Dynamic tracking and feedback control method for detecting focus by endoscope

Through dynamic tracking and feedback control methods, the kernel-related filtering algorithm and micro motor automatically adjust the front-end angle of the endoscope, solving the cumbersome and safety problems of traditional endoscopy detection lesions, and improving the accuracy and operation efficiency of diagnosis .

CN119073864AInactive Publication Date: 2024-12-06GUANGDONG UNIV OF TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411007629.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-12-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

When detecting lesions, traditional endoscopy causes the lesions to disappear due to foreign body sensation and internal physiological structure obstruction. The doctor needs to manually adjust it. The cumbersome and improper operation may lead to tissue damage, reduce diagnostic safety, and long-term operation will lead to difficulty in hand-eye coordination.

Method used

Dynamic tracking and feedback control methods are used to manually label target lesions, and a tracking model is established based on nuclear-related filtering algorithms. The position of the lesions is tracked in real time, the displacement error is calculated and the orientation of the front end of the endoscope is adjusted. The micro motor is used to automatically adjust the angle to keep the lesions in the image. center.

Benefits of technology

It improves robustness and operational ease in complex environments, reduces the visual and physical burden of doctors, significantly improves diagnostic accuracy and operational efficiency, and reduces the need for manual adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119073864A_ABST
    Figure CN119073864A_ABST
Patent Text Reader

Abstract

The invention provides a dynamic tracking and feedback control method for endoscope detection of a focus, and the method comprises the steps: carrying out the manual marking of the focus, and obtaining a target focus; establishing a tracking model based on a kernel correlation filtering algorithm, and tracking the target focus in real time through the tracking model to obtain position information of the target focus; acquiring a current transmission image transmitted by the endoscope, and calculating an error between the central point of the target focus position information and the central point of the current transmission image to obtain a displacement error; and adjusting the orientation of the front end of the endoscope according to the displacement error. Compared with the prior art, the focus position is efficiently tracked through the correlation filter, target tracking can be accurately completed even if the focus moves within a certain range of the view or has shape change, and the attention burden of doctors in the operation process is greatly relieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention is applicable to the field of medical technology, and in particular relates to a dynamic tracking and feedback control method for endoscopic lesion detection. Background Art

[0002] Medical endoscope is a widely used medical instrument. When inserted into the human body, doctors can directly observe the tissue morphology of internal organs and the changes of lesions in the body cavity, which is convenient for doctors to make accurate diagnosis of diseases. It plays an important role in medical diagnosis. With the continuous advancement of medical technology, the intelligence of medical endoscopes is getting higher and higher. Endoscopes can not only reach deep into the patient's body and act as the "eyes" inside the human body, but also intelligently identify lesions in different parts.

[0003] When doctors use endoscopes that can intelligently identify lesions to diagnose patients, the patient may move slightly or the internal physiological structure may be blocked due to the foreign body sensation of the endoscope entering the body. The image of the identified lesion may temporarily disappear from the display screen due to the above reasons. With traditional endoscopes, doctors need to first reconfirm the position of the front end of the endoscope inside the patient's body, and then manually adjust the front end of the endoscope to find the target lesion based on the previously identified lesion image. However, manually fine-tuning the positioning of the patient's target lesion will rely heavily on the doctor's experience, and the operation is quite cumbersome. Local tissue damage or inflammation caused by improper operation will cause more discomfort or even pain to the patient, reducing the safety of the diagnosis. At the same time, if the front end of the endoscope is manually adjusted, the doctor's energy will decrease during the long diagnosis process, and hand-eye coordination will be difficult.

[0004] Therefore, there is an urgent need for a new dynamic tracking and feedback control method for endoscopic lesion detection to solve the above problems. Summary of the invention

[0005] The present invention provides a dynamic tracking and feedback control method for endoscopic detection of lesions, aiming to improve the robustness of endoscopic detection of lesions in the complex environment inside the human body, enhance the convenience of using the endoscope, and reduce the visual and physical burden on doctors when holding the endoscope for a long time to detect lesions.

[0006] The present invention proposes a dynamic tracking and feedback control method for endoscopic lesion detection, and the dynamic tracking and feedback control method comprises the following steps:

[0007] S1. Manually mark the lesions to obtain the target lesions;

[0008] S2, establishing a tracking model based on a kernel correlation filter algorithm, and tracking the target lesion in real time through the tracking model to obtain the target lesion position information;

[0009] S3, obtaining a current transmission image transmitted by the endoscope, calculating the error between the center point of the target lesion position information and the center point of the current transmission image, and obtaining a displacement error;

[0010] S4. Adjust the front end direction of the endoscope according to the displacement error.

[0011] Preferably, step S2 includes the following sub-steps:

[0012] S21, determining whether the target lesion is continuously blocked within a preset time during the process of tracking the target lesion, if so, proceeding to step S22; if not, continuing to track the target lesion;

[0013] S22, obtaining the current predicted position information of the target lesion by the tracking model, and determining whether the decrease amount of the confidence corresponding to the current predicted position information within the preset time exceeds a preset threshold value, if so, proceeding to step S23; if not, continuing to track the target lesion;

[0014] S23, defining the time when the decrease in confidence corresponding to the current predicted position information exceeds the preset threshold as t frames, the previous frame when the target lesion is blocked as the nth frame, taking the target lesion position predicted by the tracking model at the n+tth frame as the starting point, starting reverse tracking to the target lesion position predicted by the tracking model at the nth frame, and calculating the confidence score of the tracking model after reverse tracking according to a preset rule, and proceeding to step S24;

[0015] S24, determining whether the confidence score exceeds a preset score threshold, if so, continuing to track the target lesion; if not, updating the tracking model and re-tracking the target lesion.

[0016] Preferably, the preset rule is to convert the Euclidean distance between the target lesion position predicted by the tracking model at the n+tth frame and the target lesion position predicted by the tracking model at the nth frame into the confidence score, which satisfies:

[0017] C n =e -α·d

[0018] Among them, α represents the adjustment coefficient, d represents the Euclidean distance, C n represents the confidence score, and e represents the base of the natural logarithm.

[0019] Preferably, in step S4, the adjustment size of the front end direction of the endoscope on the x-axis and the y-axis satisfies the following relationship respectively:

[0020] Xd =α1θ1 2 +b1θ1+c1;

[0021] Y d =α2θ2 2 +b2θ2+c2;

[0022] Among them, X d represents the displacement error of the endoscope on the x-axis, Y d represents the displacement error of the endoscope on the y-axis, b1 and b2 represent the influence of the angles of the front end of the endoscope in the x-axis direction and the y-axis direction on the displacement error respectively; c1 and c2 represent the initial error or calibration error of the angle adjustment of the front end of the endoscope in the x-axis direction and the y-axis direction respectively; α1 and α2 represent the nonlinear effect of the angle deviation on the displacement error; θ1 and θ2 represent the angles of the front end of the endoscope in the x-axis direction and the y-axis direction respectively.

[0023] Preferably, in step S4, the front end direction of the endoscope is adjusted by a micro motor.

[0024] Compared with the prior art, the present invention obtains the target lesion by manually marking the lesion; establishes a tracking model based on the kernel correlation filter algorithm, tracks the target lesion in real time through the tracking model, and obtains the position information of the target lesion; obtains the current transmission image transmitted by the endoscope, calculates the error between the center point of the target lesion position information and the center point of the current transmission image, and obtains the displacement error; adjusts the front end direction of the endoscope according to the displacement error. The present invention efficiently tracks the position of the lesion through the correlation filter, and can accurately complete the target tracking even if the lesion moves or changes in shape within a certain range of the visual field, greatly reducing the attention burden of the doctor during the operation; realizes efficient and continuous tracking of the lesion, and can still accurately lock the lesion even when the lesion moves or changes in shape due to the patient's movement or shape. This ability significantly reduces the need for manual adjustment, thereby improving the doctor's operating efficiency and diagnostic accuracy for endoscopic examinations; the front end angle of the endoscope is automatically adjusted by a micromotor to ensure that the target lesion is always in the center of the image, greatly improving the convenience of surgery and diagnosis. This mechanism is specifically designed for doctors who hold an endoscope for a long time. It reduces their visual and physical burden and allows them to focus more on the diagnosis and treatment process during complex or lengthy examinations and operations, thereby improving work efficiency and treatment outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be described in detail below in conjunction with the accompanying drawings. The above and other aspects of the present invention will become clearer and easier to understand through the detailed description made in conjunction with the following drawings. In the accompanying drawings:

[0026] Figure 1 It is a flowchart of a dynamic tracking and feedback control method for endoscopic lesion detection provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0028] Please refer to Figure 1 The present invention provides a dynamic tracking and feedback control method for endoscopic lesion detection, the dynamic tracking and feedback control method comprising the following steps:

[0029] S1. Manually mark the lesions to obtain the target lesions;

[0030] S2, establishing a tracking model based on a kernel correlation filter algorithm, and tracking the target lesion in real time through the tracking model to obtain the target lesion position information;

[0031] In an embodiment of the present invention, a kernelized correlation filter (KCF) algorithm is used. The KCF algorithm is a kernel correlation filter tracking algorithm. Target tracking can find the target position in subsequent video frame images and can be regarded as a ridge regression problem. First, fast depth segmentation is performed, and the depth information of the target area is clustered using the K-means algorithm to find the depth distribution of the target position. Then, the target area features are extracted and fused to describe the initial features of the target. Suppose the training sample is (x i ,y i ),

[0032]

[0033] x i is the training sample, y i is the label corresponding to the sample, and λ is the regularization coefficient to prevent overfitting during training. The partial derivative of w is taken and set to 0; the analytical solution to the ridge regression problem is obtained:

[0034] w=(X H X+λI) -1 X H y;

[0035] In order to separate the background from the target, it is necessary to map the sample from the low-dimensional space to a linearly separable high-dimensional space through a nonlinear mapping. The weight vector can be expressed as

[0036]

[0037] Then, in order to transform the nonlinear problem into a linear problem, the kernel function is introduced to simplify the problem as follows:

[0038] a=(k+λI) -1 y;

[0039] Where k is the kernel correlation of the two matrices, expressed as follows:

[0040]

[0041] Due to the existence of the inversion operation, the diagonalization property of the circulant matrix is ​​used to calculate the above problem in the frequency domain, and we get

[0042]

[0043] in All are calculation results in the frequency domain. Represents the kernel correlation of samples in adjacent frames, and the calculation result determines the predicted target position of the next frame.

[0044] In order to obtain a higher tracking speed, the training sample is obtained by cyclically shifting the initial target sample, and a cyclic kernel correlation matrix can be obtained. For the test sample, the corresponding response output is as follows:

[0045]

[0046] Where ⊙ is the dot product symbol, which means the corresponding elements in the matrix are multiplied. The target prediction position of the tracking algorithm can be obtained by the matrix The inverse Fourier transform is used to find the corresponding maximum response value.

[0047] In the embodiment of the present invention, step S2 includes the following sub-steps:

[0048] S21, determining whether the target lesion is continuously blocked within a preset time during the process of tracking the target lesion, if so, proceeding to step S22; if not, continuing to track the target lesion;

[0049] S22, obtaining the current predicted position information of the target lesion by the tracking model, and determining whether the decrease amount of the confidence corresponding to the current predicted position information within the preset time exceeds a preset threshold value, if so, proceeding to step S23; if not, continuing to track the target lesion;

[0050] S23, defining the time when the decrease in confidence corresponding to the current predicted position information exceeds the preset threshold as t frames, the previous frame when the target lesion is blocked as the nth frame, taking the target lesion position predicted by the tracking model at the n+tth frame as the starting point, starting reverse tracking to the target lesion position predicted by the tracking model at the nth frame, and calculating the confidence score of the tracking model after reverse tracking according to a preset rule, and proceeding to step S24;

[0051] Specifically, if the target lesion is blocked for several consecutive frames during the lesion tracking process, the target position prediction value, i.e., the matrix If the maximum response value of is lower than a given threshold, the predicted target lesion position becomes unreliable. In this case, a reverse tracking method is added: At the beginning of reverse tracking, the confidence score of the first frame before being blocked is recorded as a benchmark, including the following methods:

[0052] Frame-by-frame comparison: The target lesion starts from the frame where it is occluded. For each new frame, the confidence score of the current frame is calculated and compared with the score of the previous frame.

[0053] Calculate the decrease in confidence: The decrease in confidence refers to the change in the confidence score from one frame to the next. Assume that the confidence at time t is C t , the confidence at time t+1 is C t+1 , then the drop D can be calculated by the following formula:

[0054] D=C t -C t+1 ;

[0055] If D>0, it means that the confidence of the current frame is lower than that of the previous frame; if D<0, it means that the confidence of the current frame is higher than that of the previous frame.

[0056] Compare with the drop threshold: Set the threshold T to a number greater than 0, which is used to indicate the acceptable change in confidence. In each new frame, the calculated drop D is compared with the threshold T. If D is greater than T, it means that the drop exceeds the set threshold, and reverse tracking is started until the last frame in which the target appears is tracked.

[0057] Assume that when the endoscope is tracking the target lesion, occlusion occurs from the n+1th frame, and the confidence drop detected at the n+tth frame exceeds the threshold, then use this frame, i.e. the n+tth frame, as the new starting point to initialize the model of the tracking target. Then, use the retained historical frames and apply the KCF algorithm in reverse order to track to the nth frame.

[0058] In an embodiment of the present invention, the preset rule is to convert the Euclidean distance between the target lesion position predicted by the tracking model at the n+tth frame and the target lesion position predicted by the tracking model at the nth frame into the confidence score.

[0059] Specifically, from the position coordinates of the target lesion in the n+t frame (x n+t ,y n+t ) Start reverse tracking, use the KCF algorithm to reversely track the target lesion position in the last frame where the target appears, which is the nth frame The actual position (x n ,y n ) for comparison. The accuracy of the predicted position is evaluated by Euclidean distance. The Euclidean distance can be calculated by the following formula:

[0060]

[0061] The confidence obtained by converting the Euclidean distance obtained by reverse tracking. A higher confidence means that the current tracking result is very close to the actual target and the model performs well, while a lower confidence indicates that the model may be losing the target. When the confidence is high, the KCF algorithm continues to track according to the original model. Convert the Euclidean distance to the confidence score C n , convert the distance to a confidence score. Use the following formula:

[0062] C n =e -α·d ;

[0063] Among them, α represents the adjustment coefficient, d represents the Euclidean distance, C n represents the confidence score, and e represents the base of the natural logarithm.

[0064] Here, α is a positive coefficient that adjusts the sensitivity of the distance to the confidence level. When the distance is very small, the confidence level is close to 1, indicating a high degree of confidence; when the distance is large, the confidence level drops rapidly, indicating that the KCF algorithm has increased uncertainty in predicting the location of the target lesion due to short-term occlusion.

[0065] S24, determining whether the confidence score exceeds a preset score threshold, if so, continuing to track the target lesion; if not, updating the tracking model and re-tracking the target lesion.

[0066] Specifically, set the confidence score threshold C, if C n >C, it means that the occlusion has little effect on the target lesion to be tracked, and the target lesion can be tracked according to the original tracking model; if C n< C, it indicates that due to occlusion, the tracked target lesion has deviated significantly in position, and it is necessary to update the tracking model to re-track the target lesion.

[0067] When updating the tracking model, re - apply the new confidence to update the tracking model and perform forward tracking again. In the KCF algorithm, adjusting the regularization parameter can help the algorithm balance the fitting degree and generalization ability. The regularization parameter is used to avoid over - fitting. The regularization parameter affects the response peak during filter training, thus affecting the tracking stability. When the confidence decreases, it is necessary to increase the regularization parameter λ to reduce the risk of over - fitting. If the confidence increases and the algorithm shows sufficient generalization ability, try to reduce the regularization parameter λ so that the algorithm can fit the target more precisely. Implement an adaptive strategy according to the dynamic change of the confidence. When the confidence is low, automatically increase the regularization intensity. When the confidence is high, appropriately reduce the regularization.

[0068] S3. Obtain the current transmitted image transmitted by the endoscope, calculate the error between the center point of the target lesion position information and the center point of the current transmitted image to obtain the displacement error;

[0069] In the embodiment of the present invention, first, obtain the position center of the calibrated lesion in the image (i.e., the center point of the target lesion position information) through the above - mentioned KCF algorithm:

[0070] Next, calculate the center point of the detected lesion area (i.e., the center point of the current transmitted image). The lesion area is calibrated by a rectangular frame, and the width and height are represented by w and h respectively. The position of the rectangular frame is defined by the coordinates (x, y) of the lower - left corner and the width w and height h. Then the center coordinates of the lesion area are (x + w / 2, y + h / 2).

[0071] Based on the coordinates of the image center and the lesion center, the displacement difference between the two can be calculated. Let the image center coordinates be (X c , Y c ), and the center coordinates of the lesion area be (x + w / 2, y + h / 2). Then the displacement difference is (X d , Y d ) = (X c - x - w / 2, Y c - y - h / 2).

[0072] The front - end of the endoscope is finely controlled by the angle, and there is a certain angular constraint relationship with the two - dimensional plane. The displacement of the front - end of the endoscope is converted into an angle value for adjustment.

[0073] Therefore, set the Euclidean distance threshold to determine whether the lesion is within the range of the center of the image: If X d and Y dIf both are smaller than the threshold, the lesion is considered to be located in the center of the image; if one of them is larger than the threshold, the endoscope position needs to be further adjusted.

[0074] When calculating these variables, a simplified model is used: assuming that after the endoscope head is adjusted, the lesion moves from the initial position to the new position in the field of view of the endoscope camera, that is, the displacement difference (X d ,Y d ), and the straight-line distance d from the front end of the endoscope to the lesion remains unchanged.

[0075] S4. Adjust the front end direction of the endoscope according to the displacement error.

[0076] In an embodiment of the present invention, the front end direction of the endoscope is adjusted by a micro motor. Specifically, the control of the front end of the endoscope is controlled by two micro motors respectively, which realize independent movement on the x-axis and y-axis respectively. The center of the circle of the circular motion of the front end of the endoscope is used as the origin, and the coordinates (r, θ1, θ2) are used to describe the position of the front end of the endoscope, where r is the radius - the distance from the center of the circular path of the endoscope to the front end of the endoscope, and r is fixed; θ1 is the angle of the front end of the endoscope in the x-axis direction, which is controlled by motor 1; θ2 is the angle of the front end of the endoscope in the y-axis direction, which is controlled by motor 2. The arc lengths of the front end on the x-axis and y-axis are r·θ1 and r·θ2 respectively.

[0077] Assume that the displacements of the target lesion to be tracked on the x-axis and y-axis are X d and Y d By testing the displacement of the endoscope center field of view observed at different angles, it can be concluded that the adjustment size of the front end direction of the endoscope on the x-axis and y-axis satisfies the following relationship respectively:

[0078] X d =α1θ1 2 +b1θ1+c1;

[0079] Y d =α2θ2 2 +b2θ2+c2;

[0080] Among them, X d represents the displacement error of the endoscope on the x-axis, Y drepresents the displacement error of the endoscope on the y-axis, b1 and b2 are first-order coefficients, which respectively represent the influence of the angles of the front end of the endoscope in the x-axis direction and the y-axis direction on the displacement error; c1 and c2 are constant terms, which respectively represent the initial error or calibration error of the angle adjustment of the front end of the endoscope in the x-axis direction and the y-axis direction; α1 and α2 are second-order coefficients, which represent the nonlinear effect of the angle deviation on the displacement error; θ1 and θ2 respectively represent the angles of the front end of the endoscope in the x-axis direction and the y-axis direction.

[0081] When the angle of the endoscope tip needs to change very little, its arc length displacement is approximately equal to the displacement of the tracking target, so within a certain angle range, α1 and α2 are close to 0. At this time, the expression can be simplified to:

[0082] X d =b1θ1+c1;

[0083] Y d =b2θ2+c2;

[0084] The displacement difference X between the center point of the target lesion position information obtained by tracking the KCF algorithm and the center point of the current transmission image d and Y d The data is transmitted to the onboard control hardware board, which processes the input displacement difference signal and converts it into the field of view angle of the front end of the endoscope driven by the motor through the above formula.

[0085] In order to achieve precise control, a feedback mechanism is added to the control system. Through the encoder, the actual rotation angle of the endoscope front end driven by the motor is fed back to the control system in real time. The actual displacement difference of the endoscope field of view center is obtained by the above formula and compared with the center of the video image, and the distance difference is compared with the set threshold.

[0086] Set threshold X T and Y T ,If the Euclidean distance between the lesion and the image center is within the threshold, the ,tracking is considered successful and the transmitted image is locked for the ,next diagnosis.

[0087] Depending on how the actual displacement difference compares to the threshold, one of four situations may occur:

[0088] (1)X d <X T , Y d <Y T : The system determines that the lesion has been accurately tracked to the center of the video image, and the doctor can lock the transmitted image and proceed to the next step of medical diagnosis;

[0089] (2)X d >XT , Y d <Y T : It is considered that the tracking lesion is not located at the center of the image video, and it is necessary to readjust the front end field of view of the endoscope driven by the motor and reposition the target lesion until X d and Y d are all less than the threshold.

[0090] (3)X d <X T , Y d >Y T : It is considered that the tracking lesion is not located at the center of the image video, and it is necessary to readjust the front end field of view of the endoscope driven by the motor and reposition the target lesion until X d and Y d are all less than the threshold.

[0091] (4)X d >X T , Y d >Y T : It is considered that the tracking lesion is not located at the center of the image video, and it is necessary to readjust the front end field of view of the endoscope driven by the motor and reposition the target lesion until X d and Y d are all less than the threshold.

[0092] Compared with the prior art, the present invention obtains the target lesion by manually marking the lesion; establishes a tracking model based on the kernel correlation filter algorithm, tracks the target lesion in real time through the tracking model, and obtains the position information of the target lesion; obtains the current transmission image transmitted by the endoscope, calculates the error between the center point of the target lesion position information and the center point of the current transmission image, and obtains the displacement error; adjusts the front end direction of the endoscope according to the displacement error. The present invention efficiently tracks the position of the lesion through the correlation filter, and can accurately complete the target tracking even if the lesion moves or changes in shape within a certain range of the visual field, greatly reducing the attention burden of the doctor during the operation; realizes efficient and continuous tracking of the lesion, and can still accurately lock the lesion even when the lesion moves or changes in shape due to the patient's movement or shape. This ability significantly reduces the need for manual adjustment, thereby improving the doctor's operating efficiency and diagnostic accuracy for endoscopic examinations; the front end angle of the endoscope is automatically adjusted by a micromotor to ensure that the target lesion is always in the center of the image, greatly improving the convenience of surgery and diagnosis. This mechanism is specifically designed for doctors who hold an endoscope for a long time. It reduces their visual and physical burden and allows them to focus more on the diagnosis and treatment process during complex or lengthy examinations and operations, thereby improving work efficiency and treatment outcomes.

[0093] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0094] The embodiments of the present invention are described above in conjunction with the accompanying drawings. What is disclosed is only the preferred embodiment of the present invention. However, the present invention is not limited to the above-mentioned specific implementation manner. The above-mentioned specific implementation manner is only illustrative rather than restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms and equivalent changes without departing from the scope of protection of the purpose of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A dynamic tracking and feedback control method for endoscopic lesion detection, It is characterized in that The dynamic tracking and feedback control method comprises the following steps: S1. Manually mark the lesions to obtain the target lesions; S2, establishing a tracking model based on a kernel correlation filter algorithm, and tracking the position of the target lesion in real time through the tracking model to obtain the position information of the target lesion; S3, obtaining a current transmission image transmitted by the endoscope, calculating the error between the center point of the target lesion position information and the center point of the current transmission image, and obtaining a displacement error; S4. Adjust the front end direction of the endoscope according to the displacement error.

2. The dynamic tracking and feedback control method for endoscopic lesion detection according to claim 1, It is characterized in that Step S2 includes the following sub-steps: S21, determining whether the target lesion is continuously blocked within a preset time during the process of tracking the target lesion, if so, proceeding to step S22; if not, continuing to track the target lesion; S22, obtaining the current predicted position information of the target lesion by the tracking model, and determining whether the decrease amount of the confidence corresponding to the current predicted position information within the preset time exceeds a preset threshold value, if so, proceeding to step S23; if not, continuing to track the target lesion; S23, defining the time when the decrease in confidence corresponding to the current predicted position information exceeds the preset threshold as t frames, the previous frame when the target lesion is blocked as the nth frame, taking the target lesion position predicted by the tracking model at the n+tth frame as the starting point, starting reverse tracking to the target lesion position predicted by the tracking model at the nth frame, and calculating the confidence score of the tracking model after reverse tracking according to a preset rule, and proceeding to step S24; S24, determining whether the confidence score exceeds a preset score threshold, if so, continuing to track the target lesion; if not, updating the tracking model and re-tracking the target lesion.

3. The dynamic tracking and feedback control method for endoscopic lesion detection according to claim 2, It is characterized in that The preset rule is to convert the Euclidean distance between the target lesion position predicted by the tracking model at the n+tth frame and the target lesion position predicted by the tracking model at the nth frame into the confidence score, which satisfies: C n =and -α·d Among them, α represents the adjustment coefficient, d represents the Euclidean distance, C n represents the confidence score, and e represents the base of the natural logarithm.

4. The dynamic tracking and feedback control method for endoscopic lesion detection according to claim 1, It is characterized in that In step S4, the adjustment size of the front end direction of the endoscope on the x-axis and the y-axis satisfies the following relationship respectively: X d =a 1 i 1 2 +b 1 i 1 +c 1 ; Y d =a 2 i 2 2 +b 2 i 2 +c 2 ; Among them, X d represents the displacement error of the endoscope on the x-axis, Y d represents the displacement error of the endoscope on the y-axis, b 1 and b 2 They respectively represent the influence of the angle of the front end of the endoscope in the x-axis direction and the y-axis direction on the displacement error; c 1 and c 2 α represents the initial error or calibration error of the angle adjustment of the front end of the endoscope in the x-axis direction and the y-axis direction respectively; 1 and α 2 Represents the nonlinear effect of angular deviation on displacement error; θ 1 and θ 2 They respectively represent the angle of the front end of the endoscope in the x-axis direction and the y-axis direction.

5. The dynamic tracking and feedback control method for endoscopic lesion detection according to claim 1, It is characterized in that In step S4, the front end direction of the endoscope is adjusted by a micro motor.

Citation Information

Cited By

  • Endoscope video dynamic focus tracking method and system

    CN121258981A

  • A method and system for dynamic lesion tracking via endoscopic video

    CN121258981B