WO2015083875A1

WO2015083875A1 - Method and mobile system for estimating camera location through generation and selection of particle

Info

Publication number: WO2015083875A1
Application number: PCT/KR2013/012102
Authority: WO
Inventors: 김정호; 최병호; 황영배; 배주한
Original assignee: 전자부품연구원
Priority date: 2013-12-03
Filing date: 2013-12-24
Publication date: 2015-06-11
Also published as: KR101483549B1

Abstract

A method and a mobile system for estimating a camera location through the generation and selection of a particle are provided. The camera location estimation method according to an embodiment of the present invention estimates a location by adaptively using a global location estimation technique and a sequential location estimation technique on the basis of reliability. Therefore, the accuracy of a current location can be guaranteed without significantly reducing the estimation speed. The global location estimation is performed again when the reliability of the sequential location estimation is reduced, and thus an error accumulation problem, which may occur in the process of the sequential location estimation, can be solved.

Description

Camera Position Estimation Method and Moving System by Particle Generation and Screening

The present invention relates to camera position estimation, and more particularly, to a method for calculating the position of a mobile system such as a camera or a robot or a vehicle equipped with a camera.

The probabilistic filtering technique for camera position estimation is based on a particle filter that predicts a current camera pose from a motion model and newly updates a predicted position from data.

In the case of particle filters, the way in which particles are generated dictates the performance of the camera's position estimation. A large number of particles selects only one particle because the accuracy is improved while the processing speed is slowed down.

Although the treatment speed increases with this type of particle filter, since only one particle is sorted, there is a problem that the accuracy is very low when an error is accumulated.

The present invention has been made to solve the above problems, and an object of the present invention is to provide a method for estimating camera position through particle generation and screening that can ensure the accuracy of the current position without seriously reducing the processing speed and It is to provide a mobile system applying this.

A camera position estimation method according to an embodiment of the present invention for achieving the above object, the first estimation step of estimating the position of the camera by a global position estimation technique; A second estimating step of estimating the position of the camera from the position estimated in the first estimating step by a sequential position estimating technique; Calculating a reliability of the position estimated in the second estimating step; And re-performing from the second estimating step if the reliability is greater than or equal to the reference in the calculating step.

In addition, the camera position estimation method according to an embodiment of the present invention, if the reliability is less than the reference in the calculation step, the step of performing again from the first estimation step; may further include.

In addition, the first estimating step may estimate the position of the camera by using corresponding points found by feature point matching between the previous image and the current image.

The second estimating step may further include generating motion particles of the camera; And selecting a predetermined number of moving particles among the moving particles.

In addition, the selecting step may include calculating weights of the moving particles; And selecting the predetermined number of moving particles with reference to the weights.

The weight may be a number in which the absolute value of the difference between the observation in the current image and the corresponding observation in the previous image is less than the reference.

In the calculating step, the sum of the weights may be calculated as the reliability.

On the other hand, according to another embodiment of the present invention, a mobile system, a camera for generating an image by shooting; And estimating the position of the camera using a global position estimation technique, estimating the position of the camera using a sequential position estimation technique from the estimated position, and if the reliability of the estimated position is higher than a reference, the sequential position. And a processor that estimates the position of the camera again using an estimation technique.

As described above, according to embodiments of the present invention, the global position estimation technique and the sequential position estimation technique are adaptively used based on the reliability to ensure the accuracy of the current position without seriously reducing the position estimation speed. It becomes possible.

In particular, when the reliability of the sequential position estimation is low, the global position estimation is performed again, thereby solving the error accumulation problem that may occur in the sequential position estimation process.

In addition, even when a sudden change in motion or obstruction occurs, accurate position estimation is possible based on the reliability.

1 is a flowchart provided for explaining a camera position estimation method according to an embodiment of the present invention;

2 is a diagram illustrating a camera motion particle generation result by the RANSAC algorithm,

3 is a view illustrating a camera position estimation result according to the method illustrated in FIG. 1, and

4 is a block diagram of a robot system according to another embodiment of the present invention.

Hereinafter, with reference to the drawings will be described the present invention in more detail.

1 is a flowchart provided for explaining a camera position estimation method according to an embodiment of the present invention. The camera position estimation method according to the present embodiment selectively performs a global position estimation technique and a sequential position tracking technique in order to prevent errors from accumulating in estimating the position of the camera from an image.

By the camera position estimation method according to the present embodiment, the accuracy of the camera position estimation is improved, which will be described in detail below.

As shown in FIG. 1, first, a position of a camera is estimated by using a global position estimation technique (S110). The global position estimation technique in step S110 is a technique of estimating the movement of the camera using the corresponding points found by the feature point matching between the previous image and the current image.

The global position estimation technique in step S110 may utilize other types of global position estimation techniques, such as based on color matching, in addition to feature point matching.

Then, the position of the camera is estimated from the position estimated in step S110 by the image-based sequential position estimation technique (S120 to S140).

For sequential position estimation, first, camera motion particles are generated from the estimated position (S120). Camera movement particles include camera rotation and translation.

The camera motion particle generation in step S120 may be performed by applying a sample acquisition method by a random SAmple consensus (RANSAC) algorithm.

The left side of FIG. 2 illustrates the previous image and the current image, the center of FIG. 2 illustrates the distribution of particles by three-dimensional movement of the camera, and the right side of FIG. 2 illustrates the distribution of particles by three-dimensional rotation of the camera. It was.

Next, a weight is calculated for the camera motion particles generated in step S120 (S130). The weight for the camera moving particle can be calculated by the following equation (1).

Equation 1

here

Is the observation in the current image, m _l is the

Three-dimensional coordinates from the previous image corresponding to

Is the i-th camera motion particle, σ _l is the degree of ambiguity in the image.

Through Equation 1, the weight may be regarded as a number in which the absolute value of the difference between the observation in the current image and the observation in the previous image corresponding thereto is less than the reference.

When the weights are calculated for all camera motion particles in step S130, only some of the camera motion particles generated in step S120 are selected based on the calculated weights (S140).

That is, step S140 can be seen to function as a particle filter in the camera position estimation method according to the present embodiment. It should be noted that the number of camera motion particles selected in step S140 is not one but a plurality, and fixed to N.

Next, the reliability of the N camera motion particles selected in step S140 is calculated (S150). Reliability C in step S150 is calculated as the sum of the selected N camera motion particles, as shown in Equation 2 below.

Equation 2

This can be seen as evaluating the reliability of the particle filter using marginal likelihood.

When the reliability C calculated in step S150 is equal to or greater than the threshold Th (S160-Y), sequential position estimation is continued for the next image of the current image (S120 to S140).

This is because the reliability (C) calculated in step S150 is higher than or equal to the threshold value (Th). The threshold Th may be set to a desired value, or may be set to be adaptively changed according to needs and circumstances.

Sequential position estimation is continued, and the camera motion particles are regenerated from the camera motion particles selected in step S130 (S120), the weights are recalculated for the generated camera motion particles (S130), and based on the calculated weights. In step S120, only some of the camera motion particles generated again are selected again (S140).

On the other hand, when the reliability (C) calculated in the step S150 is less than the threshold (Th) (S160-N), the next image of the current image is performed again from step S110. That is, after estimating the position of the camera again by global position estimation, sequential position estimation is performed. This can be seen as initializing the position estimate and performing a new one again.

When the reliability (C) calculated in step S150 is less than the threshold (Th), the reliability of the sequential position estimation result is low, and the error may accumulate to prevent this.

3 illustrates a camera position estimation result according to the method illustrated in FIG. 1. In FIG. 3, the red line is the actual moving path of the camera, and the blue line is the camera position estimation path. Meanwhile, the dark spots shown in FIG. 3 have low reliability, and thus are positions that are restarted from the global position estimation.

4 is a block diagram of a robot system according to another embodiment of the present invention. As shown in FIG. 4, the robot system 200 according to the present exemplary embodiment includes a camera 210, a communication unit 220, a processor 230, a driver 240, and a storage unit 250.

The camera 210 is a means for generating an image through photographing, but may be implemented as a camera having six degrees of freedom, but there is no limitation on the type and location thereof. The driver 240 is a means for performing the movement of the robot system 200 and other functions.

The processor 230 estimates the position of the camera position 210, that is, the position of the robot system 200, by executing the position estimation algorithm illustrated in FIG. 1 based on the image generated by the camera 210.

The communicator 220 is a means for wirelessly communicating with the outside, and may transmit an image captured by the camera 210 and location information estimated by the processor 230 as well as other information to the outside.

The storage unit 250 stores an image generated by the camera 210, position information estimated by the processor 230, and the like, and a position estimation algorithm illustrated in FIG. 2 is stored as a program.

The robot system 200 shown in FIG. 4 is just an example to which the position estimation method shown in FIG. 1 is applicable. The position estimation method shown in FIG. 1 may be applied to other movable systems (eg, vehicles) in addition to the robot system 200 shown in FIG. 4.

In addition, although the preferred embodiment of the present invention has been shown and described above, the present invention is not limited to the specific embodiments described above, but the technical field to which the invention belongs without departing from the spirit of the invention claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be individually understood from the technical spirit or the prospect of the present invention.

Claims

A first estimating step of estimating the position of the camera by a global position estimation technique;

A second estimating step of estimating the position of the camera from the position estimated in the first estimating step by a sequential position estimating technique;

Calculating a reliability of the position estimated in the second estimating step; And

And re-performing from the second estimating step if the reliability is equal to or greater than a reference value in the calculating step.
The method of claim 1,

And if the reliability is less than the reference in the calculating step, performing the process from the first estimation step again.
The method of claim 1,

The first estimating step,

And estimating the position of the camera using the corresponding points found by matching the feature point between the previous image and the current image.
The method of claim 1,

The second estimating step,

Generating moving particles of the camera; And

And selecting a predetermined number of moving particles from among the moving particles.
The method of claim 4, wherein

The screening step,

Calculating weights of the moving particles; And

And selecting a predetermined number of moving particles with reference to the weights.
The method of claim 5,

The weight is,

And the absolute value of the difference between the observations in the current image and the corresponding observations in the previous image is less than a reference.
The method of claim 5,

The calculating step,

And calculating the sum of the weights as the reliability.
A camera for generating an image through shooting; And

Using the image, the position of the camera is estimated by a global position estimation technique, the position of the camera is estimated by the sequential position estimation technique from the estimated position, and the sequential position estimation when the reliability of the estimated position is higher than a reference. A processor for estimating the position of the camera again with a technique.