Target detection method based on fusion of vision, lidar, and millimeter wave radar

1. A target detection method based on fusion of vision, lidar, and millimeter wave radar, comprising:

(1) obtaining original data detected by each of a camera, a millimeter wave radar, and a lidar, and synchronizing the millimeter wave radar, the lidar, and the camera in time and space;

(2) performing a calculation on the original data detected by the millimeter wave radar according to a radar protocol;

(3) using a position based on the original data, which is detected by the millimeter wave radar and has been calculated, as a first anchor point, and generating a first region of interest, which is three-dimensional, according to a speed and a radar reflection area with the first anchor point as a center of the first region of interest;

(4) generating a second anchor point arranged according to a specified distance in a blind area in which radar points of the millimeter wave radar are not generated, and generating a second region of interest by traversing the second anchor point with the second anchor point as a center of the second region of interest;

(5) pre-processing the original data detected by the lidar to generate a point cloud bird's-eye view, and extracting a feature map of the point cloud bird's-eye view and a feature map of the original data detected by the camera;

(6) projecting the first region of interest and the second region of interest generated in step 3 and step 5, which are three-dimensional, onto the feature map of the point cloud bird's-eye view and the feature map of the original data detected by the camera; and

(7) fusing the first region of interest, the second region of interest, the feature map of the point cloud bird's-eye view, and the feature map of the original data detected by the camera, which have a same size, to form a fused image, processing the fused image through a fully connected layer, and outputting an image of a test result.

2. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

the step 1 comprises:

triggering an exposure of the camera when the lidar sweeps across a center of a Field of View (FOV) of the camera;

distributing the exposure of the camera in scans of the lidar;

obtaining synchronized frames of the camera, the millimeter wave radar, and the lidar as key frames of the camera, the millimeter wave radar, and the lidar;

superimposing millimeter wave radar data and lidar data, which are obtained by multi-frame scanning, onto the key frames; and

processing the key frames.

3. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

in step 2, the original data detected by the millimeter wave radar after being calculated contains information of position, speed, radar reflection area, and radar point status.

4. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

the step 3 comprises:

generating the first region of interest, which is three-dimensional, based on the original data detected by the millimeter wave radar that has been calculated;

using the position as the first anchor point to generate the first region of interest that is cuboid;

using a direction of a vector velocity as a direction of the first region of interest;

determining a size of the first region of interest according to a size of a Radar Cross Section (RCS);

determining a size of a three-dimensional (3D) frame according to the following table and formula:

Target

RCS[m 2 ]

RCS[dB]

Pedestrian

0.01

−20

Car

100

20

Truck

200

23

Corner reflection

20379

43

RCS dB =10 log(RCS m 2 ); and

determining a direction and an angle of the 3D frame according to a speed (V x , V y ) of a millimeter wave radar point and compensation speed information (V x_comp, V y_comp) and according to the formula:

θ

=

tan

-

1

⁢

V

x

-

V

x

_

⁢

comp

V

y

-

V

y_comp

.

5. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

the step 4 comprises:

arranging the second anchor point at a certain interval in a blind area in which radar points of the millimeter wave radar are not generated; and

generating the second region of interest by traversing the second anchor point with the second anchor point as the center of the second region of interest,

the short-wave detection range of millimeter wave radar from 0 to 30 meters is 90 degrees,

the short-beam detection range of millimeter wave radar from 30 to 70 meters is 18 degrees,

a long-wave detection angle at 70 meters is 18 degrees,

a range of detecting obstacles is [−30, 30, 0, 70], and

a method for determining the blind area comprises:

(1) projecting the first region of interest obtained in step 3 onto bird's-eye view plane;

(2) obtaining a background area according to a projected two-dimensional candidate frame; and

(3) generating candidate frames by traversing the background area.

6. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

the step 5 comprises:

processing the original data detected by the lidar to reserve lidar points of the original data detected by the lidar selected within a range of [−0.3, 2.5] meters in a direction perpendicular to a ground plane;

equally slicing the lidar points of the original data detected by the lidar selected within the range of [−0.3, 2.5] meters into four slices;

compressing each of the four slices to form a horizontal two-dimensional image;

combining the horizontal two-dimensional image with intensity information of the lidar points to obtain the point cloud bird's-eye view in a [600 700 5] dimensionality; and

using a neural network model to extract the feature map of the point cloud bird's-eye view and the feature map of the original data detected by the camera, and

a size of the feature map is unchanged from a size of an input image.

7. The target detection method based on fusion of vision, lidar, and millimeter wave radar according to claim 1 , wherein:

the step 6 comprises:

setting an index number for each of the first anchor point and the second anchor point;

projecting the first region of interest and the second region of interest onto the feature map of the point cloud bird's-eye view and the feature map of the original data detected by the camera;

determining a three-dimensional region of interest through a spatial synchronization of the original data detected by the millimeter wave radar and the lidar in step 1 since the original data detected by the millimeter wave radar and the lidar are both three-dimensional data; and

obtaining a vertex coordinate of the three-dimensional region of interest R=[x; y; z],

a conversion relationship of the three-dimensional region of interest is:

⌈

X

p

Y

p

Z

⌉

=

P

[

R

c

T

c

O

T

1

]

[

x

y

z

1

]

,

(X p , Y p ) are coordinates of projection points in an image coordinate system,

P is a matrix of camera parameters,

Rc is a rotation matrix of the camera relative to an Inertial Measurement Unit (IMU),

Tc is a translation matrix of the camera relative to the IMU,

coordinate points of a 3D area in an image are obtained through the steps 1-7,

each of the vertex coordinates, which has been obtained, is adjusted according to a ground vector to obtained an adjusted coordinate point (X, Y, Z*),

the feature maps of the same size are fused to form the fused image,

the fused image is processed through the fully connected layer,

anchor points are filtered out, and

a size and direction of anchor point boxes are regressed.