An adaptive path planning control system and method for a digestive endoscope probe

Abstract

This invention relates to the field of digestive endoscopy technology, and discloses an adaptive path planning control system and method for digestive endoscopy probes. The system includes an endoscopy probe and a handle. The handle contains an environment sensing unit, a pose estimation unit, a path planning unit, a motion control unit, and an actuator. The environment sensing unit extracts the cavity center, the probe-mucosa distance, and anatomical landmarks. The pose estimation unit calculates the probe's three-dimensional pose in real time. The path planning unit generates an adaptive motion trajectory based on the state vector. The motion control unit converts these into pulse commands to drive the actuator, which in turn drives the probe to bend or push via a wire traction mechanism. This invention achieves adaptive obstacle avoidance path planning and safety control of the endoscopy probe, reduces operational difficulty, shortens the learning cycle, effectively avoids mucosal damage, and improves the safety and efficiency of digestive endoscopy.

Description

Technical Field

[0001] This invention relates to the field of digestive endoscopy technology, and more specifically to an adaptive path planning control system and method for digestive endoscopy probes. Background Technology

[0002] Digestive endoscopy (including gastroscopy and colonoscopy) is an important tool for diagnosing and treating digestive tract diseases. Traditional digestive endoscopes are operated by a physician who holds the endoscope handle and uses a rotating control knob and a pushing / pulling lever to move the probe forward or backward within the digestive tract via a wire traction mechanism. However, current technology still has the following shortcomings:

[0003] 1. The position and orientation of the existing endoscopic probe are entirely controlled manually by the doctor, requiring the doctor to coordinate multiple degrees of freedom at the same time. The learning cycle is long, and prolonged operation can easily lead to hand fatigue, affecting the stability of the examination.

[0004] 2. During the procedure, doctors primarily rely on visual feedback to judge the distance between the probe and the mucosa, making it impossible to obtain precise probe-mucosa distance data in real time. When the intestines or stomach cavity are complex and winding, the probe can easily get too close to or even collide with the mucosa, causing patient discomfort, bleeding, or perforation risks.

[0005] 3. Doctors need to rely on their experience to identify anatomical landmarks such as the pylorus and ileocecal junction, and manually adjust the probe direction to reach the target area. For beginners or complex cases, this can easily lead to missed detections or repeated probe insertions and withdrawals, reducing examination efficiency.

[0006] 4. Existing endoscopic systems lack real-time 3D pose estimation capabilities, making it impossible to quantify the spatial position and orientation of the probe. When the probe enters a blind spot or encounters a sharp bend, doctors struggle to determine the probe's orientation, leading to blind path planning and an inability to proactively avoid dangerous areas.

[0007] 5. In automatic or semi-automatic assisted control systems, doctors cannot seamlessly intervene in control at any time. Once the automatic path planning deviates, adverse events can easily occur due to response delays.

[0008] To address the aforementioned problems, we propose an adaptive path planning control system and method for digestive endoscope probes. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides an adaptive path planning control system and method for digestive endoscope probes, overcoming the deficiencies of existing technologies.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] This invention provides an adaptive path planning control system for a digestive endoscope probe, comprising: an endoscope probe, and an endoscope handle connected to the endoscope probe via a wire traction mechanism;

[0012] The endoscope probe is equipped with an image acquisition unit for real-time acquisition of continuous images of the digestive tract lumen.

[0013] The endoscope handle contains the following units:

[0014] An environmental sensing unit, whose input end is electrically connected to the output end of the image acquisition unit, is used to extract environmental features from the image. The environmental features include the coordinates of the cavity center, the probe-mucosa distance, and anatomical landmarks.

[0015] The pose estimation unit has its input terminal electrically connected to the output terminal of the image acquisition unit, and is used to calculate the real-time pose of the endoscope probe in three-dimensional space based on the geometric changes of feature points in the continuous images;

[0016] The path planning unit has its first input terminal electrically connected to the output terminal of the environment perception unit and its second input terminal electrically connected to the output terminal of the pose estimation unit. It is used to generate an adaptive motion trajectory that avoids mucosal contact based on the current endoscopic probe pose, environmental features and preset target points.

[0017] A motion control unit, whose input terminal is electrically connected to the output terminal of the path planning unit, is used to convert the motion trajectory into pulse control commands for driving the actuator;

[0018] An actuator, the control end of which is electrically connected to the output end of the motion control unit, and the mechanical output end of the actuator forming a releasable fixed connection with the angle control knob on the endoscope handle, is used to drive the control components on the endoscope handle according to the pulse control command.

[0019] Preferably, the environmental perception unit includes: a cavity segmentation subunit, the input end of which is electrically connected to the output end of the image acquisition unit, for performing pixel-by-pixel classification of the image, dividing the image into cavity regions and tube wall regions, and outputting the geometric center coordinates of the cavity regions;

[0020] The depth estimation subunit has its input end electrically connected to the output end of the image acquisition unit. It is used to calculate the actual object distance corresponding to each pixel from a single frame image and output the probe-mucosa distance value of the central region of the endoscope probe tip.

[0021] The sign recognition subunit has its input end electrically connected to the output end of the image acquisition unit. It is used to perform feature matching between the current image and the pre-stored anatomical landmark feature data. When the matching degree exceeds a preset threshold, it outputs the pixel coordinates and sign type number of the landmark.

[0022] Preferably, the pose estimation unit includes: a feature extraction subunit, the input of which is electrically connected to the output of the image acquisition unit, for detecting significant feature points in two adjacent frames and generating corresponding feature description vectors;

[0023] The feature matching subunit has its input end electrically connected to the output end of the feature extraction subunit, and is used to pair the feature points of the current frame with the feature points of the previous frame and eliminate mismatched point pairs.

[0024] The motion calculation subunit, whose input is electrically connected to the output of the feature matching subunit, is used to calculate the translation vector and rotation Euler angle of the endoscope probe in three-dimensional space based on the pixel coordinate difference of the matching point pair and the camera intrinsic parameter matrix.

[0025] The map construction subunit has its input end electrically connected to the output end of the motion calculation subunit. It is used to accumulate and store the pose transformation at each moment with the starting point as the origin, and gradually construct a local three-dimensional spatial model expressed in the form of sparse feature point cloud.

[0026] Preferably, the path planning unit includes: a state coding subunit, whose input end is electrically connected to the output end of the environment perception unit and the output end of the pose estimation unit, respectively, for splicing the current endoscopic probe pose, cavity center coordinates, probe-mucosal distance value, and target point coordinates into a fixed-dimensional state vector;

[0027] The policy inference subunit has its input electrically connected to the output of the state encoding subunit. It internally stores a pre-trained adaptive policy network, used to output a short-term motion vector based on the state vector. ,in This represents the change in the bending angle of the endoscope probe. This refers to the displacement of the endoscope probe.

[0028] The safety verification subunit, whose input terminals are electrically connected to the output terminals of the strategy reasoning subunit and the environmental perception unit respectively, is used to determine whether the execution of the motion vector will trigger the probe-mucosal distance value to fall below a preset safety threshold. If it falls below, the motion vector is forcibly corrected. That is, stop bending and move backward in the opposite direction, where, This represents the displacement in the reverse direction.

[0029] Preferably, the actuator includes at least one drive unit for driving the angle control knob on the endoscope handle.

[0030] Preferably, it also includes a human-machine switching unit, whose first input terminal is electrically connected to the output terminal of the motion control unit, and whose second input terminal is used to receive intervention signals from the doctor's handheld controller, and whose output terminal is electrically connected to the control terminal of the actuator;

[0031] The human-machine switching unit is equipped with a signal arbitration logic circuit. When an intervention signal is received, the arbitration logic circuit cuts off the path of the automatic control command and transmits the intervention signal directly to the actuator.

[0032] This invention also provides an adaptive path planning control method for a digestive endoscope probe, applied to the aforementioned adaptive path planning control system, comprising the following steps:

[0033] Step S1: Acquire the intracavitary image of the current frame in real time through the image acquisition unit;

[0034] Step S2: The environmental perception unit extracts the cavity center coordinates, probe-mucosal distance value, and the presence of anatomical landmarks from the cavity image;

[0035] Step S3: The pose estimation unit calculates the pose change of the endoscope probe based on the current frame of the intraluminal image and the previous frame of the intraluminal image, and updates the current pose of the endoscope probe in three-dimensional space.

[0036] Step S4: The path planning unit generates the next motion vector based on the current state vector;

[0037] Step S5: The motion control unit converts the motion vector into pulse commands for the actuator.

[0038] Step S6: The actuator executes a pulse command to drive the angle control knob on the endoscope handle;

[0039] Step S7: Repeat steps S1 to S6 until the target point is reached or a stop signal is received.

[0040] Preferably, the probe-mucosal distance value in step S2 is calculated as follows:

[0041] First, for the central region of the cavity image For each pixel (u, v) within the range, calculate the estimated object distance corresponding to that pixel using the following formula. :

[0042] in, Represents pixels The grayscale value at that location ranges from 0 to 255. and For pre-calibrated constant coefficients,

[0043] Then, take the central region. The minimum estimated object distance among all pixels within the probe is taken as the probe-mucosal distance value, i.e.:

[0044] Among them, the central area The region is defined as a rectangular area with a width of one-third of the width of the cavity image and a height of one-third of the height of the cavity image, and the rectangular area is located at the center of the cavity image.

[0045] Preferably, the next motion vector in step S4 This is determined by solving the following constrained extremum problem:

[0046]

[0047] in: , For the pre-established probe kinematics function;

[0048] w1, w2, and w3 are weighting coefficients. =1mm is a small constant to prevent division by zero;

[0049] Furthermore, after each step of the movement is completed in step S6, a safety detection sub-step is added: if the current probe-mucosal distance value is detected to be less than 3 mm, the current movement is stopped immediately, a backward movement is automatically executed, the backward distance is 5 mm, and an alarm signal is sent to the doctor.

[0050] This invention provides an adaptive path planning control system and method for a digestive endoscope probe. It has the following beneficial effects:

[0051] By setting up path planning units and motion control units, adaptive motion trajectories that avoid the mucosa are automatically generated. Doctors do not need to manually fine-tune each degree of freedom, which significantly reduces the operational threshold and shortens the training time.

[0052] The environmental sensing unit outputs the probe-mucosal distance in real time. When the distance is lower than a preset threshold (such as 5mm), the safety verification subunit forcibly corrects the motion vector, causing the probe to stop moving forward or move backward, effectively preventing mucosal abrasion, bleeding or perforation.

[0053] The landmark recognition subunit uses pre-stored anatomical landmark feature data to automatically match and output the landmark coordinates. The path planning unit can set the landmark as an intermediate target point to achieve automated navigation, reduce doctors' repeated forward and backward operations, and shorten examination time.

[0054] The pose estimation unit updates the probe's pose in three-dimensional space in real time through feature extraction, matching, and motion calculation. It also combines the map construction subunit to generate a local point cloud model, providing accurate spatial reference for path planning and avoiding erroneous movement in blind spots.

[0055] The human-machine switching unit gives the intervention signals issued by the doctor's handheld controller the highest execution priority. It can cut off the automatic control command and take over the operation at any time, which retains the convenience of automation and ensures safe control in emergency situations. Attached Figure Description

[0056] Figure 1 This is a structural system architecture and signal flow diagram of the present invention.

[0057] Figure 2 This is a flowchart of the adaptive path planning control of the present invention.

[0058] Figure 3 This is a flowchart of the depth estimation and state vector construction sub-process of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] like Figure 1 As shown, in one embodiment, an adaptive path planning control system for digesting an endoscope probe includes: an endoscope probe, and an endoscope handle connected to the endoscope probe via a wire traction mechanism (i.e., several wires passing through the inside of the endoscope insertion tube for transmitting rotational or push-pull actions of the handle end to the probe tip, and a signal cable passing through the inside of the endoscope insertion tube for transmitting image signals acquired by the image acquisition unit to each unit inside the handle).

[0061] The endoscope probe is equipped with an image acquisition unit (the image acquisition unit is a miniature camera fixed to the front end of the probe, and the image acquisition unit is electrically connected to each unit inside the endoscope handle through the signal cable), which is used to acquire continuous images of the digestive tract lumen in real time;

[0062] The endoscope handle contains the following units:

[0063] An environmental sensing unit, whose input end is electrically connected to the output end of the image acquisition unit, is used to extract environmental features from the image. The environmental features include cavity center coordinates (i.e., the pixel center coordinates of the open cavity in front of the image), probe-mucosa distance (i.e., the vertical distance between the probe tip and the nearest digestive tract wall), and anatomical landmarks (i.e., fixed anatomical structures in the digestive tract that have identification significance, such as the pylorus and ileocecal junction).

[0064] The pose estimation unit, whose input is electrically connected to the output of the image acquisition unit, is used to calculate the real-time pose of the endoscope probe in three-dimensional space based on the geometric changes of feature points in the continuous images (the pose includes the three position coordinates of the probe in space). and three attitude angles ;

[0065] The path planning unit has its first input terminal electrically connected to the output terminal of the environmental perception unit and its second input terminal electrically connected to the output terminal of the pose estimation unit. It is used to generate an adaptive motion trajectory that avoids mucosal contact based on the current endoscopic probe pose, environmental features, and a preset target point (the target point is the spatial coordinate of the lesion location or anatomical landmark point pre-marked by the doctor or automatically identified by the system). The adaptive motion trajectory is a sequence consisting of a series of probe bending angle changes and push displacements.

[0066] The motion control unit, whose input terminal is electrically connected to the output terminal of the path planning unit, is used to convert the motion trajectory into pulse control commands for driving the actuator (the pulse control commands include the rotation direction, number of steps, and speed of the motor).

[0067] The actuator has its control end electrically connected to the output end of the motion control unit, and the mechanical output end of the actuator forms a releasable fixed connection with the angle control knob or push rod on the endoscope handle. It is used to drive the control components on the endoscope handle according to the pulse control command, and then indirectly control the endoscope probe to achieve bending or pushing movement through the wire traction mechanism.

[0068] Specifically, the environment perception unit includes a cavity segmentation subunit, whose input is electrically connected to the output of the image acquisition unit. This subunit is used to perform pixel-by-pixel classification of the image (i.e., using a deep learning model to label each pixel in the image as a "cavity" or "tube wall"), dividing the image into cavity regions and tube wall regions, and outputting the geometric center coordinates of the cavity regions. ;

[0069] The depth estimation subunit, whose input is electrically connected to the output of the image acquisition unit, is used to calculate the actual object distance (the actual object distance is the real distance from the probe tip to the tissue surface corresponding to that pixel) for each pixel in a single frame image, and outputs the probe-mucosa distance value of the central region of the endoscopic probe tip. (The probe-mucosal distance value is taken as the minimum value among all estimated object distances of all pixels in the central area).

[0070] The landmark recognition subunit, whose input is electrically connected to the output of the image acquisition unit, is used to perform feature matching between the current image and pre-stored anatomical landmark feature data (the feature matching uses scale-invariant feature transform or deep learning feature descriptors). When the matching degree exceeds a preset threshold, the pixel coordinates of the landmark are output. and the identifier type number.

[0071] Specifically, the pose estimation unit includes a feature extraction subunit, whose input end is electrically connected to the output end of the image acquisition unit, for detecting significant feature points (the significant feature points are unique local image locations such as corner points and edge points) in two adjacent frames of images and generating corresponding feature description vectors;

[0072] The feature matching subunit, whose input is electrically connected to the output of the feature extraction subunit, is used to pair the feature points of the current frame with the feature points of the previous frame (the pairing is achieved by calculating the Euclidean distance between the feature description vectors and selecting the nearest neighbor) and eliminate mismatched point pairs;

[0073] The motion calculation subunit, whose input is electrically connected to the output of the feature matching subunit, is used to calculate the translation vector of the endoscope probe in three-dimensional space based on the pixel coordinate difference of the matching point pairs and in conjunction with the camera intrinsic parameter matrix (which includes pre-calibrated parameters such as focal length and principal point coordinates). And rotating Euler angles ;

[0074] The map construction subunit has its input end electrically connected to the output end of the motion calculation subunit. It is used to accumulate and store the pose transformation at each moment with the starting point as the origin, and gradually construct a local three-dimensional spatial model expressed in the form of sparse feature point cloud (the sparse feature point cloud is composed of the three-dimensional coordinates of the feature points that were successfully matched in all historical frames).

[0075] Specifically, the path planning unit includes a state encoding subunit, whose input is electrically connected to the output of the environment perception unit and the output of the pose estimation unit, respectively, for concatenating the current endoscopic probe pose, cavity center coordinates, probe-mucosal distance, and target point coordinates into a fixed-dimensional state vector. (The state vector has 12 dimensions).

[0076] The policy inference subunit, whose input is electrically connected to the output of the state encoding subunit, internally stores a pre-trained adaptive policy network (the adaptive policy network is a deep reinforcement learning network, pre-trained in a simulation environment), used to output a short-term motion vector based on the state vector. ,in This represents the change in the bending angle of the endoscope probe (unit: degrees). The displacement of the endoscope probe (unit: millimeters).

[0077] The security verification subunit has its input terminals electrically connected to the output terminals of the strategy inference subunit and the environment perception unit, respectively, and is used to determine whether the execution of the motion vector will trigger the probe-mucosal distance value. Below the preset safety threshold ( (Take 5 mm), if it is lower, the motion vector will be forcibly corrected. That is, stop bending and move backward in the opposite direction, where, This is the displacement of the probe moving backward (i.e., the distance the probe retracts backward, taken as 2 mm).

[0078] Specifically, the actuator includes at least one drive unit (the drive unit is a micro stepper motor or servo motor) for driving the angle control knob or push rod on the endoscope handle to rotate, thereby driving the endoscope probe to achieve bending or pushing motion through the wire traction mechanism.

[0079] In one embodiment, an adaptive path planning control system for a digestive endoscope probe further includes a human-machine switching unit, whose first input terminal is electrically connected to the output terminal of the motion control unit, and whose second input terminal is used to receive intervention signals emitted by a doctor's handheld controller (the intervention signals are high-priority instructions generated by the doctor through buttons or joysticks), and whose output terminal is electrically connected to the control terminal of the actuator.

[0080] The human-machine switching unit is equipped with a signal arbitration logic circuit (the signal arbitration logic circuit is a hardware priority encoder or a priority judgment circuit composed of gate circuits). When an intervention signal is received, the arbitration logic circuit cuts off the path of the automatic control command and directly transmits the intervention signal to the actuator.

[0081] The following describes an adaptive path planning control method for a digestive endoscope probe provided by the present invention. The adaptive path planning control method for a digestive endoscope probe described below can be referred to in correspondence with the adaptive path planning control system for a digestive endoscope probe described above.

[0082] like Figure 2 As shown, in one embodiment, an adaptive path planning control method for a digestive endoscope probe includes the following steps:

[0083] Step S1: Acquire the intracavitary image of the current frame in real time through the image acquisition unit. (The intracavitary image is a grayscale image or a color image, with a resolution equal to the original output resolution of the endoscopic camera.)

[0084] Step S2: The environmental sensing unit obtains images from the cavity. Extract the center coordinates of the cavity. Probe-mucosal distance value And whether there are anatomical landmarks;

[0085] Step S3: The pose estimation unit estimates the cavity image in the current frame. Compared with the previous frame of the intracavitary image Calculate the pose changes of the endoscope probe And update the current pose of the endoscope probe in three-dimensional space. (in The unit is millimeters. (Unit: degrees)

[0086] Step S4: The path planning unit calculates the current state vector... Generate the next motion vector ;in The preset target point pose coordinates (the target point pose is obtained by the doctor clicking on the image marker before or during the operation);

[0087] Step S5: The motion control unit converts the motion vector into pulse commands for the actuator.

[0088] Step S6: The actuator executes a pulse command to drive the angle control knob or push rod on the endoscope handle to move, thereby indirectly controlling the endoscope probe to complete the corresponding bending or pushing action through the wire traction mechanism;

[0089] Step S7: Repeat steps S1 to S6 until the termination condition is met: the distance between the current probe pose and the target point pose. Less than the preset arrival judgment value ( (Take 2 mm), or receive a stop signal from the human-machine switching unit.

[0090] like Figure 3 As shown, specifically, the calculation method for the probe-mucosal distance value in step S2 is as follows:

[0091] First, for the central region of the cavity image For each pixel (u, v) within the range, calculate the estimated object distance corresponding to that pixel using the following formula. :

[0092]

[0093] in, Represents pixels The grayscale value at that location ranges from 0 to 255. and These are pre-calibrated constant coefficients (the calibration is obtained by taking images of the calibration plate at a known distance and fitting the gray-scale-distance relationship curve using the least squares method).

[0094] The calibration method is as follows: acquire at least 100 endoscopic images with known true distances, and obtain the calibration results through least squares fitting. and The specific value;

[0095] Then, take the central region. The minimum estimated object distance among all pixels within the probe is taken as the probe-mucosal distance value, i.e.:

[0096]

[0097] Among them, the central area It is defined as a rectangular region with a width of one-third of the width of the internal cavity image and a height of one-third of the height of the internal cavity image, and the rectangular region is located at the center of the internal cavity image (the central region corresponds to the main field of view directly in front of the probe, and is used to estimate the distance between the probe and the nearest tissue in front).

[0098] Specifically, the state vector mentioned in step S4 It consists of the following 12 values, among which:

[0099] The six values ​​in the current pose, namely ;

[0100] The two values ​​in the center coordinates of the cavity, namely The two values ​​are normalized by dividing them by the width and height of the cavity image, respectively (normalization makes the coordinate values ​​dimensionless and range between 0 and 1).

[0101] The probe-mucosal distance value is in millimeters;

[0102] The three values ​​in the target point deviation vector are: , representing the coordinate differences between the target point and the current probe in the three translational directions, respectively.

[0103] Specifically, the next motion vector described in step S4 This is determined by solving the following constrained extremum problem:

[0104]

[0105] in: , The probe kinematic function is a pre-established function (obtained by experimentally measuring the mapping relationship between the probe bending angle and the front displacement).

[0106] Constraints: ;

[0107] w1, w2, and w3 are weighting coefficients, with values ​​of w1=0.5, w2=0.3, and w3=0.2 respectively.

[0108] =1mm is a small constant to prevent division by zero;

[0109] The following constraints must be satisfied when solving the aforementioned extremum problem:

[0110] Mucosal distance constraint: ≥5 mm, meaning the distance between the probe and the mucosa must not be less than 5 mm;

[0111] Bending angle variation constraint: ≤30 degrees, meaning the change in bending angle in a single step does not exceed 30 degrees;

[0112] Push distance constraint: 0≤ ≤100 mm, meaning the single-step push distance is not less than 0 and not more than 10 mm;

[0113] Furthermore, after each step of the movement is completed in step S6, a safety detection sub-step is added: if the current probe-mucosal distance value is detected... If the distance is less than 3 mm, the current movement will stop immediately, and a backward movement will be automatically executed, with a backward distance of 5 mm. At the same time, an alarm signal will be sent to the doctor.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive path planning control system for a digestive endoscope probe, characterized in that, include: Endoscope probe, endoscope handle connected to the endoscope probe via a steel wire traction mechanism; The endoscope probe is equipped with an image acquisition unit for real-time acquisition of continuous images of the digestive tract lumen. The endoscope handle contains the following units: An environmental sensing unit, whose input end is electrically connected to the output end of the image acquisition unit, is used to extract environmental features from the image. The environmental features include the coordinates of the cavity center, the probe-mucosa distance, and anatomical landmarks. The pose estimation unit has its input terminal electrically connected to the output terminal of the image acquisition unit, and is used to calculate the real-time pose of the endoscope probe in three-dimensional space based on the geometric changes of feature points in the continuous images; The path planning unit has its first input terminal electrically connected to the output terminal of the environment perception unit and its second input terminal electrically connected to the output terminal of the pose estimation unit. It is used to generate an adaptive motion trajectory that avoids mucosal contact based on the current endoscopic probe pose, environmental features and preset target points. A motion control unit, whose input terminal is electrically connected to the output terminal of the path planning unit, is used to convert the motion trajectory into pulse control commands for driving the actuator; An actuator, the control end of which is electrically connected to the output end of the motion control unit, and the mechanical output end of the actuator forming a releasable fixed connection with the angle control knob on the endoscope handle, is used to drive the control components on the endoscope handle according to the pulse control command.

2. The system according to claim 1, characterized in that, The environmental perception unit includes a cavity segmentation subunit, whose input end is electrically connected to the output end of the image acquisition unit, for performing pixel-by-pixel classification of the image, dividing the image into cavity regions and tube wall regions, and outputting the geometric center coordinates of the cavity regions; The depth estimation subunit has its input end electrically connected to the output end of the image acquisition unit. It is used to calculate the actual object distance corresponding to each pixel from a single frame image and output the probe-mucosa distance value of the central region of the endoscope probe tip. The sign recognition subunit has its input end electrically connected to the output end of the image acquisition unit. It is used to perform feature matching between the current image and the pre-stored anatomical landmark feature data. When the matching degree exceeds a preset threshold, it outputs the pixel coordinates and sign type number of the landmark.

3. The system according to claim 1, characterized in that, The pose estimation unit includes a feature extraction subunit, whose input end is electrically connected to the output end of the image acquisition unit, and is used to detect significant feature points in two adjacent frames of images and generate corresponding feature description vectors. The feature matching subunit has its input end electrically connected to the output end of the feature extraction subunit, and is used to pair the feature points of the current frame with the feature points of the previous frame and eliminate mismatched point pairs. The motion calculation subunit, whose input is electrically connected to the output of the feature matching subunit, is used to calculate the translation vector and rotation Euler angle of the endoscope probe in three-dimensional space based on the pixel coordinate difference of the matching point pair and the camera intrinsic parameter matrix. The map construction subunit has its input end electrically connected to the output end of the motion calculation subunit. It is used to accumulate and store the pose transformation at each moment with the starting point as the origin, and gradually construct a local three-dimensional spatial model expressed in the form of sparse feature point cloud.

4. The system according to claim 1, characterized in that, The path planning unit includes a state coding subunit, whose input is electrically connected to the output of the environment perception unit and the output of the pose estimation unit, respectively, and is used to concatenate the current endoscopic probe pose, cavity center coordinates, probe-mucosal distance value, and target point coordinates into a fixed-dimensional state vector. The policy inference subunit has its input electrically connected to the output of the state encoding subunit. It internally stores a pre-trained adaptive policy network, used to output a short-term motion vector based on the state vector. ,in This represents the change in the bending angle of the endoscope probe. This refers to the displacement of the endoscope probe. The safety verification subunit, whose input terminals are electrically connected to the output terminals of the strategy reasoning subunit and the environmental perception unit respectively, is used to determine whether the execution of the motion vector will trigger the probe-mucosal distance value to fall below a preset safety threshold. If it falls below, the motion vector is forcibly corrected. That is, stop bending and move backward in the opposite direction, where, This represents the displacement in the reverse direction.

5. The system according to claim 1, characterized in that, The implementing mechanism includes: At least one drive unit is provided for driving the angle control knob on the endoscope handle.

6. The system according to claim 1, characterized in that, It also includes a human-machine switching unit, whose first input terminal is electrically connected to the output terminal of the motion control unit, and whose second input terminal is used to receive intervention signals from the doctor's handheld controller, and whose output terminal is electrically connected to the control terminal of the actuator; The human-machine switching unit is equipped with a signal arbitration logic circuit. When an intervention signal is received, the arbitration logic circuit cuts off the path of the automatic control command and transmits the intervention signal directly to the actuator.

7. An adaptive path planning control method for a digestive endoscope probe, applied to the adaptive path planning control system according to any one of claims 1 to 6, characterized in that, Includes the following steps: Step S1: Acquire the intracavitary image of the current frame in real time through the image acquisition unit; Step S2: The environmental perception unit extracts the cavity center coordinates, probe-mucosal distance value, and the presence of anatomical landmarks from the cavity image; Step S3: The pose estimation unit calculates the pose change of the endoscope probe based on the current frame of the intraluminal image and the previous frame of the intraluminal image, and updates the current pose of the endoscope probe in three-dimensional space. Step S4: The path planning unit generates the next motion vector based on the current state vector; Step S5: The motion control unit converts the motion vector into pulse commands for the actuator. Step S6: The actuator executes a pulse command to drive the angle control knob on the endoscope handle; Step S7: Repeat steps S1 to S6 until the target point is reached or a stop signal is received.

8. The method according to claim 7, characterized in that, The calculation method for the probe-mucosal distance value in step S2 is as follows: First, for the central region of the cavity image For each pixel (u, v) within the range, calculate the estimated object distance corresponding to that pixel using the following formula. : in, Represents pixels The grayscale value at that location ranges from 0 to 255. and For pre-calibrated constant coefficients, Then, take the central region. The minimum estimated object distance among all pixels within the probe is taken as the probe-mucosal distance value, i.e.: Among them, the central area The region is defined as a rectangular area with a width of one-third of the width of the cavity image and a height of one-third of the height of the cavity image, and the rectangular area is located at the center of the cavity image.

9. The method according to claim 7, characterized in that, The next motion vector described in step S4 This is determined by solving the following constrained extremum problem: in: , For the pre-established probe kinematics function; w1, w2, and w3 are weighting coefficients. =1mm is a small constant to prevent division by zero; Furthermore, after each step of the movement is completed in step S6, a safety detection sub-step is added: if the current probe-mucosal distance value is detected to be less than 3 mm, the current movement is stopped immediately, a backward movement is automatically executed, the backward distance is 5 mm, and an alarm signal is sent to the doctor.

Images