AU2005279700A1 - A method for automated 3D imaging - Google Patents

Description

WO 2006/024091 PCT/AU2005/001316 A METHOD FOR AUTOMATED 3D IMAGING TECHNICAL FIELD The present invention relates to the automated construction of 3D surface image data from digital images. 5 BACKGROUND ART The characterisation of the shape of the surface of an object (the surface topography of the object) is required to perform many tasks. Many methods for obtaining information describing the shape of the surface of an object are known. For example, the shape of the surface of -the object may be measured using 10 photogrammetric means or using a scanned laser measurement device. In many cases additional data characterising the visual nature of the surface or other properties of the surface are also required. Many methods for obtaining information describing the visual characteristics of the surface of an object are also known. 15 Using photogrammetric means, three-dimensional spatial data may be acquired from two or more two dimensional images. The pior art requires that the position and orientation of the lines of sight of the cameras or, in cases where one camera is used the position and the line of sight of the camera at each imaging location, are known. 20 According to prior art techniques, the construction of a three dimensional image from two or more two dimensional images requires the determination of points of correspondence in both images and the application of well known mathematical formulae to use knowledge of the positions of correspondence to estimate absolute or relative spatial position of points for which correspondences 25 are determined. This process requires significant processing power to be deployed. The processing is performed to determine single or multiple correspondences within images of a scene. A 'global' search for correspondences must be performed, and may produce many correspondences some of which will be false or may not 30 be unique. It is an object of the present invention to provide an improved process for constructing three dimensional surface images from sensor data where the sensor data may be two or more images.

WO 2006/024091 PCT/AU2005/001316 2 SUMMARY OF INVENTION In a broad form, the present invention uses the sensed distance or distances to a known feature or features in an image to initiate and control the 5 processing of two dimensional image data to produce three dimensional surface data. According to one aspect, the present invention provides a method for processing sets of two dimensional image data acquired using an image sensor so as to produce three dimensional surface data, the method including the steps 10 of: a) Acquiring range data indicative of the distance from the sensor to at least one known feature in each two dimensional Image; b) Processing the two dimensional image data using said range data to initiate and control the determination of correspondences between said two 15 dimensional image data; c) Integrating said processed image data to produce three dimensional surface data. Preferably but not necessarily, the two dimensional image data may be further integrated with data defining the visual characteristics or other 20. characteristics of the surface such as reflectance or emission in other bands of the electromagnetic spectrum. The three dimensional surface data consists of the spatial data defining the shape of an object (or objects), optionally integrated with the visual data or other data that has the format of a visual Image, such as spectral data from spectral 25 regions other than those of visual wavelengths defining the reflective or emissive nature of the object. The range data may be acquired directly from the image sensor, or a separate sensor, and may require integration with data about the orientation and position of the camera. 30 According to one implementation of the present invention, a range measurement device or the projection of a light patten is used to control the initiation of processing to extract three-dimensional spatial data, and as required, construct a three-dimensional image from two images.

WO 2006/024091 PCT/AU2005/001316 3 A three dimensional image is the integrated description of a surface combining spatial data and data such as visual data. The range measurement data allows the processor to more readily determine single or multiple correspondences within images of a scene. Without knowledge of the distance to 5 a feature that is present in the images being processed, as in the prior art, a 'global' search for correspondences must be performed even when the position and orientation of the cameras is known. Such a search may produce many correspondences some of which will be false or may not be unique. Knowledge of the distance to a known point, or points, combined with 10 knowledge of the relative orientation and position of the cameras (relative orientation and position may be derived mathematically or may be determined from a knowledge of absolute position and orientation) and the range measurement device or light projection device may be used to control the search for correspondences since if the position (relative or absolute) in space of a 15 feature in one image is known the position of said feature in a second image is known. This knowledge is then used to control and automate the processing that cetermines the correspondences necessary to create a three-dimensional image... According to the present invention the use of knowledge of the distance or distances to a known feature or features in an image can be obtained by the 20 projection of a known pattern of light into the space in which the 3D image is to be created.in this method the location of the image of the pattern of light is used to then determine the distance or distances to a feature or a set of features in a two dimensional image using methods known in the prior art and the process described for use with direct measurement of range to said features may then be 25 applied. The present invention further encompasses an apparatus and system for producing three dimensional surface data, and a software product operatively adapted to carry out the inventive method. It will be understood that the term object as used in relation to the subject 30 of an image is intended broadly, and may encompass geographic features and terrain as well as specific objects and features in images. The term three dimensional surface data is used broadly to mean data characterising the shape and optionally other characteristics of the surface of an object.

WO 2006/024091 PCT/AU2005/001316 4 BRIEF DESCRIPTION OF DRAWINGS The present invention will now be described with reference to the accompanying- drawings, in which: 5 Figure 1 is a representation of a typical three-dimensional imaging system (sensor) as it may be implemented using a single sensor; Figure 2 is a representation of a 3D imaging system (sensor) utilising two sensors e.g. two cameras as used in a stereo-imaging sensor wherein the field of view of the 3D sensor is defined by the overlap of the field of view of the 10 individuaI sensors; and Figure 3 is a representation of a 3D imaging system (sensor) utilising two cameras used at two locations or one camera used at two locations; and Figure 4 is a block diagram of a suitable system according to one embodiment. 15 Figure 5 is an image of one implementation of an automated 3D imaging system using two cameras and a laser range finder (centre). Figure 6 is a computer representation of a 3D image created by the implementation of an automated 3D imaging system as shown in Figure 5. Figure 7 is a representation of the 3D data component of a 3D image as a 20 spatial point cloud, i.e. a visualisation of the spatial location of each surface point of the 3D image. DETAILED DESCRIPTION Whilst the operation of the invention is described with reference to some particular implementations, it will be appreciated that many alternative 25 implementations are possible. In the conventional application of techniques of photogrammetry the relative positions and orientations of the cameras when two or more cameras are used or the positions and orientations of the camera when only one camera is used are determined and the images are sampled in accordance with the epipolar 30 geometry determined by the relative positions and orientations of the cameras or the camera when one camera is used and a disparity map is created. The disparity map is used to extract three-dimensional spatial data using the relative positions and orientations of the cameras.

WO 2006/024091 PCT/AU2005/001316 5 in the conventional application the search for correspondences is initiated by a number of means and is performed by many methods. The search for correspondences is computationally intensive. To reduce the computation the search can be constrained by the geometry of the cameras wherein the cameras $ are used in a fixed known geometrical relationship or the search can be initiated under the control of a human. Conventionally range information is not acquired and is not used to control the process of construction of a disparity map. The disparity map is generated for all points common to both images after resampling of the images in accordance 10 -with the epipolar geometry. Once the disparity map has been generated three dimensional spatial data is generated from the disparity map. In the present invention the range measurement and the relative positions and orientations of the cameras are used to initiate and control tho search process that produces the disparity information. The use of this control enables 15 the automated generation of a three dimensional image. It will be understood that while the present invention is an improvement to conventional techniques, the general principles and techniques of known systems are relevant to and form part of the practical implementation of the present invention. 20 Photogrammetry is effectively a triangulation system that requires that there are two, or more, components of the sensor which enable the determination of the direction in three dimensions to a point in the object space. The position and orientation of each component of the sensor is known and, since for each component of the sensor the angular direction to the point is known, the position 25 in space of the point is readily determined thus providing a three dimensional image. Triangulation may be subdivided into techniques: passive triangulation and active triangulation. Passive triangulation encompasses such techniques as aerial or terrestrial photogrammetry where the components of the measurement system are two or 30 more cameras, or one camera taking two or more images. Points in each image are matched and from the position in each image of the matched points the spatial position is determined. Fast systems using two television cameras and image processing systems are sometimes classified as stereovision.

WO 2006/024091 PCT/AU2005/001316 6 Active triangulation encompassed such techniques as structured light imaging where a light stripe is projected into an object space and viewed by a camera or similar sensor from another position, In some instances two cameras are used. Knowledge of the direction of projection of the light stripe and the 5 position and orientation of the camera or cameras enables calculation of the position in space of any point reflecting the light. Another form of active triangulation involves the use of a spot of light scanned over the object space. Those skilled in the art will be aware of these and other alternative 3D imaging techniques which can be employed to implement the present invention. 10 These are discussed, for example, in Besl. P.J., Active Optical Range makingg Sensors; Machine Vision and ApOications Vol 11988. The possible implementations of the present invention will now be described in more detail. It is emphasised that apart from the range determining device, the remainder of the system is conventional and may be implemented 15 using well known approaches. Figure 1 is schematic view showing the typical field of view for a single device implementation of the present invention, for example using a laser rangefinder. Such a device inherently provides three dimensional position information about the surfaces sensed. 20 Fgure 2 shows an imaging system utilising two sensors, such as digital cameras used in a stereo imaging configuration. The field of viewof the combined sensors, i.e. where they overlap, defines the 3D sensing zone. Range Information in this case is provided by the automated range determining mechanism from each camera, together with information about the alignment and position of each 25 sensor on the platform. Figure 3 shows another arrangement, using either one 'camera at two locations or two cameras at different locations. A range measurement device is collocated with the camera at each location. Again, the field of view of the 3D sensor is defined by the overlap between the fields of view of the cameras. 30 Two methods for automated creation of three dimensional images will be described below: a) Using a 3D imaging sensor that combines a range measurement device and a sensor that is capable of producing a visual presentation or other WO 2006/024091 PCT/AU2005/001316 7 representation for example a spectral representation using wavelengths other than visual light of a scene. An example of such a sensor is a two-dimensional imaging sensor as in a digital camera, and the range measurement device in the camera may provide the range measurement. The range and image data enables 5 the 3D image to be automatically constructed. b) Using a 3D imaging sensor that combines a mechanism for projecting a known pattern of light into the space in which a three dimensional image is to be constructed and a sensor that is capable of producing a visual representation or other representation for example a :spectral representation 10 using wavelengths other than visual light of a scene such as a two-dimensional imaging sensor as in a digital camera is used to acquire data from which the 3D image is automatically constructed. In the preferred embodiment, the process of creating a 3D image is undertaken as follows when a range measurement device is used. In the 15 description provided the use of two images is assumed the process may obviously be readily extended to the use of more than two images. 1. A sensor is used to acquire a two dimensional image of a scene. 2. A sensor is used to acquire a second two dimensional image of an overlapping section of the scene. 20 3. A range measurement device that may be collocated with the sensor or may be positioned at a known position relative to the sensor determines the distance to a feature or features in the scene. Knowledge of the spatial relationship is then used to predict the position of the feature to which the range has been measured in both images. 25 4. A processor locates these features in the two dimensional images using the knowledge of the distance to these features obtained by measurement using the range finder, 5. The processor uses knowledge of the position of these features in both image to control the process of image data to locate correspondences in the 30 two images that are used to create a three dimensional image. 6. The processor determines all possible correspondences between the two images WO 2006/024091 PCT/AU2005/001316 8 7. The correspondences are used with the data defining the position and orientation of the sensors to create a three dimensional image. A preferred practical implementation of the present invention requires the following components. 5 First, a means for acquiring two dimensional digital images is required. A suitable device is a digital camera or camera, with an appropriate data interface to output image data. A means for measuring distance to an object from a camera in a direction relative to the camera that is known and aligned to the camera is required. This 10 may be done using the existing range measurement device in most digital cameras. Such devices are used to provide auto-focusing in many cameras and may be an ultrasonic range measurement system or may utilise an image processing algorithm that determines when the camera is properly focussed and infers the range from the focus setting. 15 One alternative is to use a device which projects a known pattern of light (or other radiation) into the scene space, the range being determined by the reflected light sensed by the device in response to the known output. A suitable processor is required. The nature of the processor is to an extent dependent upon the size of the operating field and the volume of data 20 being processed. For processing of relatively small images an embedded digital signal processor may be used in local equipment. For large images a computer such as a personal computer may be required. Means for storing data acquired by the imaging system and other data is required, of a capacity and speed compatible with the required application. 25 The automated 3D image construction Is performed by a suitable algorithm to be executed by the processor. Such algorithms are known.inthe prior art and are embodied in systems in the following form; 1. The algorithm estimates the disparity between two images by performing correlation of similar features in the images 30 2. The disparity between two images is used with knowledge of the spatial relationship between the cameras to determine the position in space relative to the cameras of the features in the images for which the disparity has been estimated.

WO 2006/024091 PCT/AU2005/001316 9 An embodiment of the system is shown in Figure 4. In the preferred embodiment, the process of creating a 3D image is undertaken as follows, when a mechanism for projecting light pattern into the space in which a three dimensional image is to be constructed is used: 5 1. A sensor is used to acquire a two dimensional image of a scene that may include a projected light pattern. 2, A sensor is used to acquire a second two dimensional image of an overlapping section of the scene that may include a projected light pattern. 3. A processor locates features irr the imags that correspond to the 10 projected light pattern and determines the distance to these features as is well known in the prior art (e.g. Close Range Photogrammetry and Machine Vislon edited by K. Atkinson). 4. The processor uses knowledge of the position of these features in both image to control the process of image data to locate correspondences in the 15 two images that are used to create a three dimensional image. 5. The processor determines all possible correspondences between the two images 6. The correspondences are used with the data defining the position and orientation of the sensors to create a three dimensional image, 20 Figure 5 is a photograph illustrating one practical implementation of a 3D image sensing apparatus according to the present invention. The range sensor is shown at the centre. The devices to the left and right are digital cameras. It will be understood that the relationship between the cameras and the image sensor is known and can be controlled to a high levol of precision. As a consequence, the 25 geometric relationship between the three sensing components is known and can be routinely factored into the image processing. Figure 6 illustrates a 3D image produced by the system of figure 5. Figure 7 shows the same SD image data, represented as a spatial point cloud. A general process for creating a 3D image using two overlapping images 30 according to one implementation of the present invention is-, 1. The sensor system acquires two overlapping digital images of the scene from which the 3D image is to be created.

WO 2006/024091 PCT/AU2005/001316 10 2. The processing system corrects each image to remove or minimise the effects of lens distortion. 3. The sensor system acquires one or more measurements of range to objects in the scene. For a single measurement this will usually be to 5 whatever object is at the centre of the field of view. 4. The processing system uses knowledge of the relative position and orientation of the cameras to determine the degree of horizontal overlap of the digital images. For example if the distance to objects in the scene is large in comparison to the separation of the cameras and the lines of sight 10 of the cameras are parallel (or nearly so) the overlap will approach 100 per cent of the image width. 5. The processing system uses knowledge of the relative position and orientation of the cameras to determine the degree of vertical overlap of the digital images. As is well known in the prior art, if the relative lines of 15 sight of the cameras are known the vertical alignment of the image planes Is determined by the epipolar geometry. 6. Using knowledge of the horizontal and vertical overlap the area of each image that is to be processed to determine the stereo disparity is determined by the processing system. 20 7. The'processing system thus searches the processing area in each image and identifies the disparity between corresponding points in each image. 8. Using the knowledge of the relative position and orientation of the lines of sights of the cameras the processing system converts the disparity information into a spatial location relative to the cameras and thus creates 25 a SD image. It will be understood that the present invention may be implemented using alternative constructions and additional features to those specifically disclosed, but that the present invention encompasses such alternatives. 30

Claims (8)

1. A method for processing sets of two dimensional image data acquire using an image sensor so as to produce three dimensional surface data, the 5 method including at least the steps of: (a) Acquiring range data indicative of the distance from the sensor to at least one known feature in each two dimensional image; (b) Processing the two dimensional image data using said range data to initiate and control determination of correspondences between said two 10 dimensional image data so as to produce processed image data; and (c) Integrating said processed image data to produce three dimensional surface data.

2. A method for automatically generating three dimensional surface data using two dimensional imaging sensors, including at least the steps of : 15 a) obtaining a first image of the object from a first position using a first image sensor; b) obtaining a second image of the object from a second position using a second image sensor; c) obtaining a distance measurement from the first image sensor to a 20 point in the field of view of the first image d) determining the relative displacement of the first and second positions; e) using at least the distance measurement from the first image and the relative dispIacement of the first and second positions to guide the initiation of 25 a search for correspondences between the images to initiate the construction of a disparity map defining the displacement of image features between the two images; f) using the relative displacement between the cameras and the distance measurement or measurements to control the processing used to create 30 the disparity map; WO 2006/024091 PCT/AU2005/001316 12 g) using the relative displacement between the cameras and the distance measurement or measurements to control termination of the processing used to create the disparity map; and h) thereby constructing three dimensional surface data. 5 . A method according to claim 2, wherein more than 2 images are used,

4. A method according to claim 2 or claim 3, wherein more than one distance measurement is obtained.

5. A method according to any one of claims 2 to 4, wherein one or more 10 distance measurements are also made from at least the second image sensor to a point in the field of view of the second image.

6. A method according to claim 1, wherein the second image sensor is the first image sensor moved to the second position.

7. Apparatus for constructing three dimensional surface data, from sets of 15 two dimensional image data, said apparatus including at least two two dimensional image sensors, and a range sensor, said sensors arranged so as to operatively maintain a specific physical relationship, and processor means adapted to receive data from each of said image sensors and from said range sensor, such data including a set of two dimensional image data from each image 20' sensor and range data indicative of the distance from the range sensor to at least one knownfeature in each two dimensional image; said processor being adapted to process the two dimensional image data using said range data to.Initiate and control determination of correspondences between said sets of two dimensional image data so as to produce processed image data; and being further adapted to 25 integrate said processed image data to produce three dimensional surface data.

8. Apparatus according to claim 7, wherein the at least two image sensors are provided by a single image sensor which is moved between two or more positions.

9. A software product operatively adapted to implement the method 30 according to any one of claims 1 to 6.