[go: up one dir, main page]

WO2014165787A1 - Système et procédé de détection de structures - Google Patents

Système et procédé de détection de structures Download PDF

Info

Publication number
WO2014165787A1
WO2014165787A1 PCT/US2014/033017 US2014033017W WO2014165787A1 WO 2014165787 A1 WO2014165787 A1 WO 2014165787A1 US 2014033017 W US2014033017 W US 2014033017W WO 2014165787 A1 WO2014165787 A1 WO 2014165787A1
Authority
WO
WIPO (PCT)
Prior art keywords
shadow
image
image data
contour
threshold
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/US2014/033017
Other languages
English (en)
Inventor
Mohamed I. ELBAKARY
Khan M. Iftekharuddin
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.)
Old Dominion University
Original Assignee
Old Dominion 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 Old Dominion University filed Critical Old Dominion University
Publication of WO2014165787A1 publication Critical patent/WO2014165787A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/149—Segmentation; Edge detection involving deformable models, e.g. active contour models
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20116—Active contour; Active surface; Snakes
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30184—Infrastructure

Definitions

  • the present invention relates to a system and method for the detection of structures in imagery.
  • Satellite and aerial images are a common method to obtain information about objects on the Earth's sur&ce.
  • the detection of objects and targets within aerial images is of great interest for many applications, such as rescue operations and defense applications.
  • a human would be used to interpret an aerial image to identify objects such as buildings.
  • the quantity of data and the increase in applications has rendered human interpretation often expensive or impractical,
  • buildings For the purpose of most of the applications of interest buildings have simple geometry. This can reduce the likelihood of inter-building occlusion in aerial images. However, building detection in an aerial image is generally difficult because aerial imagery
  • i may contain a large number of objects other than buildings within a scene, such as vegetation, bodies of water, roads, etc., as well as the potential for similarity of imaged roofs to a background.
  • objects other than buildings within a scene such as vegetation, bodies of water, roads, etc.
  • the existence of such clutter makes the problem of detecting buildings more difficult than the detection of objects in indoor environments.
  • a structure of interest should be detected and segmented from the background and then should be represented for interpretation later.
  • a first step of focus includes the detection and isolation, This step is more difficult than the second step because the presence of natural texture of vegetation, water occupied areas, and other varieties of detail present on or near the objects of interest in the aerial images,
  • Another conventional approach employed a probability model, with the concept that spatial context constraints could increase classification accuracy.
  • height information is not available: however, multiple images may be used to infer height. Height information may be fused to spatial information to improve the efficiency of detection algorithms.
  • Polygons have also been employed to fit and to extract buildings.
  • the comers and the edges for families of polygons are matched to the set of corners and edges that may be extracted from images.
  • morphological filters for building detection such as white top-hat, black top-hat, or other geometric filters.
  • Some approaches may involve shadow as an aspect of elevated man- made structures; however, these approaches require information about the direction of the sun within the image, as well as its angular elevation.
  • Other approaches have proposed using roof color as an aspect in improving building detection methods.
  • Geometric active contour models have gained significant attention for potential in image processing and computer vision applications, such as segmentation and detection in aerial and satellite images, The inventors have been able to detect, in high-resolution panchromatic satellite images, man-made objects in two stages for a level-set algorithm.
  • the present geometric active contour model is represented by the zero-level set of the higher dimensional function in the level-set framework. These functions are able to detect the boundary of regions based on the homogeneity of local features, such as the intensity, without depending on the edges of the regions as motivation force,
  • Embodiments of the present approach involve adopting and tailoring a geometric active contour method for shadow detection of man-made structures such as buildings in high-resolution panchromatic satellite images.
  • Statistical information and prior shape information of the shadows is not universally available, as sn the ease of man-made structures, since the shadow areas are changeable during the day based on the position of the sun in the sky and the geometry of the structures that, generate the associated shadow areas.
  • the input images are the kind of the gray scale and the color information and other bands information are missing,
  • a geometric active contour may be used to solve that problem by including an additional term in the energy equation to enforce or systematically bias the contours towards the boundary of the dark regions, including the shadow regions in the input image.
  • the additional terra is based on encoding the radiometric characteristics of the dark regions relatively to the neighbors which is considered as a cue for existing of the shadows as explained in greater detail below. There is no need for the manual insertion of initial position of the contours, making the present approach automatic.
  • Figure 2 illustrates an embodiment of global thresholding.
  • Figure 3 include examples of regular and irregular boundaries with the boundary contour metric, with the upper row of examples showing regular boundaries and the lower row showing irregular boundaries,
  • Figure 4 is an embodiment, of a procedure for a geometric filter of the present approach.
  • Figure 5 is an embodiment of local processing.
  • Figure 6 is a flow chart of a method embodiment.
  • Figures 7(a) is an input sateiiite image
  • Figure 7(b) is an initial contour generated from the input image
  • Figure 7(c) is the ground truth of the shadow for man-made buildings
  • Figure 7(d) is the evolution of the geometric active contours after 20 iterations.
  • Figure 9 is an image processed by the present approach.
  • Figures 10(a)-(g) are (a) an input image, (b) a ground truth image with the man- made buildings, (c) result of segmentation by equation (10), (d) result of using Otsu's threshold with global processing, (e) result of using the geometric filter, (f) result of using Otsu's threshold with local processing, and (g) results of using the geometric filter.
  • Figures 1 1 (a)-(c) are (a) an input image, (b) a ground truth image, and (c) an image processed by the present approach.
  • Figures 12(a)-(d) are (a) an input image, (b) a ground truth image, and (c) an image processed by the present approach.
  • Figures 13(a)-(e) show different initializations for the level-set function of the present segmentation
  • Figure 14 is a schematic of a system embodiment.
  • Figure 15 is a schematic of a system embodiment.
  • Disclosed is a system and method, or overall approach, for the detection of structures involving the use of shadows depicted in high-resolution, overhead digital panchromatic imagery to determine information about the imaged objects, such as optical structure detection and recognition.
  • the approach for detection of structures involves the acquisition of a pixelated panchromatic aerial image of interest; the detection and isolation of building shadow area of each building within the image; the search of a shadow area for the pixel having a minimum value; identifying at least one neighbor pixel to the pixel having a minimum value: and application of a segmentation algorithm to determine a foot print of the building.
  • Pixelated digital images generally have a numeric value of intensity for each pixel or picture element.
  • a pixel is generally the smallest element of sampled luminance within the image.
  • each pixel may have a sample of intensity luminance value ranging from a minimum to a maximum; for an 8-bit example, a value of a minimum may be 0, with black representing a weakest intensity, and a value of a maximum may be 255, with white representing a strongest intensity, intermediate values may comprise a shade of gray,
  • the word “optical” may refer to methods, systems, objects, or other matters relating to electromagnetic radiation
  • the use of the word “light” may generally refer to electromagnetic radiation, which includes but is not limited to visible radiation, By minimum value, an absolute darkest pixel not needed if the darkest or minimal pixels generally consistently reveal a structure; in this way, occlusion is not a problem
  • An image may be acquired by use of a configured imaging system, device, radiometer, or sensor operable to capture optica! image data representative of a scene of interest, and further configured to produce a pixelated, gray scale digital image.
  • Imaging sensors may use or comprise, for example, digitizing or digital cameras, radiometers, charge coupled devices, laser scanners, CMOS image sensors, optical laser sensors or scanners, etc.
  • a first step may be acquiring a suitable input image, such as a digital pixelated gray scale aerial image of interest of sufficiently high resolution.
  • the first step may be undertaken, for example, with a configured optical imaging device or system in communication with a non-transitory memory.
  • a second step may be the detecting and isolating of shadow of each building or structure of interest within the input image.
  • the detecting and isolating may be by use of active contour models, including with a zero-level set implementation (i.e., geometric active contours class) to find the contours of structures in the image, possibly executed by a processor based control system. Localizing an object with geometric active contours may be feasible regardless the complexity of pattern of the structure.
  • Geometric active contours can localize shadows beside other structures in the image as closed contours objects.
  • Image contents may be mapped to objects defined by the closed contour approach.
  • the shadow objects may then be detected and isolated from the contents of the image tor further processing, i.e., such as finding the building that generated the shadow,
  • Active contour or snake models are energy-minimizing curves/surfaces that move in the image domain using image features, in order to accurately localize object contours.
  • the movement of the snakes is guided and influenced by external and internal forces, such that the models reach a minimal energy by segmenting the object.
  • active-contour models are characterized by a set of parameters and the evolution of the contours are performed on a predefined set of control points in the spatial domain of the image.
  • One of the major drawbacks of the parametric active contours is that the methods are greatly influenced by the initial conditions of the contours in the spatial domain close enough to the desired feature. Otherwise the contours converge to undesired objects in the image.
  • An additional drawback is thai usually each contour captures only one object in the scene.
  • Geometric active contours are represented by the zero-level set of a higher dimensional surface, such that the updating of the surface function is performed in the entire image domain.
  • Geometric active, contours may be either edge- based methods or region-based methods. In edge-based methods, the gradient of the image is employed as an attraction force to attract the contour to the edges of the objects in the image.
  • the region- based methods empioy the region features such as the gray -level intensity, and other pixels statistics that reflect the homogeneity of spatially localized regions as an attraction force to these objects.
  • the present approach offers improvements in results and low sensitivity to noise.
  • geometric active contours implicitly handle the topological changes and are not characterized by a set of parameters,
  • the image domain - in a simple case ⁇ consists of a background and foreground (i.e., two) regions, which are characterized by homogeneity of the gray level.
  • the gray levels in these two regions are approximately pieeewise-eonstani intensities of different values u 0 and u £ -, and C is the boundary or curve between the two regions.
  • This geometric active contour method may- obtain good results whether the boundaries between regions are well distinguished or not. Boundaries that are not well distinguished between two regions means that the intensity of the interior region is 3 ⁇ 4. ⁇ u and the intensity of the exterior region is u 0 ⁇ u.
  • the energy equation of the geometric active-contour model that extracts the object boundary may be defined by the following equation:
  • C denotes the curve of the active contour
  • u is the image intensity
  • the constants /1 ⁇ 2 and ⁇ ⁇ are the average intensities inside and outside of C, respectively
  • x and v are the planar coordinates of the contour.
  • the curve that minimizes the energy equation, equation (1), is the contour that would fit to the edges of the object of interest and distinguish it from the boundaries.
  • the shadow information may be used as described above to detect the building and to isolate it.
  • the relative intensity of the dark regions to the neighbor may provide a cue for shadow detection. This step involves, for each shadow area, finding or detecting the spatial location of the building that generated that shadow by searching within the shadow pixels for the pixel that has a minimal value of luminescence.
  • a building should be a spatial neighbor to that pixel.
  • the curve C that minimizes the energy in equation (1) is the contour that would fit the edges between the two regions. Since the shadows in the image cannot be described by spectral information, prior templates, and statistical information, the relative intensity of the dark regions to the neighbor regions are used as a cue for shadow detection. Reformulating equation (1) may accommodate this strategy, such that the active contours during the process of evolution such that the energy function is selectively biased to enclose the dark regions, including the shadows, from their neighbors. Having enforced the contours to enclose the shadow and other dark regions, then separating them from the rest of the image contents is feasible.
  • the present system and method may further process these regions to isolate the shadow among the dark regions by employing the integrating of the Otsu's threshold with the boundary complexity metric to build a framework for later use, as discussed below.
  • the present method and system may thus involve reformulating equation ( 1) to detect the shadows and the dark regions in the input image, as follows. Assuming an image with the domain ⁇ e R 2 and a level-set representation, i.e., ⁇ H + , an energy functional E total may be formed that favors the contours that surrounding shadows and the dark region by adding an additional energy term E shadow that favors the contours surrounding shadows and the dark region, as follows:
  • ⁇ is the global mean of the input image
  • k is a constant.
  • the value of k is systematically chosen to encode the radiometric characteristics of shadow regions relative to the surrounding pixels in the input image as follows. Comparison of equations (3) and (1 ) shows that ⁇ % is equivalent to k * ⁇ . Therefore, following notations in equation (1 ), k * ⁇ is the average intensity inside C in equation (3), Note that, for k ⁇ I , the average intensity for any image inside C in equation (3) is always less than the global mean of the input image (i.e., the average intensity outside Q, In other words, the value of k in equation (3) controls the average intensity inside C and, hence, the selection of regions by equation (3).
  • Etotai bP Eseg + + « ⁇ length(C) + ⁇ area(inside (Q), (4)
  • a, ⁇ ⁇ 0 are constant parameters relating to the internal energy of the contour, namely tension and stiffness, respectively.
  • the terms length (Q and area (inside ( )) are the length of contour and the area of the region inside contour C, respectively, Equation (4) specifically considers the radiometric characteristics of the dark regions when compared to the radiometric characteristics of the surrounding pixels in the input panchromatic image.
  • the present geometric active-contour model in equation (4) is systematically biased to enclose the regions that exhibit lower average intensity values.
  • the active-contour model detects and segments these regions along with the shadow regions in an image, Therefore, a post-processing processing step may alleviate and remove such clutter regions such as vegetation and water bodies from the segmented image.
  • the level-set implementation method may then be employed to compute the energy function over the input image domain ⁇ ,
  • the curve C is represented by the zero-level set of a function : fl ⁇ R ⁇ , such that:
  • equation (10) discretization of equation (4) is implemented as shown in equation (10) below.
  • h and At are the space iteration step and the time iteration step, respectively, ⁇ ], ⁇ 2 , ⁇ 3 , ' are constants.
  • ⁇ 1 and ⁇ 2 are the average intensities inside and outside of C, respectively.
  • the dark regions obtained by the proposed model for the geometric active contours include shadow and clutter such as water bodies, vegetation, and dark grounds.
  • the following steps may be employed to remove clutter.
  • the result of the given modified model in equation ( 10) is a set of potential shadow regions R ::::: ⁇ 3 ⁇ 4, R-.,... Rtile ⁇ ⁇ ) that may include clutter.
  • further processing of the result may proceed as follows.
  • One may construct Otsu's threshold on the global histogram of the detected regions after adding to each region a strip, with thickness of three pixels, from the surrounding, as shown in Figure 1.
  • Otsu's threshold is an optimal threshold method to automatically segment a bimodal histogram. This shows that three pixels from the surrounding for each region are enough to obtain a global bimodal histogram for the detected regions.
  • one peak is expected to include the shadow and similar intensity pixels in the segmented regions.
  • the other peak will include the surroundings' pixels and similar intensity pixels.
  • the global Otsu's threshold in this step removes the clutter which fall within the peak of the surrounding pixels while at the same time increases the irregularity and complexity of the boundaries of the remaining clutter that fail between the two peaks of the bimodal histogram.
  • High boundary complexity helps in removing the clutter by using the geometric filter as discussed below.
  • the geometric filter may be implemented by comparing the BC measure of each region with a threshold to filter out the regions of values higher than the threshold.
  • the BC measure reflects the regularity and the complexity of the boundary.
  • Figure 3 shows examples of the BC of regular and irregular regions. The regular boundaries have smaller BC values than irregular boundaries, and the BC value increases as the complexity of the boundary increases.
  • the BC measure is higher for the clutter because the clutter, such as vegetation and the water body regions, has boundaries that tend to be irregular and complicated.
  • the shadows of the buildings and other man-made structures have boundaries that tend to be regular; in general, zero BC is associated to straight lines, as shown in the first row in Figure 3,
  • the value of the threshold for the geometric filter is chosen to be 0, 1 5 after extensive testing with all the regions in all the input images considered, so as to keep the shadows of interest and filter out the maximum amount of clutter.
  • a threshold equal to zero may not be ideal in all places for the shadow boundary of man-made structures, such as a building but may be associated with complexity that increases the BC measure, as shown in the example of the first row in Figure 3,
  • the operation of the geometric filter is summarized in Figure 4,
  • a region may be declared true shadow after processing that region with another local Otsu's threshold and then testing its BC using the above geometric filter.
  • This step may be explained as follows.
  • the two steps discussed in Figures 2 and 4 help in removing mainly the bright clutter,
  • bright clutter means that it does not fall within the shadow peak in the global bimodal histogram in procedure GIoha!OtsuThresholdQ in Figure 2 and the clutter that have a high BC measure using procedure GeometricFilterQ in Figure 4,
  • the global processing for the region boundary using the processing in Figures 2 and 4 may not be enough to clearly identify the shadow from the clutter.
  • the present approach includes a method with an approach or algorithm for detection and segmentation of shadow of man-made buildings in panchromatic satellite images, which may be embodied in several forms.
  • the algorithm may be fully automated and systematically biased for shadow detection.
  • the present approach removes the clutter associated with shadow by using an optimal segmentation methodology integrated with a geometric filter scheme to distinguish the clutter from the shadow of man-made structures.
  • This section presents results of the present approach for shadow detection of man- made objects tested on actual panchromatic satellite images for various challenging scenes.
  • the images are from the United State Cjeoiogical Survey (USGS), and each image consists of 4 bands (RGBIR) with a resolution of one pixel per one meter in both directions. For these color images, corresponding panchromatic images were created.
  • the images contained various objects including vegetation, buildings, towers, water bodies, roads and other man- made objects such as cars and trucks.
  • the scenes were selected for containing isolated buildings, connected buildings, small buildings, and large buildings.
  • the shadow content in the input images varied in shape and size such that the shadows might be represented by a few pixels to a large region.
  • the shadow regions could be connected to vegetation and/or bodies of water.
  • the performance of the present approach was evaluated in terms of quantitative metrics as well as qualitative evaluation.
  • the quantitative evaluation of the results was based on ground truth data of the shadows in the input images, which were derived from manual shadow detection; however it is worth note that the prior labeling of each pixel of the shadow regions for each input image was challenging.
  • the initial contour of the geometric active contour was generated automatically in the middle of the input image as a regular circle with a diameter proportional with the size of the input image, as shown in Figure 7(b), After a few iterations, the active contours began to surround all the dark regions, including the shadow regions. In other words, the contours favored and enclosed the regions which were darker than the surrounding region, such as the shadows, vegetation, and bodies of water, as may be seen in Figure 7(d).
  • Figure 7(c) shows the ground truth depiction of the shadow in the input image of Figure 7(a).
  • contours surrounded clutter in addition to the shadow of vegetation and bodies of water (e.g., lakes, canals), as these regions were darker than the surroundings (in the gray level images.)
  • a further processing step may be introduced after the convergence of contours, such that the output of the contours detection is the input to the further processing,
  • Figure 8(a) shows the shadow regions segmented by using the present contour model in equation (10), and the shadow region segmented by the contour model in equation (1 ) is presented in Figure 8(b).
  • the results of segmentation in Figures 8(a) and (b) demonstrate the ability of the present contour model to distinguish the isolated shadow and the dark regions in the input image significantly more efficiently than the regular control model in equation (3 ).
  • Figure 9 shows the final result of the present approach for the input image in Figure 7(a). Note that the present approach shows good detection of the shadows of the man-made buildings in the scene when compared with the ground truth in Figure 7(c). Further note that the present approach detects the shadow of the isolated buildings; however, the shadow of the connected buildings is segmented as one region. In addition, the approach detects the shadows of large buildings such as towers and that of most of the small buildings. The present approach is able to remove all the clutter from the image, except the lake that is connected to the shadow of the lower tower. Therefore, this approach is robust for the clutter, which is common in overhead scenes.
  • the water body is difficult to separate from the shadow of the tower since the water body is adjacent and connected to the shadow of the tower, and these exhibit similar intensity feature characteristics.
  • the water bod and the shadow of the tower are treated as one object by the geometric active-contour method.
  • the Otsu's threshold was determined on the histogram of the detected regions by the geometric active contours after adding for each region a strip from the surroundings, as explained above.
  • Figure 10(d) presents the resuit of using Otsu's threshold with global processing. This step alleviated clutter in the result by removing the clutter which is brighter than the shadow, and this is clear in the roof of the building in the bottom of the image and in some bodies of water. The result of this step was a set of potential shadow regions.
  • the present approach may be evaluated using well-known quantitative metrics.
  • Three types of metrics may be employed as follows: The first type of metrics is named producer's accuracies, which measure the correctness of the approach and indicate how well the true shadow and non-shadow pixels are correctly classified.
  • the second type of metrics is the user's accuracies that measure the precision of the approach and indicate the probabilities of correctly detected and classified pixels (i.e., shadow and non-shadow).
  • the third type of metrics is the overall accuracy that measures the percentage correct.
  • the producer's accuracy of the shadow r s and producer's accuracy of non-shadow T rs are defined by: (1 1)
  • true positive is the number of shadow pixels, which are correctly detected and identified when compared with the ground truth
  • false negative is the number of the true shadow pixels, which are detected and identified by a method as non- shadow pixels.
  • True negative denotes the number of true non-shadow pixel that are correctly detected and identified by a method
  • false positive is the number of non- shadow pixels that are detected and identified by a method as true shadow pixels.
  • the user's accuracy of shadow as and the user's accuracy of non-shadow 0 s are defined by:
  • the overall accuracy is defined by;
  • Table I shows the values of the quantitative metrics for the present approach applied to images in Figures 9-3 :2,
  • the overall accuracies of the present approach exceed those of conventional efforts, such as fixed masking, for example,
  • the present approach outperformed two prior approaches in three categories (x RS , a s , p) and produced comparable performance in ⁇ ⁇ 8 , Moreover, the present approach offered lower performance in x s when compared with one prior approach and comparable results to a fixed masking method.
  • the lower performance was due to changing the label of some pixels from shadow to non-shadow during a clutter detection process.
  • the change in pixel labels was caused by higher brightness values of those pixels in the shadow area, and this, in turn, was reflected in the increased FN value and hence reduced T. s values, respectively.
  • the increased FN value is not desirable, it did not affect the overall accuracy p of the present approach
  • Embodiments of the present geometric active-contour model may be derived from a level-set-based model, in general, the re-initialization step in the level set may be time- consuming and carry expensive computational cost. The re-initialization step is optional; hence, the results here were obtained without re-initialization.
  • Table 2 presents the elapsed time for processing the images in this paper. All algorithms are implemented in computational environment computer software provided by Mathworks, Inc. under the Matiab® brand name. Note that the processing time is highly dependent ors the way scripts may be written, the details in the image contents, and the dimension of the image.
  • a system embodiment is generally adapted to taking pixeiated panchromatic gray scale images of a scene of interest, the image comprising image data representative of that scene of interest.
  • Embodiments of the system raay thus be adapted to effect a first step of optically acquiring, detecting, or capturing a suitable input digital pixeiated gray scale aerial image of interest of sufficiently high resolution.
  • Such an imaging sensor may include a lens, a corresponding optical image system, an analog to digital converter, all in electronic or operable communication with a non-transitory memory.
  • imaging sensors may include, for example, a radiometer, digitizing cameras, CCDs, CMOS image sensors, and optical laser sensors or scanners, etc.
  • the building detection system may include an imaging system comprising a pixelated panchromatic gray scale imaging sensor, an imaging power supply, and an imaging memory, with the imaging system capable of capturing the image data.
  • the imaging sensor may be a pixelated imaging array sensor, with a plurality of photon accumulating light sensors or pixels.
  • the input image or image data may be stored in the non-transitory imaging memory as image data representative of the scene of interest and/or communicated to other aspects of the system,
  • the detection system may include a control or image processing system having a control processor, a non-transient control memory, a control power supply, the control system in operable communication with the imaging system, the imaging system having an imaging output and the control system having a control input by which the image data may be communicated to the control system,
  • the control memory and imaging memory may be shared or common storage structure, a first and second memory, or separate and physically distinct memories, depending on the application. The separate denomination is simply to illustrate the potential for remote or prior imaging as well as local processing. Further, the control memory (and/or imaging memory) may be specially configured with computer software directed to the above process or method embodiments.
  • control system may be responsive to the image data, with the non-transient control memory storing or having control computer code or software executable by the control processor for the detecting and isolating of one or more shadows within the scene of interest using a geometric active contour, followed by the application of the following:
  • ⁇ are internal energy tension and stiffness constants for a contour
  • ⁇ is an external energy constant for the contour
  • h is a space iteration step
  • At is a time iteration step, and finite differences of are calculated from;
  • Control memory may store or hold control computer code or software executable by the control processor for the application of the variety of process embodiments disclosed herein, for the application to or processing of image data,
  • the non-transient memory may further store computer code, when executed by the computer processor, for: (i) applying to each of one or more shadow regions R a three pixel strip from the surroundings of that shadow, and constructing an Otsu's threshold on a global histogram of R; (ii) geometrically filtering image data of the gray scale image to remove data having a boundary complexity greater than a predetermined global threshold 3 ⁇ 4 (hi) applying to each of one or more shadows R a three pixel strip from the surroundings of that shadow, and constructing an Otsu's threshold on a local histogram of R; and (iv) geometrically filtering image data of the gray scale image to remove data having a boundary complexity greater than a predetermined local threshold 2), As noted above, the image data may he drawn from a satellite image.
  • the imaging system may include imaging circuitry placing imaging sensor, imaging memory, and imaging power supply in operable communication (e.g., electronic, optical, etc.). Imaging circuitry may permit individual access of each photosensor pixel or element of the imaging sensor. In some embodiments, imaging system may further comprise an analog to digital converter to convert analog image data to digital data, A control system may also include control circuitry placing the control processor, control memory, and the control power supply in operable communication. Control system and imaging system may communicate in a variety of wired, wireless, or non-transient media transfer formats. In some embodiments, imaging circuitry and control circuitry may constitute a single system circuitry,
  • Embodiments of the control system may be adapted to processing image data algorithmically as disclosed herein, either in total or as a reduced portion of image data.
  • the isolation of image data may be by implemented by a control processor configured for the use of active contour models, such as a zero-level set implementation (i.e., geometric active contours class), to find the contours of structures in the image.
  • the control processor may thus read the pixelated image data to localize or process an object with geometric active contours.
  • the geometric active contours permit localization of shadows or other structures in the image as closed contours objects.
  • Image contents or data may then be mapped to objects defined by a closed contour approach. Shadow objects may then be detected and isolated from the reference contents of an image for further processing, he,, such as finding the building that generated the shadow.
  • Embodiments of the control processor may use the shadow information to detect and isolate a building of interest from image data. This step may involve, for each shadow area, finding the spatial location of the building that generated that shadow by searching the data within the shadow pixels for the pixel that has a minimal value of luminescence, A building should be a spatial neighbor to that pixel. Once the spatial location of the building is determined in the image plane, a segmentation algorithm can be employed to localize the foot print of the building,
  • the control processor may be adapted or configured to estimate the dimensions of a building from the shape and size of the shadow image areas. Embodiments of the approaches discussed above use this information to predict the size and shape of the shadow of the object from the pixel values captured from the overhead sensor.
  • Another, secondary, object's shadow image may be used to compare the size and shape of a first shadow image. Minimal values of each shadow may be compared to determine which building may be taller and which is shorter. The minimal pixel values of image data in each shadow may be compared with the other pixel values of that shadow image.
  • the present system, media, and methods employ a new shadow detection and extraction approach for high-resolution panchromatic satellite image, using a gray level satellite image.
  • the system is based on employing the geometric active contour after tailoring the traditional model to tackle the problem between hands.
  • the new model of the geometric active contours favors the shadow and the similar dark regions in the input image. Having obtained the potential shadow regions, further processing steps are introduced to isolate the shadow from clutter such as vegetation and water bodies.
  • the experimental or evaluation results which have been conducted by using real images, quantitatively and qualitatively show the outperformanee of the present approach compared with a conventional approach.
  • the present approach may be embodied as a method, system, or on a computer readable medium. Accordingly, the present approach may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a "circuit," "module” or “system.” See, for example. Figure 1. Furthermore, the present approach may take the form of a computer program product on a computer readable medium having computer-usable program code embodied in the medium. [0084] Any suitable computer readable medium may be utilized.
  • the computer-usable or computer-readable medium may be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a transmission media such as those supporting the Internet or an intranet, or a magnetic storage device.
  • the computer-usable or computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.
  • a computer-usable or computer-readable medium may be any medium that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
  • Computer program code for carrying out operations of the present approach may be written in an object oriented programming language such as Java, Smalltalk, C++ or the like. However, the computer program code for carrying out operations of the present approach may also be written in conventional procedural programming languages, such as the "C" programming language or similar programming languages.
  • the program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.
  • the remote computer may be connected to the user's computer through a local area network (LAN) or a wide area neivvork (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
  • LAN local area network
  • WAN wide area neivvork
  • These computer program instructions may also be stored in a computer-readable memory, including a networked or cloud memory, that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function/act specified in the flowchart and/or block diagram block or blocks.
  • the computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
  • Any prompts associated with the present approach may be presented and responded to via a graphical user interface (GUI) presented on the display of the mobile communications device or the like. Prompts may also be audible, vibrating, etc.
  • GUI graphical user interface
  • any flowcharts and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present approach.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s).
  • the functions noted in the block may occur out of the order noted in the figures.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

L'invention concerne un système et un procédé de détection et d'isolation de l'ombre de bâtiments ou de structures d'intérêt au sein d'une image d'entrée. Le bâtiment ou la structure d'intérêt est localisé par le système en utilisant la modélisation active des contours géométriques basée sur des régions, et la présente approche peut surmonter la complexité du motif et de la structure par une séquence de traitement global et local pour enlever les données d'image du fouillis.
PCT/US2014/033017 2013-04-05 2014-04-04 Système et procédé de détection de structures Ceased WO2014165787A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201361809056P 2013-04-05 2013-04-05
US61/809,056 2013-04-05

Publications (1)

Publication Number Publication Date
WO2014165787A1 true WO2014165787A1 (fr) 2014-10-09

Family

ID=51659236

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2014/033017 Ceased WO2014165787A1 (fr) 2013-04-05 2014-04-04 Système et procédé de détection de structures

Country Status (1)

Country Link
WO (1) WO2014165787A1 (fr)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017177284A1 (fr) * 2016-04-15 2017-10-19 University Of Southern Queensland Procédés, systèmes et dispositifs se rapportant à la détection d'ombre pour l'identification d'objets en temps réel
CN111179293A (zh) * 2019-12-30 2020-05-19 广西科技大学 一种基于颜色和灰度特征融合的仿生型轮廓检测方法
US20220374634A1 (en) * 2021-05-18 2022-11-24 Here Global B.V. Identifying canopies
CN118552532A (zh) * 2024-07-26 2024-08-27 湖南元数科技有限公司 一种动态演化的车辆智能定损方法与系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6546117B1 (en) * 1999-06-10 2003-04-08 University Of Washington Video object segmentation using active contour modelling with global relaxation
US20090202124A1 (en) * 2006-10-11 2009-08-13 Olympus Corporation Image processing apparatus, image processing method, and computer program product
US7831081B2 (en) * 2005-08-15 2010-11-09 Boston Scientific Scimed, Inc. Border detection in medical image analysis
US20110257514A1 (en) * 2008-10-03 2011-10-20 Centre National De La Recherche Scientifique (Cnrs) Image processing method for estimating a brain shift in a patient
US20120148103A1 (en) * 2009-08-23 2012-06-14 Iad Gesellschaft Fur Informatik, Automatisierung Und Datenverarbeitung Mbh Method and system for automatic object detection and subsequent object tracking in accordance with the object shape
US8289318B1 (en) * 2008-08-29 2012-10-16 Adobe Systems Incorporated Determining three-dimensional shape characteristics in a two-dimensional image
US8391576B2 (en) * 2007-03-22 2013-03-05 Fujifilm Corporation Device, method and recording medium containing program for separating image component, and device, method and recording medium containing program for generating normal image

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6546117B1 (en) * 1999-06-10 2003-04-08 University Of Washington Video object segmentation using active contour modelling with global relaxation
US7831081B2 (en) * 2005-08-15 2010-11-09 Boston Scientific Scimed, Inc. Border detection in medical image analysis
US20090202124A1 (en) * 2006-10-11 2009-08-13 Olympus Corporation Image processing apparatus, image processing method, and computer program product
US8391576B2 (en) * 2007-03-22 2013-03-05 Fujifilm Corporation Device, method and recording medium containing program for separating image component, and device, method and recording medium containing program for generating normal image
US8289318B1 (en) * 2008-08-29 2012-10-16 Adobe Systems Incorporated Determining three-dimensional shape characteristics in a two-dimensional image
US20110257514A1 (en) * 2008-10-03 2011-10-20 Centre National De La Recherche Scientifique (Cnrs) Image processing method for estimating a brain shift in a patient
US20120148103A1 (en) * 2009-08-23 2012-06-14 Iad Gesellschaft Fur Informatik, Automatisierung Und Datenverarbeitung Mbh Method and system for automatic object detection and subsequent object tracking in accordance with the object shape

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017177284A1 (fr) * 2016-04-15 2017-10-19 University Of Southern Queensland Procédés, systèmes et dispositifs se rapportant à la détection d'ombre pour l'identification d'objets en temps réel
US10740610B2 (en) 2016-04-15 2020-08-11 University Of Southern Queensland Methods, systems, and devices relating to shadow detection for real-time object identification
CN111179293A (zh) * 2019-12-30 2020-05-19 广西科技大学 一种基于颜色和灰度特征融合的仿生型轮廓检测方法
US20220374634A1 (en) * 2021-05-18 2022-11-24 Here Global B.V. Identifying canopies
US11854257B2 (en) 2021-05-18 2023-12-26 Here Global B.V. Identifying canopies
CN118552532A (zh) * 2024-07-26 2024-08-27 湖南元数科技有限公司 一种动态演化的车辆智能定损方法与系统

Similar Documents

Publication Publication Date Title
Arévalo et al. Shadow detection in colour high‐resolution satellite images
Huang et al. Individual tree crown detection and delineation from very-high-resolution UAV images based on bias field and marker-controlled watershed segmentation algorithms
Zhou et al. Object-based land cover classification of shaded areas in high spatial resolution imagery of urban areas: A comparison study
US10489674B2 (en) Information extraction using image data
Benarchid et al. Building extraction using object-based classification and shadow information in very high resolution multispectral images, a case study: Tetuan, Morocco
AU2016315938A1 (en) Systems and methods for analyzing remote sensing imagery
US12118807B2 (en) Apparatus and method for three-dimensional object recognition
US20170178341A1 (en) Single Parameter Segmentation of Images
CN117058557B (zh) 基于物理特征与深度学习模型的云和云阴影联合检测方法
CN112465735A (zh) 行人检测方法、装置及计算机可读存储介质
Grigillo et al. Automated building extraction from IKONOS images in suburban areas
Tang et al. Object-oriented change detection based on the Kolmogorov–Smirnov test using high-resolution multispectral imagery
WO2014165787A1 (fr) Système et procédé de détection de structures
Parr et al. Grape yield estimation with a smartphone’s colour and depth cameras using machine learning and computer vision techniques
Gauci et al. A Machine Learning approach for automatic land cover mapping from DSLR images over the Maltese Islands
Azevedo et al. Shadow detection using object area-based and morphological filtering for very high-resolution satellite imagery of urban areas
CN105631849A (zh) 多边形目标的变化检测方法及装置
Aytekin et al. Automatic and unsupervised building extraction in complex urban environments from multi spectral satellite imagery
Koc-San et al. A model-based approach for automatic building database updating from high-resolution space imagery
Albanwan et al. 3D iterative spatiotemporal filtering for classification of multitemporal satellite data sets
Koc-San et al. Support vector machines classification for finding building patches from IKONOS imagery: the effect of additional bands
Ke et al. Individual tree crown detection and delineation from high spatial resolution imagery using active contour and hill-climbing methods
CN116563445B (zh) 基于虚拟现实的动漫场景渲染方法及装置
Liu et al. Identification of Damaged Building Regions from High-Resolution Images Using Superpixel-Based Gradient and Autocorrelation Analysis
Jiao et al. Individual Building Rooftop and Tree Crown Segmentation from High‐Resolution Urban Aerial Optical Images

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: 14779547

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14779547

Country of ref document: EP

Kind code of ref document: A1