存储介质中, 这里所称得的存储介质, 如: ROM/RAM、 磁碟、 光盘等。
以上所述仅为本发明的较佳实施例而已, 并非用于限定本发明的保护范 围。 凡在本发明的精神和原则之内所作的任何修改、 等同替换、 改进等, 均包 含在本发明的保护范围内。

Claims

权 利 要 求
1、 一种识别图像中障碍物的方法, 其特征在于, 包括:
获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像 按相同方式进行划分, 对于每帧图像均获得若干个划分后的块区域;
计算当前帧以及与当前帧最近的之前 N帧所对应的每个块区域的运动障 碍物置信度;
根据所述当前帧以及与当前帧最近的之前 N帧的所述每个块区域的运动 障碍物置信度, 依次确定当前帧图像中每个块区域是否为障碍物;
根据每个块区域确定图像中的障碍物。
2、 根据权利要求 1所述的方法, 其特征在于, 所述计算当前帧的每个块 区域的运动障碍物置信度的步骤包括:
i )获取运动参数,生成当前帧对应之前 n-k时刻的假想图像; n >=2, k>=l ; ii )对于一个块区域, 计算该块的当前帧与所述假想图像的相似程度, 获 得第一运动置信度 C_M_A1;
iii )对于一个块区域, 计算该块的当前帧与当前帧之前 n-k时刻图像的相 似程度, 获得第二运动置信度 C_M_A2;
iv )对于一个块区域,若所述第一运动置信度 C_M_A1的值大于第一运动 阈值, 且所述第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值大 于第二运动阈值, 则当前帧的该块区域的运动障碍物置信度 C_M为 1 , 否则 当前帧的该块区域的运动障碍物置信度 C_M为 0;
V )重复执行步骤 ii )至 iv ),直到计算出每个块区域的运动障碍物置信度。
3、 根据权利要求 2所述的方法, 其特征在于, 所述确定当前帧图像中每 个块区域是否为障碍物的步骤包括:
若当前帧及与当前帧最近的之前 N帧所对应的某个相同块区域的运动障 碍物置信度的值等于 1的数量大于第一数量阈值,则当前帧图像中该块区域的 总置信度 _11(^1为 1 , 否则该块区域的总置信度 C_Total为 0;
若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。
4、 一种识别图像中障碍物的方法, 其特征在于, 包括:
获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像 按相同方式进行划分, 对于每帧图像均获得若干个划分后的块区域;
计算当前帧以及与当前帧最近的之前 N帧所对应的每个块区域的运动障 碍物置信度, 和每个块区域的特征障碍物置信度;
根据所述当前帧以及与当前帧最近的之前 N帧的所述每个块区域的运动 障碍物置信度, 和每个块区域的特征障碍物置信度,依次确定当前帧图像中每 个块区域是否为障碍物;
根据每个块区域确定图像中的障碍物。
5、 根据权利要求 4所述的方法, 其特征在于, 所述计算当前帧的每个块 区域的运动障碍物置信度的步骤包括:
i )获取运动参数,生成当前帧对应之前 n-k时刻的假想图像; n >=2, k>=l; ii )对于一个块区域, 计算该块的当前帧与所述假想图像的相似程度, 获 得第一运动置信度 C_M_A1;
iii )对于一个块区域, 计算该块的当前帧与当前帧之前 n-k时刻图像的相 似程度, 获得第二运动置信度 C_M_A2;
iv )对于一个块区域,若所述第一运动置信度 C_M_A1的值大于第一运动 阈值, 且所述第一运动置信度 C_M_A1与第二运动置信度 C_M_A2的比值大 于第二运动阈值, 则当前帧的该块区域的运动障碍物置信度 C_M为 1 , 否则 当前帧的该块区域的运动障碍物置信度 C_M为 0;
V )重复执行步骤 ii )至 iv ),直到计算出每个块区域的运动障碍物置信度。
6、 根据权利要求 4所述的方法, 其特征在于, 所述特征障碍物置信度包 括基于垂直特性的特征障碍物置信度、 或者基于纹理特性的特征障碍物置信 度。
7、 根据权利要求 6所述的方法, 其特征在于, 所述根据垂直特性计算当 前帧的每个块区域的特征障碍物置信度的步骤包括:
a )判断当前帧一块区域是否具有垂直属性, 若有则当前帧该块区域的特 征障碍物置信度为 1 , 否则当前帧的该块区域的特征障碍物置信度 C_F为 0; b )重复执行步骤 a ), 直到计算出每个块区域的特征障碍物置信度。
8、 根据权利要求 7所述的方法, 其特征在于, 所述判断当前帧一块区域 是否具有垂直属性的步骤包括:
aOl )计算一个块区域的垂直方向的强度 /v; 其
Figure imgf000023_0001
第 列像素的灰度值, k为整数, i' j e R , 为该图像块区域, N为图像宽度, M为图像高度。
a02 )若步骤 aOl )所述垂直方向的强度 /v大于强度阈值, 则所述块区域 具有垂直属性, 否则所述块区域不具有垂直属性。
9、 根据权利要求 7所述的方法, 其特征在于, 所述确定当前帧图像中每 个块区域是否为障碍物的步骤包括:
若当前帧及与当前帧最近的之前 N帧所对应的某个相同块区域的运动障 碍物置信度的值等于 1的数量大于第一数量阈值, 且, 当前帧及与当前帧最近 的之前 N帧所对应的所述某个相同块区域的特征障碍物置信度的值等于 1的 数量大于第二数量阈值, 则当前帧图像中该块区域的总置信度。—^^^ 为 1 , 否则该块区域的总置信度 C_Total为 0;
若当前帧图像中所述某个块区域的总置信度( _11(^1为 1 , 则该块区域为 障碍物, 否则该块区域为非障碍物。
10、 一种识别图像中障碍物的装置, 其特征在于, 包括:
图像分割单元, 用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划分,对于每帧图像均获得若干个划分后 的块区域;
运动障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的运动障碍物置信度,
第一块区域障碍物识别单元,用于根据所述当前帧以及与当前帧最近的之 前 N帧的所述每个块区域的运动障碍物置信度, 依次确定当前帧图像中每个 块区域是否为障碍物;
障碍物确定单元, 用于根据每个块区域确定图像中的障碍物。
11、 根据权利要求 10所述的装置, 其特征在于, 所述运动障碍物置信度 计算单元包括: 假想图像生成单元, 用于获取运动参数, 生成当前帧对应之前 n-k时刻的 i想图像; n >=2, k>=l;
第一运动置信度计算单元,用于对于一个块区域计算当前帧与所述假想图 像的相似程度, 获得第一运动置信度 C_M_A1;
第二运动置信度计算单元, 对于一个块区域计算当前帧与当前帧之前 n-k 时刻图像的相似程度, 获得第二运动置信度 C_M_A2;
运动障碍物置信度确定单元, 用于在所述第一运动置信度 C_M_A1 的值 大于第一运动阈值, 且所述第一运动置信度 C_M_A1 与第二运动置信度 C_M_A2的比值大于第二运动阈值时,确定当前帧的该块区域的运动障碍物置 信度 C_M为 1 , 否则确定当前帧的该块区域的运动障碍物置信度 C_M为 0。
12、 根据权利要求 10所述的装置, 其特征在于, 所述第一块区域障碍物 识别单元包括:
第一总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值时, 确定当前帧图像中该块区域的总置信度 C_Total为 1 , 否则确定该块区 域的总置信度 C_Total为 0;
第一识别单元, 用于获知当前帧图像中所述某个块区域的总置信度 C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。
13、 一种识别图像中障碍物的装置, 其特征在于, 包括:
图像分割单元, 用于获取当前帧以及与当前帧最近的之前 N帧的图像, 对所获取的每帧图像按相同方式进行划分,对于每帧图像均获得若干个划分后 的块区域;
运动障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的运动障碍物置信度,
特征障碍物置信度计算单元,用于计算当前帧以及与当前帧最近的之前 N 帧所对应的每个块区域的特征障碍物置信度;
第二块区域障碍物识别单元,用于根据所述当前帧以及与当前帧最近的之 前 N帧的所述每个块区域的运动障碍物置信度, 和每个块区域的特征障碍物 置信度, 依次确定当前帧图像中每个块区域是否为障碍物; 障碍物确定单元, 用于根据每个块区域确定图像中的障碍物。
14、 根据权利要求 13所述的装置, 其特征在于, 所述运动障碍物置信度 计算单元包括:
假想图像生成单元, 用于获取运动参数, 生成当前帧对应之前 n-k时刻的 i想图像; n >=2, k>=l;
第一运动置信度计算单元,用于对于一个块区域计算当前帧与所述假想图 像的相似程度, 获得第一运动置信度 C_M_A1;
第二运动置信度计算单元,用于对于一个块区域计算当前帧与当前帧之前 n-k时刻图像的相似程度, 获得第二运动置信度 C_M_A2;
运动障碍物置信度确定单元, 用于在所述第一运动置信度 C_M_A1 的值 大于第一运动阈值, 且所述第一运动置信度 C_M_A1 与第二运动置信度 C_M_A2的比值大于第二运动阈值时,确定当前帧的该块区域的运动障碍物置 信度 C_M为 1 , 否则确定当前帧的该块区域的运动障碍物置信度 C_M为 0。
15、 根据权利要求 13所述的装置, 其特征在于, 所述特征障碍物置信度 包括基于垂直特性的特征障碍物置信度、基于垂直边缘特性的特征障碍物置信 度, 或者基于纹理特性的特征障碍物置信度。
16、 根据权利要求 15所述的方法, 其特征在于, 在所述特征障碍物置信 度基于垂直特性时, 所述特征障碍物置信度计算单元包括:
垂直属性判断单元, 用于判断当前帧的某一块区域是否具有垂直属性, 将 判断结果通知给特征障碍物置信度确定单元;
特征障碍物置信度确定单元, 用于获知当前帧某块区域具有垂直属性时, 确定当前帧该块区域的特征障碍物置信度为 1 , 否则确定当前帧的该块区域的 特征障碍物置信度 C_F为 0。
17、 根据权利要求 13所述的装置, 其特征在于, 所述第二块区域障碍物 识别单元包括:
第二总置信度确定单元, 用于在当前帧及与当前帧最近的之前 N帧所对 应的某个相同块区域的运动障碍物置信度的值等于 1 的数量大于第一数量阈 值, 且, 当前帧及与当前帧最近的之前 N帧所对应的所述某个相同块区域的 特征障碍物置信度的值等于 1的数量大于第二数量阈值时,确定当前帧图像中 该块区域的总置信度 C_Total为 1 ,否则确定该块区域的总置信度( _11(^1为 0; 第二识别单元, 用于获知当前帧图像中所述某个块区域的总置信度 C_Total为 1时, 确定该块区域为障碍物, 否则确定该块区域为非障碍物。
PCT/CN2009/071577 2008-12-15 2009-04-30 识别图像中障碍物的方法和装置 WO2010069167A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2011539877A JP5276721B2 (ja) 2008-12-15 2009-04-30 画像中の障害物を識別する方法および装置
US13/133,546 US8406474B2 (en) 2008-12-15 2009-04-30 Method and apparatus for identifying obstacle in image
DE112009003144.7T DE112009003144B4 (de) 2008-12-15 2009-04-30 Verfahren und Vorrichtung zum Feststellen eines Hindernisses in einem Bild

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2008101858334A CN101419667B (zh) 2008-12-15 2008-12-15 识别图像中障碍物的方法和装置
CN200810185833.4 2008-12-15

Publications (1)

Publication Number Publication Date
WO2010069167A1 true WO2010069167A1 (zh) 2010-06-24

Family

ID=40630453

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2009/071577 WO2010069167A1 (zh) 2008-12-15 2009-04-30 识别图像中障碍物的方法和装置

Country Status (5)

Country Link
US (1) US8406474B2 (zh)
JP (1) JP5276721B2 (zh)
CN (1) CN101419667B (zh)
DE (1) DE112009003144B4 (zh)
WO (1) WO2010069167A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2431917A1 (en) * 2010-09-21 2012-03-21 Mobileye Technologies Limited Barrier and guardrail detection using a single camera
US9959595B2 (en) 2010-09-21 2018-05-01 Mobileye Vision Technologies Ltd. Dense structure from motion

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101419667B (zh) 2008-12-15 2010-12-08 东软集团股份有限公司 识别图像中障碍物的方法和装置
US9129523B2 (en) * 2013-05-22 2015-09-08 Jaybridge Robotics, Inc. Method and system for obstacle detection for vehicles using planar sensor data
US9118816B2 (en) 2011-12-06 2015-08-25 Mobileye Vision Technologies Ltd. Road vertical contour detection
US9297641B2 (en) 2011-12-12 2016-03-29 Mobileye Vision Technologies Ltd. Detection of obstacles at night by analysis of shadows
US9256791B2 (en) 2012-12-04 2016-02-09 Mobileye Vision Technologies Ltd. Road vertical contour detection
CN103413308B (zh) * 2013-08-01 2016-07-06 东软集团股份有限公司 一种障碍物检测方法和装置
CN104424648B (zh) * 2013-08-20 2018-07-24 株式会社理光 对象跟踪方法和设备
CN104299244B (zh) * 2014-09-26 2017-07-25 东软集团股份有限公司 基于单目相机的障碍物检测方法及装置
US10635909B2 (en) * 2015-12-30 2020-04-28 Texas Instruments Incorporated Vehicle control with efficient iterative triangulation
WO2017147747A1 (en) * 2016-02-29 2017-09-08 SZ DJI Technology Co., Ltd. Obstacle avoidance during target tracking
US10163015B2 (en) * 2016-11-16 2018-12-25 Ford Global Technologies, Llc Detecting foliage using range data
CN107049717B (zh) * 2016-11-23 2019-03-01 杭州视氪科技有限公司 一种基于rgb-d相机和立体声的视障人士斑马线检测系统及方法
WO2020024234A1 (zh) * 2018-08-02 2020-02-06 深圳前海达闼云端智能科技有限公司 路径导航方法、相关装置及计算机可读存储介质
CN109658450B (zh) * 2018-12-17 2020-10-13 武汉天乾科技有限责任公司 一种基于无人机的快速正射影像生成方法
CN110135278B (zh) * 2019-04-23 2023-05-12 腾讯科技(上海)有限公司 一种障碍物检测方法、装置及电子设备
CN110674705B (zh) * 2019-09-05 2022-11-29 北京智行者科技股份有限公司 基于多线激光雷达的小型障碍物检测方法及装置
CN110765970B (zh) * 2019-10-31 2022-08-09 北京地平线机器人技术研发有限公司 最近障碍物的确定方法和装置、存储介质、电子设备
CN111231982B (zh) * 2020-01-08 2021-05-04 中国第一汽车股份有限公司 一种智能驾驶的障碍物识别方法、装置、车辆和存储介质
CN113771846B (zh) * 2020-10-28 2023-08-04 北京京东乾石科技有限公司 用于车辆的自动控制方法、装置、系统和介质
CN113077657B (zh) * 2021-03-30 2022-07-05 上海华兴数字科技有限公司 车辆间安全距离报警方法及装置
US11624831B2 (en) 2021-06-09 2023-04-11 Suteng Innovation Technology Co., Ltd. Obstacle detection method and apparatus and storage medium
CN114488072A (zh) * 2021-06-09 2022-05-13 深圳市速腾聚创科技有限公司 障碍物检测方法、装置及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1688157A (zh) * 2005-06-13 2005-10-26 北京中星微电子有限公司 一种运动检测方法
CN101123681A (zh) * 2007-09-20 2008-02-13 宝利微电子系统控股公司 一种数字图像降噪方法及装置
CN101179725A (zh) * 2007-12-12 2008-05-14 北京中星微电子有限公司 一种运动检测方法与装置
CN101419667A (zh) * 2008-12-15 2009-04-29 东软集团股份有限公司 识别图像中障碍物的方法和装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH10222679A (ja) * 1997-02-04 1998-08-21 Toyota Motor Corp 車両用画像処理装置
JP4004653B2 (ja) * 1998-08-03 2007-11-07 カスタム・テクノロジー株式会社 動きベクトル検出方法および装置、記録媒体
US7446798B2 (en) * 2003-02-05 2008-11-04 Siemens Corporate Research, Inc. Real-time obstacle detection with a calibrated camera and known ego-motion
DE102006001033A1 (de) 2005-03-03 2006-09-07 Volkswagen Ag Vorrichtung und Verfahren zur Erkennung eines Objektes für ein Kraftfahrzeug

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1688157A (zh) * 2005-06-13 2005-10-26 北京中星微电子有限公司 一种运动检测方法
CN101123681A (zh) * 2007-09-20 2008-02-13 宝利微电子系统控股公司 一种数字图像降噪方法及装置
CN101179725A (zh) * 2007-12-12 2008-05-14 北京中星微电子有限公司 一种运动检测方法与装置
CN101419667A (zh) * 2008-12-15 2009-04-29 东软集团股份有限公司 识别图像中障碍物的方法和装置

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2431917A1 (en) * 2010-09-21 2012-03-21 Mobileye Technologies Limited Barrier and guardrail detection using a single camera
US9280711B2 (en) 2010-09-21 2016-03-08 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
EP3301612A1 (en) * 2010-09-21 2018-04-04 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US9959595B2 (en) 2010-09-21 2018-05-01 Mobileye Vision Technologies Ltd. Dense structure from motion
US10078788B2 (en) 2010-09-21 2018-09-18 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US10115027B2 (en) 2010-09-21 2018-10-30 Mibileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US10445595B2 (en) 2010-09-21 2019-10-15 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US10685424B2 (en) 2010-09-21 2020-06-16 Mobileye Vision Technologies Ltd. Dense structure from motion
EP3751457A1 (en) * 2010-09-21 2020-12-16 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US11087148B2 (en) 2010-09-21 2021-08-10 Mobileye Vision Technologies Ltd. Barrier and guardrail detection using a single camera
US11170466B2 (en) 2010-09-21 2021-11-09 Mobileye Vision Technologies Ltd. Dense structure from motion

Also Published As

Publication number Publication date
CN101419667B (zh) 2010-12-08
JP2012512446A (ja) 2012-05-31
DE112009003144B4 (de) 2019-05-29
CN101419667A (zh) 2009-04-29
DE112009003144T5 (de) 2012-07-05
JP5276721B2 (ja) 2013-08-28
US20110262009A1 (en) 2011-10-27
US8406474B2 (en) 2013-03-26

Similar Documents

Publication Publication Date Title
WO2010069167A1 (zh) 识别图像中障碍物的方法和装置
WO2021147546A1 (zh) 多传感器融合slam系统、多传感器融合方法及机器人、介质
WO2021196294A1 (zh) 一种跨视频人员定位追踪方法、系统及设备
Zaarane et al. Distance measurement system for autonomous vehicles using stereo camera
US9420265B2 (en) Tracking poses of 3D camera using points and planes
KR101184097B1 (ko) 얼굴 정면포즈 판단 방법
WO2010060287A1 (zh) 一种基于单目视觉的障碍物检测方法及装置
JP2016029564A (ja) 対象検出方法及び対象検出装置
JPWO2015029982A1 (ja) 画像処理装置、画像処理方法、及びプログラム
JP2007263669A (ja) 3次元座標取得装置
BR112016010089A2 (pt) dispositivo de estimação de posição de corpo em movimento e método de estimação de posição de corpo em movimento
WO2015014111A1 (zh) 一种光流跟踪方法和装置
US9727776B2 (en) Object orientation estimation
WO2023016271A1 (zh) 位姿确定方法、电子设备及可读存储介质
JP2020042772A (ja) 深度マップに対して画像位置合せを行って深度データを最適化することができる深度データ処理システム
JP7173471B2 (ja) 3次元位置推定装置及びプログラム
WO2021000390A1 (zh) 一种点云融合方法、装置、电子设备和计算机存储介质
JP3786618B2 (ja) 画像処理装置及びその方法
WO2010066123A1 (zh) 障碍物分割方法及装置
CN111739144A (zh) 一种基于深度特征光流的同时定位和建图的方法及装置
WO2023016182A1 (zh) 位姿确定方法、装置、电子设备和可读存储介质
CN112017212A (zh) 人脸关键点跟踪模型的训练、跟踪方法及系统
JP4850768B2 (ja) 3次元の人の顔の表面データを再構築するための装置及びプログラム
WO2022012425A1 (zh) 目标检测方法、装置及电子设备
CN110009683B (zh) 基于MaskRCNN的实时平面上物体检测方法

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: 09832849

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 13133546

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 2011539877

Country of ref document: JP

WWE Wipo information: entry into national phase

Ref document number: 1120090031447

Country of ref document: DE

Ref document number: 112009003144

Country of ref document: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 10/10/11)

122 Ep: pct application non-entry in european phase

Ref document number: 09832849

Country of ref document: EP

Kind code of ref document: A1