CN105844700A - System for acquiring three-dimensional point clouds in outdoor scene - Google Patents

Abstract

The invention discloses a system for acquiring three-dimensional point clouds in an outdoor scene based on a laser radar, and belongs to the field of computer vision. The invention solves the problem of how to solve for dimensions in the single-line radar outdoor scene measurement by using a single-line radar 102 and a camera 103, conducting synchronous acquisition via a synchronous model 104, and by splicing radar data using the movement path of the camera in a computer 101. The core of the primary algorithm in the invention lies in that the point clouds formed by images and the radar are used for point cloud fusion to solve for the dimensions iteratively. The method provided by the invention can attain a relatively accurate outdoor scene three-dimensional model in a simple and cheap way.

Description

A kind of outdoor scene three-dimensional point cloud acquisition system

Technical field

The present invention relates to a kind of outdoor scene three-dimensional point cloud acquisition system, relate to computer vision field outdoor scene three-dimensional Point cloud acquisition.

Background technology

Being paid attention in each field along with 3D technology and play, outdoor scene 3D rebuilds and becomes the heat into vision research field Point.High-quality outdoor scene 3D model has a wide range of applications at computer vision field.

Existing outdoor scene three-dimensional point cloud acquisition system substantially has two kinds: 1, utilize binocular camera, by characteristic point The mode joined, it is thus achieved that depth information.2, utilizing laser radar, carry camera, the equipment such as GPS, IMU, to outdoor scene collection point Cloud.First method precision is the highest.Second method majority uses face battle array laser radar, and obtains movement locus by GPS, Relatively costly.

Summary of the invention

The invention discloses a kind of outdoor scene three-dimensional point cloud acquisition system, its object is to solve in whole or in part one Individual or multiple the problems referred to above or shortcoming, to provide at least advantages described below.

The conventional system of ratio, only need to use a common camera and single line laser radar, and system composition is fairly simple, Need not use the equipment such as GPS, IMU, cost is relatively low.

Need not too much manual operation, make full use of cloud data, some cloud precision is higher.

In order to entirely or partially realize above-mentioned purpose, outdoor scene three-dimensional point cloud acquisition system disclosed by the invention, its It is characterised by comprising the following steps:

(1) synchronization module, function is that the lock-out pulse blood pressure lowering realizing sending laser radar divides and transmits to camera, makes Camera carry out hardware level with laser radar synchronization, be built into acquisition system.

(2) outdoor collection, obtains multiple image data and the radar data of measured target.

(3) view data is carried out points cloud processing, obtain the scale free three-dimensional point cloud of target and every two field picture spatial information.

(4) by the every two field picture spatial information splicing radar data obtained, another block scale free obtaining target is three-dimensional Point cloud.

(5) two pieces of different three-dimensional point clouds carry out a cloud to merge, try to achieve the concrete yardstick of target.

(6) yardstick calculated is brought in the some cloud after fusion, obtain the high-quality outdoor scene having concrete size Three-dimensional point cloud.

Accompanying drawing explanation

Technical solution of the present invention is further appreciated by by accompanying drawing explanation for offer, and constitutes a part for description, with The enforcement of the present invention is used for explaining technical scheme together, is not intended that the restriction to technical solution of the present invention.Accompanying drawing It is described as follows:

Fig. 1 is the hardware composition diagram of outdoor scene three-dimensional point cloud acquisition system.

Fig. 2 is flow chart of data processing figure.

Detailed description of the invention

Embodiments of the present invention are described in detail, whereby to the present invention how application technology means below with reference to accompanying drawing Solve problem, and the process that realizes reaching technique effect can fully understand and implement according to this.

Acquisition system.Fig. 1 is the hardware composition of acquisition system, and wherein 101 parts are computer, and 102 parts are that single line swashs Optical radar, 103 parts are camera, and 104 parts are synchronization module.

Flow chart of data processing.This flow process mainly includes step: 301, data acquisition；302, image real time transfer；303, thunder Reach data to process；304 clouds merge.

301, data acquisition: mobile collection system, carries out data acquisition to the measured target in outdoor scene around a circle, To view data and radar data, preserve in a computer.Image data acquiring and radar data acquisition are Tong Bu carried out.

302, image real time transfer: utilize the general method for solving of PnP problem, solve camera coordinates system and radar fix Rotation translation matrix [R between system_s|t_s].Utilize from moving to structure (Struct From Motion) algorithm, to collecting A series of color image datas carry out three-dimensional reconstruction, obtain a three-dimensional point not having dimensional information under world coordinate system Cloud P_c=(x_c, y_c, z_c), the image coordinate simultaneously obtaining each two field picture is tied to the rotation translation matrix [R of world coordinate system_j| t_j], wherein R_jFor the rotation of jth two field picture, t_jFor translation, this is a scalar.Assume that scale factor is λ₀, bring P into_cIn obtain Point cloud P_c=λ₀·(x_c, y_c, z_c)。

303, radar data processes: the radar data collected can be expressed asWhereinFor the i-th point in jth frame radar data under radar fix system Three-dimensional coordinate,For the depth value of i-th point, θ_iAngle formed by i-th point to radar center and horizontal plane.By radar number It is transformed in camera coordinates system according to being moved by rotary flat, L_j=R_s·l_j+t_s.Due to every frame radar data and every frame image data It is relation one to one, therefore, utilizes the rotation translation transformation [R of every two field picture_j|t_j], by radar data splicing in unification Under world coordinate system, and scale factor λ to be entered₀, obtain a cloud P_L=R_jL_j+λ₀·t_j。

304, some cloud merges: obtain two pieces of some clouds under world coordinate system in 301, P_cAnd P_L.At P_cMiddle search P_L's Closest approach, total n, P_L(n)→P_C(n), then iterative computation

$\arg \min \left(\underset{k=1}{\overset{n}{\Σ}}{||R_{\mathrm{\λ}}{P}_{L\left(k\right)}+T_{\mathrm{\λ}}-{\mathrm{\λP}}_{C\left(k\right)}||}^2\right)$

Wherein R_λFor P_cAnd P_LBetween rotation, due to camera and radar surveying is same under world coordinate system Object, so after by a series of rotation translation transformations, P_cAnd P_LBetween not do not rotate, R_λFor unit matrix.By P_cAnd P_LSubstitute into Obtain formula

$\arg min\left(\underset{k=1}{\overset{n}{\Σ}}{||R_{\mathrm{\λ}}(R_{j\left(k\right)}{L}_k+\mathrm{\λ}\·t_{j\left(k\right)})+T_{\mathrm{\λ}}-\mathrm{\λ}\·P_{C\left(k\right)}||}^2\right)$

And if only if T_λWhen=0, minima could be obtained,

Obtain formula

$\arg min\underset{k=1}{\overset{n}{\Σ}}{||R_{\mathrm{\λ}}(R_{pk}+\mathrm{\λ}\·t_k)-\mathrm{\λ}\·m_k||}^2$

Wherein

λ derivation is obtained

$\mathrm{\λ}=\frac{\underset{k=1}{\overset{n}{\Σ}}({t_k}^{,}\·R_{pk}-m_k^{,}\·R_{\mathrm{\λ}}\·R_{pk})}{\underset{k=1}{\overset{n}{\Σ}}(2\·m_k^{,}\·R_{\mathrm{\λ}}\·t_k-m_k^{,}\·m_k-t_k^{,}\·t_k)}$

The λ substitute point cloud P that will solve_cAnd P_LMiddle repeat the above steps, solves repeatedly until λ no longer changes or reaches Maximum iteration time, solves final scale factor lambda.

Those skilled in the art should be understood that the system structure of the above-mentioned present invention and each step can be with general Calculating device to realize, they can concentrate on single calculating device, or is distributed in the net of multiple calculating device composition On network, alternatively, they can realize, it is thus possible to be stored in depositing with calculating the executable program code of device Storage device is performed by calculating device, or they are fabricated to respectively each integrated circuit modules, or by them Multiple modules or step are fabricated to single integrated circuit module and realize.So, the present invention is not restricted to any specific hardware Combine with software.

Although the present invention is shown with the embodiment described as above, but described content is only to facilitate understand this The embodiment invented and use, is not limited to the present invention.Technical staff in any the technical field of the invention, On the premise of the spirit and scope that disclosed herein, in form and any repairing can be done in details implement Change and change, but the scope of patent protection of the present invention, still must be defined in the range of standard with appending claims.

Claims (2)

1. an outdoor scene three-dimensional point cloud acquisition system, this system includes 101, computer；102, radar；103, camera； 104, synchronization module.It is characterized in that comprising following key step: 301, data acquisition；302, image real time transfer；303, radar Data process；304, some cloud merges.

2. described in claim 1, the step 304 in method is characterised by: obtain two pieces in 301 under world coordinate system Some cloud, P_CAnd P_L。

At P_CMiddle search P_LClosest approach, total n, P_L(n)→P_C(n), then iterative computation minimizes cost function:

$\arg \min \left(\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\right||R_{\mathrm{\λ}}{P}_{L\left(k\right)}+T_{\mathrm{\λ}}-{\mathrm{\λP}}_{C\left(k\right)}|{|}^2),$

Wherein R_λFor P_CAnd P_LBetween rotation, R_λFor unit matrix, launch cost function, solve scale factor.

Obtain scale factor solved function:

$\mathrm{\λ}=\frac{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\left({t_k}^{,}\·R_{pk}-m_k^{,}\·R_{\mathrm{\λ}}\·R_{pk}\right)}{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}\left(2\·m_k^{,}\·R_{\mathrm{\λ}}\·t_k-m_k^{,}\·m_k-t_k^{,}\·t_k\right)}$

Wherein

The λ substitute point cloud P that will solve_CAnd P_LMiddle repeat the above steps, solves repeatedly until λ no longer changes or reaches maximum Iterations, solves final scale factor lambda.