Method and apparatus for detecting target object in blind area of vehicle
A method for detecting a target object in a blind area of a vehicle is provided by embodiments of the application. The method includes: acquiring at least one frame of an image generated by shooting a blind area of a vehicle by a shooting device; extracting an image feature of an area where at least one predetermined detection gate is located in the image; and matching the image feature of the area where the predetermined detection gate is located with a matching template of the predetermined detection gate to determine a target object. An apparatus for detecting a target object in a blind area of a vehicle is further provided by embodiments of the application.
1. A method for detecting a target object in a blind area of a vehicle, comprising:
acquiring a frame of an image generated by shooting the blind area of the vehicle by a shooting device;
extracting an image feature of an area where a predetermined detection gate is located in the image, wherein the area where the predetermined detection gate is located in the image is determined according to a shape and a size of the predetermined detection gate, a depth distance from the target object to the vehicle and a lateral distance from the target object to the vehicle; and the shape and the size of the predetermined detection gate are determined according to a type of the target object; and
matching the image feature of the area where the predetermined detection gate is located with a matching template of the predetermined detection gate, to determine the target object,
wherein a count of the number of the predetermined detection gates is 2n×m, wherein n is the number of the predetermined detection gates corresponding to the target objects which have same lateral distances and different depth distances on one side of the vehicle, and m is the number of the predetermined detection gates corresponding to the target objects which have same depth distances and different lateral distances on one side of the vehicle, wherein n and m are natural numbers larger than or equal to 1.
2. The method according to claim 1 , wherein the area where the predetermined detection gate is located comprises a complete image of the target object.
3. The method according to claim 2 , wherein the area where the predetermined detection gate is located comprising the complete image of the target object comprises: the range of the target object corresponding to the detection gate in the blind area of the vehicle meeting a requirement of:
an arctangent value of a ratio of the depth distance to the lateral distance from the target object included in the detection gate to the vehicle being equal to or less than a horizontal view angle of the shooting device, wherein the depth distance is a vertical distance from the shooting device arranged on the vehicle to a front plane of the target object, and the lateral distance is a vertical distance from a depth axis of the shooting device arranged on the vehicle to a left plane or a right plane of the target object.
4. The method according to claim 3 , wherein in a case that the shooting device arranged on the vehicle is a fisheye camera, a height of the fisheye camera is Cam_H, a Euclidean distance from the target object to the fisheye camera is D, a depth distance from the target object to the fisheye camera is VerD, a lateral distance from the target object to the fisheye camera is HonD, an actual height of the target object is Obj_3DH, an actual width of the target object is Obj_3DW, a parameter of a cylinder image converted from a fisheye image is ay, a pixel coordinate of a center of the cylinder image is (u0,v0) a pixel width of the fisheye image is imgW, a pixel height of the fisheye image is imgH, a pixel coordinate of a lower left corner of the detection gate comprising the complete image of the target object in the cylinder image is (u,v) a pixel height of the detection gate is H, and a pixel width of the detection gate is W, the detection gate is determined by the following equations:
u =Foot_col−Obj_pix W
v =Foot_row
W =Obj_pix W
H =Obj_pix H
wherein Foot_row is a row in the cylinder image corresponding to a lower edge of the front plane of the target object, which is determined by the following equation:
Foot_row
=
Cam_H
·
ay
D
+
v
0
,
wherein D=√{square root over (HonD 2 +VerD 2 )}
Foot_col is a column in which a far-end pixel of the lower edge of the front plane of the target object in the cylinder image is located, which is determined by the following equation:
Foot_col
=
u
0
+
1
d
θ
arctan
HonD
VerD
wherein dθ is a horizontal angle of each pixel in the cylinder image,
d
θ
=
φ
imgW
,
φ is a horizontal view angle;
Obj_pixH is a pixel height of the target object in the cylinder image, which is determined by the following equation:
Obj_pixH
=
Obj_
3
DH
·
ay
VerD
Obj_pixW is a pixel width of the target object in the cylinder image, which is determined by the following equation:
Obj_pixW
≈
1
d
θ
·
Obj_
3
DW
VerD
.
5. The method according to claim 3 , wherein in a case that the shooting device arranged on the vehicle is a plane camera, the arctangent value of the ratio of the depth distance to the lateral distance from the target object included in the detection gate to the vehicle being equal to or less than the horizontal view angle of the shooting device is met by meeting a requirement of:
the depth distance from the target object included in the detection gate to the vehicle being equal to or larger than
ay
·
Cam_H
imgH
-
v
0
,
and the lateral distance from the target object included in the detection gate to the vehicle being equal to or less than
ay
·
(
imgW
-
u
0
)
·
Cam_H
ax
·
(
imgH
-
v
0
)
,
wherein Cam_H is a mounting height of the plane camera, ax, ay are inter parameters of the plane camera, and ax=f/dx, ay=f/dy, f is a focal length of the plane camera, dx is a minimum physical size of a pixel of a plane image in a horizontal direction, dy is a minimum physical size of a pixel of the plane image in a vertical direction, imgW is a pixel width of the plane image, imgH is a pixel height of the plane image, and (u0,v0) is a pixel coordinate of a center of the plane image.
6. The method according to claim 5 , wherein in a case that a pixel coordinate of a lower left corner of the detection gate comprising the complete image of the target object in the plane image is (u,v), a pixel height of the detection gate is H, a pixel width of the detection gate is W, the depth distance of the target object included in the detection gate is VerD, the lateral distance of the target object included in the detection gate is HonD, an actual height of the target object is Obj_3DH, and an actual width of the target object is Obj_3DW, the detection gate is determined by the following equations:
u =Foot_col−Obj_pix W
v =Foot_row
W =Obj_pix W
H =Obj_pix H
wherein Foot_row is a row in the plane image corresponding to a lower edge of the front plane of the target object, which is determined by the following equation:
Foot_row
=
v
0
+
Cam_H
·
ay
VerD
Foot_col is a column in which a far-end pixel of the lower edge of the front plane of the target object in the plane image is located, which is determined by the following equation:
Foot_col
=
u
0
+
HonD
·
ax
VerD
Obj_pixH is a pixel height of the target object in the plane image, which is determined by the following equation:
Obj_pixH
=
Obj_
3
DH
·
ay
VerD
Obj_pixW is a pixel width of the target object in the plane image, which is determined by the following equation:
Obj_pixW
=
Obj_
3
DW
·
ax
VerD
.
7. The method according to claim 1 , wherein differences between (1) the lateral distances from the target objects having the same depth distances included in the predetermined detection gates to the vehicle and (2) a lane width, fit a normal distribution.
8. The method according to claim 1 , wherein
matching templates of the detection gates corresponding to the target objects which have the same depth distances are the same, or
matching the image feature of the area where the predetermined detection gate is located with the matching template of the predetermined detection gate to determine the target object comprises: matching the image feature of the area where the predetermined detection gate is located with the matching template of the predetermined detection gate; determining a matching degree between the matching template of the predetermined detection gate and the image feature of the area where the predetermined detection gate is located; judging whether the matching degree is larger than a predetermined judging threshold; and determining that the target object corresponding to the detection gate exists if the matching degree is larger than the predetermined judging threshold.