JP2010187130A - Camera calibrating device, camera calibration method, camera calibration program, and recording medium having the program recorded therein - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To facilitate measurement of internal and external parameters of a camera. <P>SOLUTION: A silhouette image creation means 12 creates a silhouette image resulting from extracting an area of an object from an image inputted from an image acquisition means 11 like a camera. A parameter estimation means 14 selects a preliminarily prepared change type of the object in accordance with an arbitrary probability and generates a state candidate according with the selected change type. A simulation image is created by projecting the silhouette image on a virtual camera having the same camera internal and external parameters as the generated state candidate on the basis of the state candidate and simulating the silhouette image. An evaluation value is obtained which results from product operation between a result of comparison between both images and a determination result of likelihood of the object itself based on prior knowledge. It is determined whether the state candidate can be accepted as the final change state or not in accordance with the evaluation value. Internal and external parameters of the camera are estimated from the accepted change state. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for estimating internal and external parameters of a camera using an image obtained by photographing an object including a person / animal with a camera.

  In the field of computer vision, much research has been conducted on tracking moving objects such as people using camera information. This person tracking technology is mainly applied to video surveillance technology and the like. Here, for example, the following processing is realized.

  That is, a target object is observed using an image acquisition device such as a camera, and a silhouette image obtained by extracting a target region (silhouette) on the image is prepared. When the internal and external parameters of the camera used for tracking are measured in advance and obtained, a person model is placed in the three-dimensional space, and this scene is shot with a virtual camera having the previously calculated parameters, a silhouette image is obtained. Can be simulated.

  A method for tracking a plurality of persons by comparing a simulation image generated using an ellipsoid with a silhouette image and a silhouette image has been proposed. As an example of this, the technique of Non-Patent Document 1 is known.

Tatsuya Osawa, Atsuko Kudo, Hiroyuki Arai, Hideki Koike “Tracking multiple adjacent objects using monocular moving images” IEICE Tech. 108 no. 93 Published on June 12, 2008 P1-P6

  According to the method of non-patent document 1, a person directly predicts a position in a three-dimensional space, and captures the predicted result with a virtual camera having the same internal and external parameters as the camera actually installed. By simulating the image and comparing the consistency with the actually captured image, the position of the person in the three-dimensional space is tracked using only one camera information.

  However, in order to simulate a silhouette image, it is necessary to measure the internal and external parameters of the camera used for shooting in advance. However, since measurement of the internal and external parameters of the camera requires specialized knowledge, there is a possibility that only a trained skilled person can perform the measurement.

  The present invention has been made to solve the above-described problems of the prior art, and an object of the present invention is to provide a technique capable of easily measuring internal and external parameters of a camera.

  Therefore, in order to solve the above problems, the present invention automatically evaluates the internal / external parameters of the camera by evaluating the prediction result of the change state of the object (number of objects, size of the object, position of the object). Estimated.

  One aspect of the present invention is a camera calibration apparatus for obtaining internal and external parameters of a camera used for photographing from a photographed image of an object, and a silhouette image obtained by extracting a region where an object is photographed from the photographed image of the camera A silhouette image creating means for creating a parameter estimation means for estimating internal and external parameters of the camera by evaluating the predicted state according to the change type of the object using the silhouette image. Prepare.

  Another aspect of the present invention is a camera calibration method for obtaining internal and external parameters of a camera used for photographing from a photographed image of an object, wherein the silhouette image creating means captures an object from the photographed image of the camera. A first step of creating a silhouette image in which a region is extracted, and the parameter estimation means evaluates the state predicted according to the change type of the object using the silhouette image, thereby A second step of estimating external parameters.

  In both aspects, the state of the object is represented by a set of ellipsoidal models, and approximates to the body, the two-dimensional coordinate values on the model top, the short axis of the ellipsoid that approximates the head, and the body. It is preferable to represent the person by the ellipsoidal short axis and the height of the model indicating the height.

  The present invention may be provided in the form of a program that causes a computer to function as the camera calibration device, or may be provided in the form of a recording medium on which the program is recorded.

  According to the present invention, since the internal and external parameters of the camera are automatically estimated, it is easy to measure the parameters.

Schematic which shows the imaging | photography condition of a camera. The coordinate diagram showing the position and orientation of the camera. Schematic of an ellipsoid model showing a three-dimensional model. The block diagram of the camera structure apparatus which concerns on embodiment of this invention. The chart figure which shows the process of the camera calibration apparatus. (A) is an input image, (b) is a silhouette image. (A) is an input image, (b) is a vertical edge image, and (c) is a horizontal edge image. The flowchart which shows the process step of estimation of the parameter. Probability distribution map of horizontal head position calculated from the same silhouette image. The probability distribution map of the head position in the vertical direction calculated from the edge detection image. (A) is an input image of the embodiment, (b) is a silhouette image of the embodiment, (c) is an edge image (horizontal) of the embodiment, (d) is an edge image (vertical) of the embodiment, and (e) is Simulation image of the example, (f) is the three-dimensional position of the object estimated in the example

(1) Basic Principle First, the basic principle of the present invention will be described below. The present invention captures one or a plurality of objects (including a person / animal) existing in a three-dimensional space, thereby changing the state of each object, that is, the number of objects, the position of the object, the size of the object, and the imaging. The camera internal and external parameters of the image acquisition means used for the estimation are estimated.

  Here, it is assumed that one or more objects to be photographed exist on the same plane. As an example, as shown in FIG. 1, a camera 1 (for example, a digital camera or a video camera) that targets a plurality of persons walking on a plane and whose positions and postures are fixed may be used. In this case, the coordinates that define the three-dimensional space can be set arbitrarily, so the XY plane should be the plane on which the person is walking and the Z axis should be set to represent the height from the floor. Is possible.

  As shown in FIG. 2, if the camera 1 is placed at the origin of the X axis and the Y axis, the three-dimensional position of the camera 1 is represented by (0, 0, Tz) and one variable indicating the height from the plane. The posture of the camera 1 can also be expressed by two variables using rotation angles (φ, θ) around the X axis and the Y axis. These three variables (Tz, φ, θ) indicate the external parameters of the camera 1.

  Next, the internal parameters of the camera 1 can be expressed only by the focal length f of the lens, assuming that the distortion of the camera lens is ignored and the projection center indicates the center of the image.

  Using these internal and external parameters of the camera 1, a projection matrix P necessary for calculating where on the two-dimensional image a point on the three-dimensional space is projected can be calculated by the following equation (1). Is possible. In Equation 1, the matrix I is a 3 × 3 unit matrix.

  In order to represent the target object, as shown in FIG. A model approximating the shape of a person was defined using Q. Here, the model represents the two-dimensional coordinate value (x, y) on the image of the top of the model, the short axis r2 of the ellipsoid P approximating the head, the short axis r2 of the ellipsoid Q approximating the body, and the height. A person is represented by a height h. The major axis and the minor axis of the ellipsoid P are considered to maintain a constant ratio, and the major axis radius r = k1 × r1. Here, k1 is an arbitrary constant, and for example, a numerical value of 1.5 may be used. Similarly, the major axis radius of the ellipsoid Q is set to r = h−k2 × r1. k2 is an arbitrary constant, and for example, a numerical value of 1.2 may be used.

  Therefore, the state of one person can be expressed by five parameters (x, y, r1, r2, h). The target state representing the states of a plurality of persons can also be represented by arranging vectors M (person models) representing the positions and sizes of the persons. That is, the state S including the internal / external parameters of the camera 1 and the states of a plurality of persons is expressed by the following Expression 2.

Here, the person model M _k representing the position, size, and shape of the person k constituting the state S is a five-dimensional vector represented by the following Expression 3.

Here, the five parameters constituting the person model M _k are parameters of a model that approximates the shape of the person described above.

  As described above, if the state S including the internal / external parameters of the camera 1 and the states of a plurality of objects is known, it is possible to simulate a silhouette image in the same manner as in the method of Non-Patent Document 1. Is self-explanatory.

  The present invention evaluates the predicted state S by predicting the state S, simulating a silhouette image, comparing it with an actual silhouette image, and evaluating consistency.

  The basic principle of the present invention is to estimate a state S representing a correct combination of parameters by using the MCMC (Markov Chain Monte Carlo) method which is a kind of probability statistical state distribution estimation method for the above processing. This basic principle is reflected in the camera calibration apparatus according to the embodiment of the present invention.

(2) Specific Embodiment Next, an embodiment of the present invention will be described with reference to FIGS. FIG. 4 shows a configuration example of the camera configuration device 10 according to the embodiment of the present invention. Assume that this camera configuration apparatus 10 is connected to an image acquisition unit 11 that acquires an image of an object via a network.

  The image acquisition means 11 corresponds to a means for acquiring image data taken at each time, that is, the camera 1. More specifically, the image sensor includes an image sensor that converts an image of an object captured by a digital camera or a video camera into a digital signal. Here again, it is assumed that an object to be imaged (one or a plurality of persons / animals, etc.) exists on the same plane.

  Specifically, the camera calibration device 10 is configured by a computer, and image data is input from the image acquisition unit 11 in time series. Here, the camera calibration device 10 is a result of cooperation between computer hardware resources (CPU, memory, hard disk drive, communication interface, etc.) and software, resulting in silhouette image creation means 12, edge image creation means 13, parameter estimation. It functions as the means 14.

  The silhouette image creating means 12 creates a silhouette image obtained by extracting only the region where the target object is shown from the input image data. The edge image creating means 13 creates an edge image obtained by extracting an edge from input image data.

  The parameter estimation unit 14 estimates the state S using the silhouette image created by the silhouette image creation unit 12. That is, the number of objects present in the silhouette image, the size of the object, the three-dimensional position of the object, and the internal / external parameters (focal length f, three-dimensional position (0, 0, Tz) and X A posture (φ, θ)) expressed by an axis and a rotation angle around the Y axis is estimated. This estimation result is output to a monitor or a printer. The processing steps (S501 to S504) of the camera calibration device 10 will be described below based on the chart of FIG.

  S501: When processing is started, captured image data (Im) is input from the image acquisition means 11 via a communication interface (S501). In this embodiment, description will be made on the assumption that the data is input via a network as an example. However, any acquisition means may be used as long as the image data (Im) can be acquired.

  S502: The silhouette image creating means 12 creates a silhouette image (Sil) for the image data (Im) input in S501. The silhouette image is a binary image in which the luminance value of the region where the target object of the photographed image is reflected is 1 and the other region is 0. FIG. 6A shows an example of the input image. b) shows an example of a silhouette image. Such a silhouette image can be easily generated by using a known method such as a background difference method or an inter-frame difference method.

  S503: The edge image creation means 13 creates an edge image for the image data (Im) input in S501. Here, the edge detects the vertical direction and the horizontal direction, and (Eh, Ev) having the vertical direction Eh and the horizontal direction Ev as components is stored for each pixel, and this is used as an edge image.

  The edge image can be easily generated by using a well-known Sobel operator or the like. As an example, FIG. 7A shows an input image, FIG. 7B shows a vertical edge detection image, and FIG. 7C shows a horizontal edge detection image. Note that S502. The silhouette image / edge image created in S503 may be stored / saved in a memory (RAM) or a hard disk drive.

  S504: Using the silhouette image (Sil) created in S502 by the parameter estimation means 14, the number of objects existing in the image, the three-dimensional position, the size, the inside / outside of the camera in the image acquisition means 11 The state S including the parameters [focal length f of the camera, three-dimensional position (0, 0, Tz), posture (φ, θ) represented by rotation angles around the X axis and Y axis] is estimated.

Hereinafter, the specific processing content (S601 to S607) of S504 will be described based on the flowchart of FIG. Here, the estimation of the state S is performed using (B + P) state candidates generated by a plurality of (B + P) iterations. Further, the initial state is represented as S ′ ₀ , and the state candidate generated in the n-th iterative calculation is represented as S ′ _n . The state candidates generated in the first B times are discarded, and the state parameters obtained in the last P times can be generated with equal probability, and the parameters are estimated. This is because the state candidate generated at the beginning is considered to largely depend on the initial state S ′ ₀ .

  Then, the processing is started as the number of updates (n = 1). The initial state of S is arbitrary, but can be set as follows, for example. (F, Tz, φ, θ) = (X_SIZE, 200, 0.0, 0.0), and the state of the person is an empty set. That is, S = (X_SIZE, 200, 0.0, 0.0). X_SIZE is the horizontal size of the input image.

  S601: A change type is selected. The change type represents the type of processing for changing the target state, and consists of the following seven types.

  1: Add person, 2: Delete person, 3: Change person's position, 4: Change person's size, 5: Change focal length, 6: Change camera position, 7: Change camera posture.

  First, when a person is not included in the current state, (1: Add person) is always selected. In other cases, the probability that each type is selected is set arbitrarily in advance, and the probability is selected from four types. For example, (p1, p2, p3, p4, p5, p6, p7) = (0.05, 0.05, 0.2, 0.2, 0.1, 0.2, 0.2) (Pi is the probability that type i will be selected).

S602: A state candidate S ^* is generated according to the change type selected in S601. The processing will be described below for each change type.

<1: Add person>
When the addition of a person is selected, it is necessary to newly set model parameters. For example, the following method described in Document 1 may be used.

First, regarding the parameters r1, r2 and h representing the size of an ellipsoid that approximates a person, the range of values that can be taken from empirical prior knowledge about the size of the person (such as no one with a height exceeding 3 m) is limited. can do. Here, this range is set to r1: r1 _min ≦ r1 ≦ r1 _max
r2: r2 _min ≤ r2 ≤ r2 _max
h: h _min ≤ h ≤ h _max
By generating uniform random numbers under these conditions, parameters r1, r2, and h can be newly generated.

  The parameter relating to the head position (x, y) can be generated as follows. First, a horizontal head position x is generated. Assuming that a person is photographed so as to stand in the vertical direction on the image as shown in FIG. 9, if a silhouette image (Sil) in which only the object is reflected in the vertical direction at a certain position x is added, This value V increases at a position where the head of the person exists. The addition amount V (x) at a certain position x is expressed by the following Equation 4, which is calculated for all x on the image.

  However, Sil (x, y) indicates the x row and y column components of the silhouette image, and Y indicates the vertical size of the image. In order to make this V (x) a head position probability distribution in the horizontal direction, normalization is performed so that the sum of V (x) of all x on the image becomes 1.0, and the generated horizontal direction The position x according to the head position probability distribution V (x) is extracted. This can be realized, for example, as follows. That is, the value val is obtained using a uniform random number between 0.0 and 1.0. It can be realized by adding V (x) to the value sum in order from x = 0 and extracting x when sum ≧ val for the first time.

  Next, a vertical head position y is generated. Since the parameter representing the size of the ellipsoid that approximates the person has already been generated, the projection matrix P is calculated using this and the camera internal and external parameters (f, Tz, φ, θ), and the human head An ellipsoid representing a part (hereinafter referred to as a head model) can be projected on the image. As shown in FIG. 10, when the value obtained by adding the values of the edge detection images (Ev, Eh) on the head model and taking the average is W (y), the value increases in the vicinity of the human head. W (y) is calculated by the following formula 5.

  Here, HL is the head contour length, n (i) is the normal vector of the head model, and g (i) = (Ev (i), Eh (i)) is the value of the edge detection image in the head contour. Is a gradient vector.

  W (y) is calculated for all y on the image, and finally, normalization is performed with the sum of W (y) being 1.0 so that this becomes the vertical head position probability distribution at position x. The position y is extracted according to the head position probability distribution W (y) in the vertical direction. This can be extracted in the same manner as in the horizontal direction.

A new model parameter M _n can be generated as described above. The provisional state candidate S ^* is a state in which a new target state M _n is added to S ′ _n−1 .

<2: Erase person>
When the deletion of the person is selected, first, a target is randomly selected from the previously generated state candidates S ′ _n−1 . That is, when there are K objects, each object is selected with a probability of 1 / K. Next, the selected object M _k is deleted. The provisional state candidate S ^* is a state in which the target state M _k is deleted from S ′ _n−1 .

<3: Person position change>
When the position change of the person is selected, first, a target is randomly selected from the previously generated state candidates S ′ _n−1 . That is, when there are K objects, each object is selected with a probability of 1 / K. Next, among the selected elements of M _k , the provisional position (x ′ _k , y ′ _k ) changed from the position (x _k , y _k ) of the person on the two-dimensional plane according to the following equation 6 ).

However, δ _x and δ _y are one-dimensional Gaussian noises, and their parameters can be experimentally determined to arbitrary values. M ′ _k is defined as in Equation 7 below.

The provisional state candidate S ^* is a state in which the target state M _k of S ′ _n−1 is updated to M ′ _k .

<4: Changing the size of a person>
If the magnitude of change of the person is selected, first select a target at random from the state candidate S _'n-1 that was last generated. That is, when there are K objects, each object is selected with a probability of 1 / K. Next, among the selected elements of M _k , the provisional values changed from (r1 _k , r2 _k , h _k ) representing the radius r1, radius r2 and height h of an ellipsoid approximating a person according to the following equation (8): The parameters of the general shape and size (r1 ′ _k , r2 ′ _k , h ′ _k ) are calculated.

However, δ _r1 , δ _r2 , and δ _h are one-dimensional Gaussian noises, and their parameters can be experimentally determined to arbitrary values. Using this provisional shape and size parameter, M ′ _k is defined as in Equation 9 below.

The provisional state candidate S ^* is a state in which the target state M _k of S ′ _n−1 is updated to M ′ _k .

<5: Change focal length>
If the focal length change is selected, the focal length parameter of the camera is updated from f to a temporary focal length f ′ changed according to the following expression 10.

However, δ _f is a one-dimensional Gaussian noise, and its parameter can be determined to an arbitrary value experimentally.

<6: Camera position change>
When the camera position change is selected, the camera position parameter is updated from Tz to the provisional camera position Tz ′ changed according to the following Expression 11.

However, δ _Tz is a one-dimensional Gaussian noise, and its parameter can be determined to an arbitrary value experimentally.

<7: Camera posture change>
When the camera posture change is selected, the camera posture parameter is updated with the temporary camera posture rotation angle (φ ′, θ ′) changed from (φ, θ) according to the following Expression 12.

However, δ _Φ and δ _Θ are one-dimensional Gaussian noises, and their parameters can be experimentally determined to arbitrary values.

S603: An image simulating the silhouette image (Sil) created in S503, that is, a simulation image (Sim) is created using the temporary state candidate S ^* in S602. Here, the projection matrix P representing the projection relationship of the object existing in the three-dimensional space onto the camera image can be calculated from the parameters (f, Tz, φ, θ).

By using this relationship, it is possible to calculate the two-dimensional coordinate value (X, Y) on the plane of the object of height h standing upright on the plane projected to the two-dimensional coordinate value (x, y) on the image. It is. Thereby, a plurality of 3D models composed of ellipsoids based on the provisional target state S ^* calculated in S602 can be arranged in the 3D space.

  A simulation image is created by projecting the virtual scene generated as described above onto a virtual camera. By using a binary image in which the value of the ellipsoidal region on the image is 1 and the value of the other region is 0, a simulation image (Sim) simulating a silhouette image (Sil) can be created.

S604: The likelihood L is calculated in order to determine whether or not the provisional target state S ^{* in} S602 is likely. This is performed by determining the probability of the target state itself based on prior knowledge and comparing the silhouette image (Sil) created in S503 with the simulation image (Sim) generated in S603.

  Here, the determination of the certainty of the target state itself based on prior knowledge is a condition that persons do not overlap in a three-dimensional space. Now, since the person is approximated by an ellipsoid model, it is necessary to satisfy the condition that the individual ellipsoids do not overlap, that is, the distance between the centers of two different ellipsoids is greater than the sum of the radii of the two ellipsoids. There is.

In order to prevent two persons from overlapping each other in a three-dimensional space, a penalty function E corresponding to the distance R expressed by the following equation 13, that is, the distance between the centers of ellipsoids representing the two persons. Set (S ^* ).

Here, X _k and Y _k indicate the position X coordinate and Y coordinate on the plane of the person k in the three-dimensional space calculated when the simulation image is generated. The function E that gives a penalty when the distance between the ellipsoids becomes short can be set as in the following Expression 14, for example. However, α is a constant and can be arbitrarily set experimentally.

  Next, a comparison between the silhouette image (Sil) and the simulation image (Sim) generated in S603 will be described. This can be calculated by, for example, the following evaluation formula (15). However, (u, v) indicates the u row v column component of the image.

Finally, the likelihood L of the provisional state S ^* is calculated by a product operation represented by Expression 16.

  By using such likelihood, it becomes possible to combine the accuracy of the person position in the three-dimensional space and the tracking on the two-dimensional image.

S605: Using the likelihood L calculated in S604, the provisional state candidate S ^* is accepted or rejected. This is determined by calculating the acceptance rejection judgment probability A. The acceptance refusal judgment probability A can be calculated by, for example, the following Expression 12.

Here, if it is the nth update at present, L ′ represents the likelihood of the _n− 1th candidate state S ′ _n−1 . That is, the current state is always accepted if the likelihood is higher than the previous state, and if the likelihood is lower than the previous state, the provisional state S ^* is adopted with a probability of L / L ′.

Here, if the provisional state S ^* is adopted, S ′ _n = S ^* is set, and if adoption is rejected, S ′ _n = S ′ _{t, n−1} is set and the process proceeds to the condition determination step of S606.

  S606: It is determined whether the update process has been performed a prescribed number of times. If the number of updates n exceeds B + P times represented by the sum of constants B and P that can be arbitrarily determined, the iterative calculation process ends, and the process proceeds to the target state estimation step of S607.

  Conversely, if the number of updates n does not exceed B + P times, the process returns to S601 and the process is repeated again. It is assumed that the prescribed number of update processes is defined in a program or the like.

  S607: A state having the maximum probability is calculated. Since it is assumed that the last P states of the state candidates generated by the iterative calculation all occur with equal probability, the state S having the maximum probability can be calculated by the following Expression 18.

  The state S calculated in this manner is obtained by photographing one or more objects or persons existing in the three-dimensional space, and the number, position, and size of each object, and the image acquisition unit 11 used for photographing. This is an estimation result including the internal and external parameters of the camera at.

  Since this estimation result includes the internal and external parameters of the camera, it is easy to measure the parameters. In addition, since the state S calculated here has the highest probability of the person position in the three-dimensional space, the internal and external parameters of the camera included here are used as appropriate parameters for camera calibration and the like. It can be used to simplify work.

  In particular, until now, the measurement of the internal and external parameters of the camera has been problematic in that only a trained skilled person can measure it. According to the camera construction device 10, it is automatically estimated from the image. Therefore, application such as estimation and tracking of a three-dimensional position of an object appearing in an image can be performed only by installing image acquisition means 11 such as a camera.

  It is assumed that the intermediate data in S601 to S606 is appropriately stored and saved in a memory (RAM) or a hard disk drive, and Equations 4 to 18 are defined in a program or the like.

  The present invention is not limited to the above-described embodiment. For example, an evaluation function that takes similarity between images instead of Equation 15 may be used to compare the silhouette image and the simulation image in S604. Even if a simple product operation is used for this evaluation function, the processing of S604 can be realized in the same manner.

  Further, the present invention can also be configured as a program that causes a computer to function as a part or all of the means 12 to 14 of the camera calibration apparatus 10. In this case, S501 to S504. Cause the computer to execute all or some of steps S601 to S607.

This program can be provided through a network such as a website or e-mail. The program is recorded in a recording medium 10 such as a CD-ROM, DVD-ROM, CD-R, CD-RW, DVD-R, DVD-RW, MO, HDD, Blu-ray Disk (registered trademark). It can also be stored and distributed. The recording medium 10 is read using a recording medium driving device, and the program code itself realizes the processing of the above embodiment, so that the recording medium also constitutes the present invention.
(3) Embodiment Hereinafter, the camera calibration device 10 is used to process the video obtained from the video camera, and the result of tracking using the photographed person will be described with reference to the processing flow of FIG. . First, when processing is started, a camera image is acquired (S501), and a silhouette image and an edge image are created (S502, S503). FIG. 11A shows an input image from a video camera, FIG. 11B shows a silhouette image created using the background subtraction method, and FIGS. 11C and 11D show a silhouette image created using the Sobel operator. An edge image is shown.

  Next, parameters are estimated in S504. Here, a simulation image simulating the state is generated while the state candidates are sequentially generated, and the evaluation is performed. FIG. 11E shows a simulation image representing a target state that takes the calculated maximum probability, and FIG. 11F shows a bird's eye view of the target state. As can be seen from FIG. 11F, the three-dimensional positional relationship of the person is correctly estimated. Here, if the internal and external parameters of the camera are not accurately estimated, the three-dimensional positional relationship of the person cannot be estimated correctly, and the accuracy of the estimation results of the internal and external parameters of the camera is also shown.

  According to the processing described above, in this embodiment, estimation results of parameters such as the three-dimensional position, the number of objects, the size of the object, and the focal length, position, and posture of the video camera are automatically output from the image of the person photographed. The

1.11 Image acquisition means (camera)
DESCRIPTION OF SYMBOLS 10 ... Camera calibration apparatus 12 ... Silhouette image creation means 13 ... Edge image creation means 14 ... Parameter estimation means

Claims (10)

A camera calibration device for obtaining internal and external parameters of a camera used for photographing from a photographed image of an object,
Silhouette image creation means for creating a silhouette image obtained by extracting a region where an object is captured from a photographed image of the camera;
Parameter estimation means for estimating internal and external parameters of the camera by evaluating the predicted state according to the change type of the object using the silhouette image;
A camera calibration device comprising: The parameter estimation means selects a prepared object change type according to an arbitrary probability, and generates a state candidate according to the selected change type;
Means for projecting onto a virtual camera having the same internal and external parameters of the camera as the state candidate based on the state candidate, and generating a simulation image simulating the silhouette image;
Means for evaluating the state candidate from the result of comparing the silhouette image and the simulation image and the determination result of the probability of the target state itself based on prior knowledge;
The camera calibration apparatus according to claim 1, further comprising: The means for evaluating the state candidate includes means for obtaining an evaluation value obtained by calculating the comparison result and the determination result;
Determining whether to accept the state candidate as a final change state based on the evaluation value,
3. The camera calibration apparatus according to claim 2, wherein internal and external parameters of the camera are estimated from the final change state.   Representing the state of the object as a set of ellipsoidal models, two-dimensional coordinate values on the image of the top of the model, the short axis of the ellipsoid that approximates the head, the short axis of the ellipsoid that approximates the body, The camera calibration apparatus according to claim 1, wherein a person is represented by a height of a model indicating height. A camera calibration method for obtaining internal and external parameters of a camera used for photographing from a photographed image of an object,
A first step in which a silhouette image creating means creates a silhouette image obtained by extracting an area in which an object is captured from a photographed image of the camera;
A second step of estimating internal and external parameters of the camera by evaluating a state predicted by the parameter estimation unit according to the change type of the object using the silhouette image;
A camera calibration method characterized by comprising: The second step includes selecting a change type of an object prepared in advance according to an arbitrary probability, and generating a state candidate according to the selected change type;
Based on the state candidate, projecting to a virtual camera having the same camera internal / external parameters as the state candidate, and generating a simulation image simulating the silhouette image;
Evaluating the candidate state from the determination result of the probability of the target state itself based on the comparison result of the silhouette image and the simulation image and prior knowledge;
The camera calibration method according to claim 5, further comprising: The step of evaluating the state candidate includes obtaining an evaluation value obtained by calculating the comparison result and the determination result;
Determining whether to accept the state candidate as a final change state based on the evaluation value,
The camera calibration method according to claim 6, wherein internal and external parameters of the camera are estimated from the final change state.   Representing the state of the object as a set of ellipsoidal models, two-dimensional coordinate values on the image of the top of the model, the short axis of the ellipsoid that approximates the head, the short axis of the ellipsoid that approximates the body, The camera calibration method according to claim 5, wherein a person is represented by a height of a model indicating height.   A camera calibration program for causing a computer to function as the camera calibration device according to claim 1.   A recording medium on which the camera calibration program according to claim 9 is recorded.

Images