Eye fundus examination pupil positioning system and method based on yov8 target detection

Abstract

The invention relates to the technical field of pupil positioning, in particular to a eyeground examination pupil positioning system based on yov8 target detection. The pupil detection algorithm based on the yov8, the phase detection automatic focusing algorithm and the fundus examination robot cooperate with one another, full-automatic fundus examination can be achieved, namely, fundus examination can be conducted on a patient without operation of medical staff, and even self-service fundus examination of the patient can be further achieved. A patient only needs to turn on or turn off the system by himself without other operations, then the head is placed on the head fixing frame, the fundus examination robot can move the fundus camera to conduct pupil positioning under assistance of a myov8-based pupil detection algorithm, a phase detection automatic focusing algorithm is called to conduct automatic focusing, and the front auxiliary work of fundus examination is completed; afterwards, the system controls the fundus camera to shoot, and the whole process automation from positioning to shooting inspection is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of pupil positioning, and more specifically, to a fundus examination pupil positioning system based on yolov8 object detection. Background Art

[0002] When the remote control system of a traditional fundus examination robot is used, a doctor needs to manually control the movement of the fundus camera to achieve focusing on the pupil. The doctor controls the movement of the three axes of the remote control system of the fundus examination robot to make the fundus camera reach a suitable position, and the doctor needs to frequently adjust the position for focusing, which prolongs the time used for fundus examination; and during this period, the patient's head needs to be kept stable, which brings fatigue to the patient. Summary of the Invention

[0003] The present invention provides a fundus examination pupil positioning system based on yolov8 object detection, and the purpose is to reduce the operation of fundus examination and improve the efficiency of fundus examination.

[0004] The above purpose is achieved by the following technical solutions:

[0005] A fundus examination pupil positioning system based on yolov8 object detection, characterized in that it includes a fundus examination robot with a remote control system and a computer. The computer uses a pupil detection algorithm model based on yolov8, a phase detection autofocus algorithm, and a motion control algorithm;

[0006] The fundus examination robot with a remote control system includes a fundus camera and an xyz three-axis motion mechanism for driving the displacement of the fundus camera; the fundus camera inputs image information to a pupil detection algorithm module based on yolov8 and a phase detection autofocus algorithm module; the phase detection autofocus algorithm controls the focusing of the fundus camera; the motion control algorithm converts the motion instructions of the fundus camera into control instructions for the multi-axis motion mechanism, and the pupil detection algorithm model based on yolov8 is used to control the spatial position of the fundus camera.

[0007] A fundus examination pupil positioning method based on yolov8 object detection, using the above-mentioned fundus examination pupil positioning system based on yolov8 object detection, includes the following steps:

[0008] S1. The fundus camera takes a picture;

[0009] S2. Detect the pupil position by the pupil detection algorithm based on yolov8;

[0010] S3. Judge whether the fundus camera is aligned with the pupil. If so, jump to S5; if not, jump to S4;

[0011] S4. The computer outputs an instruction to the fundus examination robot, causing the position of the fundus camera to change, and jumps to S1;

[0012] S5. The fundus camera is automatically focused through the phase detection autofocus algorithm;

[0013] S6. Perform fundus examination;

[0014] S7. When the examination is over, the system shuts down.

[0015] The beneficial effects of the fundus examination pupil positioning system and method based on yolov8 object detection of the present invention are as follows:

[0016] The pupil detection algorithm based on yolov8, the phase detection autofocus algorithm and the fundus examination robot cooperate with each other, and can realize fully automatic fundus examination, that is, the fundus examination of patients can be carried out without the operation of medical staff, and even further, the self-service fundus examination of patients can be realized. The patient only needs to turn on or off the system by himself without other operations, and then place his head on the head fixing frame. The fundus examination robot will move the fundus camera for pupil positioning with the assistance of the pupil detection algorithm based on yolov8, call the phase detection autofocus algorithm for automatic focusing, and complete the pre-assist work of fundus examination. After that, the system controls the fundus camera to take pictures, realizing the full process automation from positioning to shooting and examination.

[0017] Compared with a single fundus examination robot, this system omits the process of manual position control, and has the advantages of saving time and reducing the operation difficulty. At the same time, it reduces the pupil focusing time, thereby improving the comfort. Brief Description of the Drawings

[0018] Figure 1 It is a schematic diagram of a fundus examination pupil positioning system based on yolov8 object detection.

[0019] Figure 2 It is a flowchart of a fundus examination pupil positioning method based on yolov8 object detection. Detailed Embodiments

[0020] A pupil positioning system for a remote control system of a fundus examination robot, referring to Figure 1 , includes a computer and a fundus examination robot with a remote control system;

[0021] Among them, the computer uses a pre-deployed and trained pupil detection algorithm model based on yolov8, a phase detection autofocus algorithm and a motion control algorithm of the remote control system of the fundus examination robot;

[0022] Among them, the fundus examination robot with a remote control system includes a control end device and an operation end device;

[0023] Furthermore, the control end device includes a first control module, as well as an operation module, a display module, and a first communication module connected to the first control module;

[0024] Furthermore, the operation end device includes a second control module, as well as a drive module, a camera module, and a second communication module connected to the second control module, and a motion module connected to the drive module and the camera module; the second control module can process the input instructions and perform calculations, and convert them into a format that can be sent through the second communication module;

[0025] Furthermore, the operation module can be one or more of an operation handle, a keyboard, and a virtual reality wearable device to achieve human-machine interaction; the two control modules can use industrial control computers in the prior art.

[0026] Furthermore, the first communication module and the second communication module are connected, and the connection method is wired or wireless connection, preferably wireless communication, such as but not limited to 3G, 4G, 5G, Wi-Fi, and Bluetooth; the two communication modules are used to achieve the transmission of instructions from the control end device to the operation end device, and the image transmission from the operation end device to the control end device.

[0027] Specifically, the camera module includes a fundus camera, which is the main body for fundus examination. The camera module transmits the image information back to the second control module, and the second control module transmits the image information through the second communication module, the first communication module, and the first control module in sequence, so as to be transmitted to the display module. The display module is used to display the image transmitted back by the camera module of the operation end device, facilitating doctors to remotely observe the image;

[0028] Furthermore, the second control module is used to analyze the received motion instructions and convert them into control instructions for the drive module, so as to control the drive module;

[0029] Specifically, the drive module includes an elmo driver, and the motion module includes a multi-axis motion mechanism, preferably an xyz three-axis motion mechanism. The xyz three-axis motion mechanism can use a ball screw xyz three-axis slide table in the prior art. The motor drives the screw to rotate, and the screw drives the slide table to move linearly. At this time, the control instructions of the drive module act on the motor of the multi-axis motion mechanism to displace the slide table, thereby driving the fundus camera to move and changing the spatial position of the fundus camera. The drive module drives the motion module, the xyz three-axis motion mechanism, according to the instructions of the second control module to move the camera module to the specified position.

[0030] The fundus camera inputs image information to the pupil detection algorithm module based on YOLOv8 and the phase detection autofocus algorithm module. YOLOv8: You Only Look Once, which can be used for object detection, image classification, and instance segmentation tasks, means that it can identify the category and location of objects in the image by only browsing once. YOLO is an object detection algorithm based on the Region-free method.

[0031] The phase detection autofocus algorithm controls the focusing of the fundus camera; the pupil detection algorithm model based on YOLOv8 controls the spatial position of the fundus camera through the motion control of the remote control system;

[0032] Specifically, the pupil detection algorithm based on YOLOv8 includes three parts: the backbone extraction network (BACKBONE), the feature extraction network (NECK), and the detection head (HEAD). BACKBONE is mainly used to extract the information of the input image and provide it to NECK and HEAD, including multiple Conv, C2f modules, and SPPF. NECK adopts the structure of FPN+PAN and mainly plays the role of feature fusion. HEAD identifies and obtains the position information of the pupil according to the features processed by BACKBONE and NECK. In terms of Loss calculation, a positive and negative sample dynamic allocation strategy is adopted, and TaskAlignedAssigner is used. Its matching strategy can be simply summarized as: select positive samples according to the score weighted by the classification and regression scores. The formula is as follows:

[0033] t = s α ×μ β

[0034] Among them, s is the Cls score of all pixels - all categories, μ is the Reg score (IOU) of the predicted box of all pixels and all GT boxes, α and β are weight hyperparameters, and their product can measure the alignment degree, and t is used as the weighted score.

[0035] Yolov8 uses a method combining CIoU and DFL losses to calculate the regression loss of the bounding box. The calculation formula of CIoU is as follows:

[0036]

[0037] Among them, IoU represents the intersection ratio of the predicted box and the true box, ρ 2 (b, b gt ) represents the Euclidean distance between the predicted box and the true box; h and w respectively represent the height and width of the predicted box; w gt and h gt respectively represent the height and width of the true box; c w and ch Indicates the height and width of the smallest bounding box formed by the predicted box and the ground truth box.

[0038] Among them, a fundus examination pupil localization method based on yolov8 object detection refers to Figure 2 , including the following steps:

[0039] S1. The fundus camera takes an image;

[0040] S2. The pupil detection algorithm based on yolov8 uses the taken image to detect the pupil position;

[0041] S3. Judge whether the fundus camera is aligned with the pupil. If so, jump to S5; if not, jump to S4;

[0042] S4. The computer outputs an instruction to the remote control system, and the motion control algorithm converts the spatial motion instruction of the fundus camera into the control instruction of the xyz three-axis motion mechanism. The xyz three-axis motion mechanism adjusts the position of the fundus camera and jumps to S1;

[0043] S5. The fundus camera is automatically focused through the phase detection autofocus algorithm;

[0044] S6. Perform fundus examination;

[0045] S7. After the examination is completed, the system is shut down.

Claims

1. A fundus examination pupil positioning system based on yolov8 target detection, characterized in that: It includes a fundus examination robot and a computer with a remote control system. The computer uses a pupil detection algorithm model based on YOLOv8, a phase detection automatic focusing algorithm, and a motion control algorithm.

2. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 1, the fundus examination robot with a remote control system includes a fundus camera and an xyz three-axis motion mechanism for driving the displacement of the fundus camera.

3. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 2, the fundus camera inputs image information to the pupil detection algorithm module based on yolov8 and the phase detection automatic focusing algorithm module.

4. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 2, the phase detection autofocus algorithm controls the focus of the fundus camera.

5. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 2, the motion control algorithm converts the motion instructions of the fundus camera into control instructions of the multi-axis motion mechanism, and the pupil detection algorithm model based on yolov 8 is used to control the spatial position of the fundus camera.

6. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 1, the pupil detection algorithm model based on yolov8 includes BACKBONE, NECK and HEAD; BACKBONE is used to extract information about the input image and provide it to NECK and HEAD, including multiple Conv, C2f modules and SPPF.

7. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 6, NECK adopts the structure of FPN+PAN to play the role of feature fusion.

8. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 6, HEAD identifies and obtains the position information of the pupil based on the features obtained by BACKBONE and NECK processing.

9. According to the fundus examination pupil positioning system based on yolov8 target detection in claim 6, the pupil detection algorithm model based on yolov8 uses Loss calculation, adopts a positive and negative sample dynamic allocation strategy, and adopts TaskAlignedAssigner, and the formula is as follows: t=s α ×μ β ; in, s is the Cls score of all pixels - all categories, μ is the IOU of all pixel prediction boxes and all GT boxes, α and β are weight hyperparameters, and the multiplication of the two can measure the degree of alignment, and t is the weighted score; Yolov8 uses a combination of CIoU and DFL loss to calculate the regression loss of the bounding box. The calculation formula of CIoU is as follows: Where IoU represents the intersection ratio between the predicted box and the real box, ρ 2 (b,b gt ) represents the Euclidean distance between the predicted box and the true box; h and w represent the height and width of the predicted box respectively; w gt and h gt Represents the height and width of the real box respectively; c w and c h Indicates the height and width of the minimum bounding box composed of the predicted box and the true box.

10. A pupil positioning method for fundus examination based on yolov8 target detection, characterized in that: The fundus examination pupil positioning system based on yolov8 target detection according to any one of claims 2 to 9 comprises the following steps: S1, fundus camera captures images; S2, pupil detection algorithm based on yolov8 detects pupil position; S3, determine whether the fundus camera is aligned with the pupil, if so, jump to S5; if not, jump to S4; S4, the computer outputs instructions to the fundus inspection robot, so that the position of the fundus camera changes, and jumps to S1; S5, automatically focusing the fundus camera through a phase detection autofocus algorithm; S6. Perform fundus examination; S7. Inspection is completed and the system is shut down.

Images