Heat source target motion tracking method applied to AGV

Abstract

The invention discloses a heat source target motion tracking method applied to an AGV, and belongs to the technical field of AGVs, and the method comprises the following steps: 1, determining an initial heat source target orientation through a monocular infrared camera; 2, accurate target positioning is carried out through a laser radar; 3, modeling the motion trail of the target object; step 4, target matching: at any T + 1 moment, respectively calculating the predicted orientation and predicted distance of the target object 1, detecting all the objects through a monocular infrared camera and a laser radar to obtain the actual azimuth sequence {alpha n} and actual distance {dn} of all the objects, respectively calculating the matching degree of each object and the target object 1 at the T + 1 moment, and calculating the matching degree of each object and the target object 1 at the T + 1 moment; and the object with the matching degree Qn closest to 1 is screened out. According to the invention, the AGV can effectively track heat source targets with similar appearances under a poor light condition, especially when a plurality of moving targets with the same appearances are briefly overlapped; the method is low in system hardware resource consumption and can be implemented on a low-configuration embedded system.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV, and particularly to a method for tracking the movement of heat source targets applied to AGV. Background Art

[0002] AGV (Automated Guided Vehicle), that is, an automated guided vehicle, plays an important role and significance in industrial production in many aspects. An AGV can operate continuously for 24 hours according to a preset path, quickly and accurately complete material handling tasks, greatly reducing the time and labor costs of manual handling and improving the overall production efficiency. It can be seamlessly docked with other equipment on the production line to achieve automatic material distribution and loading and unloading, making the production process smoother and reducing the waiting time in the production process. AGV is an important part of industrial automation and intelligence, and can be integrated with other intelligent devices and systems to achieve intelligent management and control of the entire production process, promoting the development of industrial production towards intelligence and digitization.

[0003] When handling multiple moving heat source targets in a poorly lit environment, the AGV needs to have strong moving target tracking ability, that is, moving target locking ability. In current technologies, most AGVs track moving targets by machine vision. The target is identified by machine vision and the target coordinates are calculated. However, in a poorly lit environment, machine vision becomes very unreliable. Especially when multiple moving targets with the same appearance overlap briefly, it is difficult to track the original target relying on machine vision. Summary of the Invention

[0004] In order to solve the problem that an AGV cannot effectively track heat source targets with similar appearances under poor light conditions, the present invention proposes a method for tracking the movement of heat source targets applied to AGV, including the following steps:

[0005] Step 1: Determine the initial orientation of the heat source target through a monocular infrared camera.

[0006] Step 2: Accurately locate the target through a lidar.

[0007] Step 3: Model the movement trajectory of the target object.

[0008] Step 4: Target matching.

[0009] Preferably, the step 1 includes the following steps:

[0010] Step 1.1: Collect environmental images through a monocular infrared camera, identify target object 1 through an image processing algorithm, and determine the geometric center position P1(x1, y1) of target object 1 in the image.

[0011] Step 1.2: Determine the azimuth angle corresponding to P1(x1, y1) according to the known geometric model of the infrared camera Then the target object 1 must be located on the ray in the azimuth of α, and the azimuth angle of the target object 1 is α. Where c represents the width of the image in pixels; H represents the horizontal field of view angle of the camera in degrees.

[0012] Preferably, the said Step 2 includes the following steps:

[0013] Step 2.1: At time T, detect the objects within the azimuth range of α±Δe through the lidar installed directly above the infrared camera vertically to obtain the accurate azimuth angle α T and the accurate distance d T . Where Δe is the error threshold, generally taking a value of 1 - 2°.

[0014] Step 2.2: Through continuous observation by the lidar, obtain the accurate azimuth angle data of the target object 1 at the nearest consecutive N moments, forming a sequence {α T-N+1 ……α T-2、 α T-1、 α T}, and obtain the accurate distance data of the target object 1 at the nearest consecutive N moments, forming a sequence {d T-N+1 ……d T-2、 d T-1、 d T}. When the observed data exceeds N, the oldest data is automatically deleted and the new data is added. N generally takes a value between 10 and 20.

[0015] Preferably, the said Step 3 includes the following steps:

[0016] Step 3.1: Fit the accurate azimuth angle sequence {α T-N+1 ……α T-2、 α T-1、 α T} with a polynomial of degree k1 to obtain the relationship between the accurate azimuth angle α T of the target object 1 and time T, that is, α T = f1(T). Generally, it is recommended that k1 takes values of 2, 3, 4, 5.

[0017] Step 3.2: Fit the accurate distance sequence {d T-N+1 ……d T-2、 d T-1、 d T} with a polynomial of degree k2 to obtain the relationship between the accurate distance d T of the target object 1 and time T, that is, d T = f2(T). Generally, it is recommended that k2 takes values of 2, 3, 4, 5.

[0018] Preferably, step 4 includes the following steps:

[0019] Step 4.1: At any T+1 moment, calculate the predicted azimuth α' T = f1(T) and the predicted distance d T = f2(T) of the target object 1 respectively. T+1 and the predicted distance d'. T+1 .

[0020] Step 4.2: At the T+1 moment, use the monocular infrared camera and lidar to detect all objects again, and obtain the actual azimuth sequence {α n} and the actual distance sequence {d n} of all objects. n represents the serial number of the current observed object.

[0021] Step 4.3: Calculate the matching degree of each object with the target object 1 at the T+1 moment according to the formula . w1 represents the azimuth weight, and the default value is 0.5; w2 represents the distance weight, and the default value is 0.5. The matching weights of the azimuth and distance can be fine-tuned according to the principle of w1 + w2 = 1. Q n represents the matching degree of the nth detected object with the target object 1.

[0022] Step 4.4: Screen out the object with the matching degree Q n closest to 1, and this object is the target object 1, that is, the tracking of the target object 1 at the T moment is realized at the T+1 moment.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The present invention enables the AGV to effectively track heat source targets with similar appearances under poor light conditions, especially when multiple moving targets with the same appearance overlap briefly.

[0025] 2. The method of the present invention consumes low system hardware resources and can be implemented on a low-configuration embedded system. Brief Description of the Drawings

[0026] Figure 1 is the main view of the principle of the present invention.

[0027] Figure 2 is the top view of the principle of the present invention. Detailed Embodiment

[0028] The technical solution of the present invention will be further explained below in conjunction with the drawings in the specification and specific embodiments.

[0029] Such as Figure 1 , 2As shown in the figure, a heat source target motion tracking method applied to an AGV includes the following steps:

[0030] Step 1: Determine the initial heat source target orientation through a monocular infrared camera. The specific process is as follows:

[0031] Step 1.1: Collect environmental images through a monocular infrared camera, identify the target object 1 through an image processing algorithm, and determine the geometric center position P1(x1, y1) of the target object 1 in the image. In some cases with less demanding requirements, the image recognition algorithm can also be not used, and the target object 1 can be directly specified manually as the tracking object.

[0032] Step 1.2: Determine the azimuth angle corresponding to P1(x1, y1) according to the known geometric model of the infrared camera. Then the target object 1 must be located on the ray in the α azimuth, and the azimuth angle of the target object 1 is α. Here, c represents the width of the image, with the unit of pixel; H represents the horizontal field of view angle of the camera, with the unit of degree. Although the infrared camera can obtain the infrared image of the heat source object, due to its structural characteristics, the obtained azimuth angle α is not accurate and has a large error. In the present invention, the infrared camera is only used to obtain the initial position of the target object 1.

[0033] Step 2: Accurately locate the target through a lidar. The specific process is as follows:

[0034] Step 2.1: At time T, detect the objects within the azimuth range of α±Δe through the lidar installed directly above the infrared camera vertically, and obtain the accurate azimuth angle α T and the accurate distance d T . Here, Δe is the error threshold, generally taking a value of 1 - 2°. The lidar has high precision and can be used to measure the accurate azimuth angle and distance of the target object 1. The lidar is preferably installed coaxially above and below the infrared camera for convenient measurement.

[0035] Step 2.2: Through continuous observation by the lidar, obtain the accurate azimuth angle data of the target object 1 at the nearest consecutive N moments, forming a sequence {α T-N+1 ... α T-2、 α T-1、 α T}, and obtain the accurate distance data of the target object 1 at the nearest consecutive N moments, forming a sequence {d T-N+1 ... d T-2、 d T-1、 d T}. When the observed data exceeds N, the oldest data is automatically deleted and the new data is added. N generally takes a value between 10 and 20.

[0036] Step 3: Model the motion trajectory of the target object. The specific process is as follows:

[0037] Step 3.1: Fit the accurate azimuth angle sequence {α T-N+1 ……α T-2、 α T-1、 α T} with a polynomial of degree k1 to obtain the relationship between the accurate azimuth angle α T of the target object 1 and time T, that is, α T = f1(T). Generally, it is recommended that k1 take values of 2, 3, 4, 5.

[0038] Step 3.2: Fit the accurate distance sequence {d T-N+1 ……d T-2、 d T-1、 d T} with a polynomial of degree k2 to obtain the relationship between the accurate distance d T of the target object 1 and time T, that is, d T = f2(T). Generally, it is recommended that k2 take values of 2, 3, 4, 5.

[0039] Step 4: Target matching, the specific process is as follows:

[0040] Step 4.1: At any T + 1 moment, calculate the predicted azimuth α' T and predicted distance d' T of the target object 1 through α T+1 = f1(T) and d T+1 = f2(T) respectively. This value is a predicted value inferred from the model and has the motion attributes of the target object 1.

[0041] Step 4.2: At the T + 1 moment, detect all objects again through the monocular infrared camera and lidar to obtain the actual azimuth angle sequence {α n} and actual distance {d n} of all objects. n represents the current target object number.

[0042] Step 4.3: Calculate the matching degree between each object and the target object 1 at the T + 1 moment according to the formula . w1 represents the azimuth angle weight, and the default value is 0.5; w2 represents the distance weight, and the default value is 0.5. The matching weights of the azimuth angle and distance can be fine-tuned according to the principle of w1 + w2 = 1. Q n represents the matching degree between the nth object detected and the target object 1.

[0043] Step 4.4: Screen out the object with the matching degree Q n closest to 1. This object is the target object 1, that is, the tracking of the target object 1 at the T moment is achieved at the T + 1 moment.

[0044] The following data are the actual measured data in the experiment. Combining the experimental data, the implementation manner of this solution is further explained.

[0045] In the experiment, c = 800, H = 90°, P1(320, 280).

[0046] Furthermore, the value of α is calculated.

[0047]

[0048] Taking N = 10, then the sequence {α T-N+1 ……α T-2、 α T-1、 α T} = {0.479 0.840 1.011 1.259 1.212 1.505 1.894 2.182 2.141 2.371}. Using a 5th-degree polynomial fitting, we get α T = f1(T) = 1258.3T 5 - 1967.5T 4 + 1126.8T 3 - 287.09T 2 + 35.244T - 0.7104, R 2 = 0.9851,

[0049] The sequence {d T-N+1 ……d T-2、 d T-1、 d T} = {3.338 3.494 3.401 3.881 3.996 4.056 4.141 4.639 4.711 4.992}. Using a 3rd-degree polynomial for fitting, we get d T = f2(T) = 0.5246T 3 + 2.7212T 2 + 2.0541T + 3.2181, R 2 = 0.9639,

[0050] Then, based on the above results, predicting at the T + 1 moment, α' T+1 = 2.590, d' T+1 = 5.258;

[0051] If at the T + 1 moment, the measured α1 = 2.451 and d1 = 5.205 for the first object,

[0052] So

[0053] If at the T + 1 moment, the measured α2 = 2.306 and d2 = 4.855 for the second object,

[0054] So

[0055] Then Q1 is closer to 1 than Q2. Therefore, the probability that the first object is the target object 1 is higher, and it is determined that the first object is the target object 1, thereby realizing the tracking of the target object 1.

[0056] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A heat source target motion tracking method applied to an AGV, characterized in that, It includes the following steps: Step 1: Determine the azimuth of the initial heat source target through a monocular infrared camera; Step 2: Accurately locate the target through lidar; Step 3: Model the motion trajectory of the target object; Step 4: Target matching.

2. The method for tracking the movement of a heat source target applied to an AGV according to claim 1, wherein The specific steps of Step 1 include the following steps: Step 1.1: Collect environmental images through a monocular infrared camera, identify the target object 1 through an image processing algorithm, and determine the geometric center position P1(x1,y1) of the target object 1 in the image; Step 1.2: Determine the azimuth angle corresponding to P1(x1, y1) according to the known geometric model of the infrared camera Then the target object 1 must be located on the ray in the azimuth of α, and the azimuth angle of the target object 1 is α, where c represents the width of the image in pixels; H represents the horizontal field of view angle of the camera in degrees.

3. The method for tracking the movement of a heat source target applied to an AGV according to claim 1, wherein The specific steps of Step 2 include the following steps: Step 2.1: At time T, use the lidar installed directly above the infrared camera vertically to detect the objects within the azimuth range of α±Δe, and obtain the accurate azimuth angle α of the target object 1 at the current moment T and the accurate distance d T , where Δe is the error threshold with a value range of 1 - 2°; Step 2.2: Through continuous observation by the lidar, obtain the accurate azimuth angle data of the target object 1 at the most adjacent consecutive N moments, forming a sequence {α T-N+1 ……α T-2、 α T-1、 α T}, and obtain the accurate distance data of the target object 1 at the most adjacent consecutive N moments, forming a sequence {d T-N+1 ……d T-2、 d T-1、 d T}. When the observed data exceeds N, automatically delete the oldest data and add the new data. N ranges from 10 to 20.

4. A heat source target motion tracking method applied to an AGV according to claim 1, characterized in that, The specific steps of Step 3 include the following steps: Step 3.1: Fit the accurate azimuth angle sequence {α T-N+1 ……α T-2、 α T-1、 α T} with a polynomial of degree k1 to obtain the relationship between the accurate azimuth angle α T of the target object 1 and time T, that is, α T = f1(T), where k1 takes values of 2, 3, 4, 5; Step 3.2: Fit the precise distance sequence {d T-N+1 ……d T-2、 d T-1、 d T} with a k2-degree polynomial to obtain the relationship between the precise distance d T of the target object 1 and the time T, that is, d T = f2(T), where k2 takes values of 2, 3, 4, and 5.

5. The heat source target motion tracking method applied to an AGV according to claim 1, wherein The specific steps of Step 4 include the following steps: Step 4.1: At any T+1 moment, through α T = f1(T), d T = f2(T) respectively calculate the predicted azimuth α' T+1 and the predicted distance d' T+1 ; Step 4.2: At time T+1, all objects are detected again by the monocular infrared camera and the lidar, and the actual azimuth angle sequence {α n} and the actual distance {d n} of all objects are obtained, where n represents the current observation object serial number; Step 4.3: According to the formula Calculate the matching degree between each object at time T+1 and the target object 1 respectively. w1 represents the azimuth weight, with a default value of 0.5; w2 represents the distance weight, with a default value of 0.

5. According to the principle of w1 + w2 = 1, fine-tune the matching weights of the azimuth and distance, and Q n represents the matching degree between the nth detected object and the target object 1; Step 4.4: Screen out the matching degree Q n The object closest to 1, and this object is the target object 1, that is, the tracking of the target object 1 at time T is achieved at time T+1.

Images