AI identification and driving-off control system and method of mobile inspection device

Abstract

The application discloses an AI identification and driving control system and method of a mobile inspection device, and relates to the technical field of driving control.The system comprises an image acquisition module, a biological identification and orientation analysis module, a same-direction pursuit judgment module, a driving direction execution module, and a continuous tracking and speed adjustment module;inspection images are acquired, target organisms are identified, and escape directions are predicted according to orientation information;whether a same-direction pursuit state is formed is determined by combining the moving direction of the device and the path area position;the driving direction is adjusted so that the included angle meets the threshold requirement and driving is performed;meanwhile, the position of the organism is continuously tracked;and collaborative control is realized by adjusting the speed and dynamically updating the driving direction.The application can effectively break the same-direction pursuit phenomenon, quickly drive the organism away from the inspection path, avoid the obstruction of the inspection process, and improve the continuous inspection operation efficiency and operation stability.

Description

Technical Field

[0001] This invention relates to the field of drive-away control technology, specifically to an AI recognition and drive-away control system and method for mobile inspection devices. Background Technology

[0002] Mobile inspection devices are widely used in outdoor settings such as substations, factories, and industrial parks, and can automatically perform environmental monitoring, equipment inspection, and anomaly handling. To address biological disturbances such as birds and small animals along the inspection path, existing devices are often equipped with visual recognition modules and acoustic, light, and ultrasonic repellency modules. By using AI to identify biological targets, they initiate repellency actions, ensuring the safe and stable conduct of inspection operations.

[0003] Existing technologies typically treat device movement and organism removal as independent execution steps. The device stops moving after reaching the target location, completes the removal process, and confirms the organism's departure before continuing its inspection along the path. This approach does not consider the dynamic relationship between the removal process and device movement. In actual continuous inspections, when the device detects and removes an organism during its journey, the removal method can easily guide the organism to escape along the device's subsequent direction of travel, keeping the organism continuously within the device's path. Existing control methods cannot predict the organism's escape direction or coordinate adjustments with the device's movement path. When the organism's escape direction is similar to the device's movement direction, the removed organism will remain in front of the device, easily leading to a chase and preventing effective removal of the organism from the inspection path. This results in continuous congestion of the inspection process and a significant decrease in overall operational efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based identification and disengagement control system and method for mobile inspection devices, in order to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The AI-based identification and decoy control method for mobile inspection devices includes the following steps: S1. Acquire inspection images from the mobile inspection device, identify target organisms in the inspection images, and acquire the location and orientation information of the target organisms in the inspection images; based on the orientation information of the target organisms in the inspection images, predict the escape direction of the target organisms after they are driven away. S2. Obtain the moving direction of the mobile inspection device. Based on the spatial angle between the predicted escape direction of the target creature and the moving direction of the mobile inspection device, and the positional relationship between the target creature's position information in the inspection image and the preset direction determination area, determine whether the mobile inspection device and the target creature meet the chasing formation conditions. If the chasing formation conditions are met, determine that a chasing state in the same direction has been formed. The chasing formation conditions are that the spatial angle is less than a preset angle threshold, and the target creature's position in the inspection image is within the preset direction determination area. S3. For the determination of forming a chasing state, the driving direction parameter of the driving module is determined according to the escape direction of the target organism after being driven away and the moving direction of the mobile inspection device, so that the spatial angle between the driving direction and the moving direction of the mobile inspection device is greater than or equal to the preset angle threshold; the driving module is controlled to perform the driving action according to the driving direction parameter. S4. After the driving action is executed, the target creature is continuously tracked, and the mobile inspection device is judged based on the real-time relative position between the target creature and the mobile inspection device to determine whether the mobile inspection device and the target creature have broken away from the same-direction chasing state. If they have not broken away, the mobile inspection device's speed is reduced or its movement is stopped, and the driving direction parameters are readjusted until the target creature deviates from the mobile inspection device's forward path, at which point the original inspection speed and original inspection route are restored.

[0006] Furthermore, S1 includes the following: The mobile inspection device acquires inspection images, which are then input into a trained target organism detection model. The target organism detection model outputs the detection results of the target organisms in the inspection images, including the category of the target organisms and the position coordinates of the target detection boxes in the inspection images. The position coordinates of the target detection box in the inspection image are used as the position information of the target organism. The position information includes the planar coordinates of the target organism in the inspection image. The planar coordinates are taken as the origin of the upper left corner of the inspection image, in pixels, and are represented by the center coordinates of the target detection box. Based on the position coordinates of the target detection box in the inspection image, the bounding box region of the target organism is cropped from the inspection image and used as the input region for posture key point extraction. Posture key point extraction is performed on the bounding box region to obtain the head key point coordinates (uh, vh) and torso center point coordinates (ub, vb) of the target organism, where the head key point is the geometric center point of the head region of the target organism, and the torso center point is the geometric center point of the torso region of the target organism. The posture key point extraction is implemented using a pre-trained posture key point recognition model, which takes the bounding box region of the target organism as input and the head key point coordinates and torso center point coordinates as output. Construct an orientation vector Vh in the image plane, where Vh = (uh - ub, vh - vb), where uh and vh are the x and y coordinates of the head keypoint in the image plane, respectively, and ub and vb are the x and y coordinates of the torso center point in the image plane, respectively. The starting point of the orientation vector Vh is the torso center point coordinates (ub, vb), and the ending point is the head keypoint coordinates (uh, vh). The direction of the orientation vector Vh in the image plane is the natural orientation of the target organism. Calculate the orientation angle θh of the orientation vector, where θh = atan2(uh - ub, vh - vb), where atan2 is the arctangent function in the four quadrants, and the value of θh ranges from -180° to 180°, with the horizontal rightward direction of the image as 0° and counterclockwise as positive. Use the orientation angle of the orientation vector as the orientation information of the target organism, and use the orientation information θh as the predicted escape direction of the target organism after being driven away.

[0007] Furthermore, S2 includes the following: The current movement direction of the mobile inspection device is obtained. The movement direction is provided in real time by the navigation system of the mobile inspection device and is represented by the movement direction angle θm of the device on the horizontal ground. The movement direction angle θm is 0° with due north as the reference direction and clockwise as the reference direction, and the value range is 0° to 360°. The offset angle θc of the current orientation of the inspection device's camera relative to due north is obtained. The offset angle θc is determined by the fixed installation angle of the camera or the real-time rotation angle of the pan-tilt unit. θm is converted to the image plane coordinate system. The conversion relationship is θm' = θm - θc. The converted movement direction angle θm' is 0° with the horizontal rightward direction of the image as the reference direction and counterclockwise as the reference direction. The angle is adjusted to the range of -180° to 180° by adding or subtracting 360°. Based on the predicted escape direction θh of the target organism and the angle θm' of the moving direction of the mobile inspection device after conversion, the angle difference Δθ = θh - θm' is calculated. By adding or subtracting 360°, the angle difference Δθ is adjusted to the range of [-180°, 180°] to obtain Δθn. Then the spatial angle θ1 = |Δθn|, and the value range of θ1 is from 0° to 180°. A rectangular region centered on the geometric center of the inspection image is obtained and used as the preset direction determination region, with a horizontal half-width of Dx and a vertical half-height of Dy. The specific values ​​of Dx and Dy are pre-calibrated based on the inspection image resolution and the camera's field of view. Based on the obtained target detection box center point coordinates (xc, yc) and the target detection box's width w and height h, the ratio P of the intersection area Si of the target detection box and the preset direction determination region to the total area Sb of the target detection box is calculated. Here, the total area Sb of the target detection box = w × h, and the intersection area Sb of the target detection box and the preset direction determination region is... i = max(0, min(xc+w / 2, W / 2+Dx)-max(xc-w / 2, W / 2-Dx))×max(0, min(yc+h / 2, H / 2+Dy)-max(yc-h / 2, H / 2-Dy)), where W is the width of the inspection image, H is the height of the inspection image, and the ratio P = Si / Sb. When P > P0, the target organism is determined to be located within the preset direction determination area, i.e., within the forward path direction of the mobile inspection device; otherwise, it is determined to be located outside the preset direction determination area. Here, P0 is the preset area ratio threshold. An initial cumulative counter C is set to 0. The chasing formation condition is determined frame by frame. The calculated spatial angle θ1 is compared with a preset angle threshold θ0, which is pre-set based on the mobile inspection device's speed and scene safety requirements. When θ1 < θ0, the first condition is satisfied; otherwise, it is not. The determination of whether the target organism is within a preset direction determination area is used as the second condition. Within a single frame, if both the first and second conditions are satisfied, the cumulative counter C is incremented by 1; if either condition is not satisfied, the cumulative counter C is reset to 0. A preset frame count threshold N is set based on the image acquisition frame rate and determination stability requirements. When the value of the cumulative counter C is greater than or equal to the preset frame count threshold N, the mobile inspection device and the target organism are determined to meet the chasing formation condition, forming a chasing state in the same direction.

[0008] Furthermore, S3 includes the following: To determine if a unidirectional chasing state has formed, the calculated spatial angle θ1 is obtained. Since θ1 < θ0 when a unidirectional chasing state is formed, the driving direction is adjusted so that the spatial angle between the driving direction and the moving direction is greater than or equal to the preset angle threshold θ0. The coordinates of the target detection box center point (xc, yc) and the coordinates of the inspection image center (W / 2, H / 2) are obtained, and the deflection angle θp of the target organism relative to the image center in the image plane is calculated, where θp = atan2(yc-H / 2, xc-W / 2). The value of θp ranges from -180° to 180°, with the image horizontal to the right. θp is the deflection angle of the target organism's current direction relative to the front of the mobile inspection device. It is directly calculated from the center position of the target detection box and represents the spatial orientation of the target organism at the current moment. Unlike the escape direction θh predicted based on the orientation of the organism's head in S1, θp does not involve the prediction of the organism's movement trend, but reflects the actual position and orientation of the target organism in the image plane. When determining the driving direction, θp should be used as the pointing basis for the driving direction, so that the driving module acts towards the current orientation of the target organism, thereby driving the organism laterally away from the direction of movement of the mobile inspection device. Calculate the angle difference Δθp = θp - θm'. Adjust the angle difference Δθp to the range of [-180°, 180°] by adding or subtracting 360°, and take the absolute value to obtain the spatial angle θ2 between the direction of the target organism and the moving direction angle θm'. If θ2 ≥ θ0, the target driving direction angle θd is directly set to θp. If θ2 < θ0, the target driving direction angle θd is determined according to θd = θp + sign(θp - θm') × (θ0 - θ2 + Δ), where sign is the sign function, which takes 1 when the input is greater than 0, -1 when it is less than 0, and 0 when it is equal to 0. Δ is the redundant angle preset according to the control accuracy of the driving mechanism. The obtained θd is adjusted to the range of [-180°, 180°] by adding or subtracting 360°, and the driving module is controlled to perform the driving action according to the adjusted target driving direction angle θd.

[0009] Furthermore, S4 includes the following: After the driving action is executed, the mobile inspection device continuously tracks the target organism. It achieves tracking by acquiring the coordinates (xc, yc) of the center point of the target detection box frame by frame. Based on the position change of the center point of the target detection box in multiple consecutive frames, it determines whether the target organism has deviated from the forward path of the mobile inspection device. The departure determination cumulative counter Ce is set with an initial value of 0. In each frame of the inspection image, the intersection area ratio P of the target detection box and the preset direction determination area is calculated according to the area ratio calculation formula in S2. When P≤P0, it is determined that the target organism has deviated from the preset direction determination area, and Ce=Ce+1. When P>P0, it is determined that the target organism is still within the preset direction determination area, and Ce=0. A preset threshold M for the number of frames for determining escape is set according to the sensitivity requirements for escape determination. When Ce≥M, it is determined that the target creature has stably deviated from the forward path of the mobile inspection device, the same-direction pursuit state has been released, the driving action is stopped, the mobile inspection device resumes the original inspection speed and original inspection route, and the cumulative counter C in S2 is reset to 0, and the frame-by-frame determination of the pursuit formation condition is re-executed; when Ce<M, it is determined that the target creature has not stably deviated from the preset direction determination area, and the same-direction pursuit state has not been released. The current travel speed v of the mobile inspection device is obtained and compared with the preset minimum effective deportation speed vmin. vmin is preset according to the effective operation conditions of the deportation module. When v > vmin, the mobile inspection device is controlled to reduce its travel speed to vmin. When v ≤ vmin, the mobile inspection device is controlled to pause its movement. The deflection angle θpc of the target organism relative to the center of the image in the current frame is obtained. The angular deviation Δθe between θpc and the target deportation direction angle θd in S3 is calculated. The angular deviation Δθe is adjusted to the range of [-180°, 180°] by adding or subtracting 360° and the absolute value is taken to obtain the deviation angle θdc. θdc is compared with the preset deviation threshold θdt, where θdt is preset according to the target organism's movement speed and the response accuracy of the deportation module. When θdc > θdt, calculate the spatial angle θ2 between the current frame target creature's direction and the movement direction angle θm' in the same way as in S3, and update the target removal direction angle θdn according to θdn = θpc + sign(θpc - θm') × (θ0 - θ2 + Δ). After adjusting θdn to the range of [-180°, 180°] by adding or subtracting 360°, control the removal module to perform the removal action according to the updated removal direction angle. When θdc ≤ θdt, maintain the current removal direction unchanged; continue to execute the detachment judgment process. The preset total removal duration is preset according to the target creature category and scene safety requirements. If Ce does not reach M within the preset total removal duration, send an alarm message to the background monitoring system to prompt manual intervention.

[0010] The AI ​​recognition and expulsion control system of the mobile inspection device includes: an image acquisition module, a biometric recognition and orientation analysis module, a same-direction pursuit judgment module, an expulsion direction execution module, and a continuous tracking and speed adjustment module; The image acquisition module acquires inspection images in real time during the inspection process of the mobile inspection device; The biometrics and orientation analysis module identifies target organisms in the inspection images, obtains the target organism's location and orientation information, and predicts the escape direction of the target organism after it is driven away based on the orientation information. The same-direction pursuit judgment module obtains the moving direction of the mobile inspection device, and determines whether a same-direction pursuit state has been formed based on the spatial angle between the escape direction and the moving direction, as well as the positional relationship between the target creature and the preset direction judgment area. When a chasing state is formed in the same direction, the driving direction execution module determines the driving direction parameters of the driving module, so that the spatial angle between the driving direction and the moving direction is greater than or equal to the preset angle threshold, and controls the driving module to perform the driving action. The continuous tracking and speed adjustment module continuously tracks the target organism during the driving process, determines whether the target organism has broken away from the same-direction pursuit state, and reduces the moving inspection device's speed or stops moving and readjusts the driving direction parameters if the target organism has not broken away. After the target organism deviates from the forward path, the original inspection speed and original inspection route are restored.

[0011] Furthermore, the biometrics and orientation analysis module includes a target detection unit and a pose and orientation extraction unit; The target detection unit inputs the inspection image into the trained target organism detection model and outputs the category of the target organism and the position coordinates of the target detection box in the inspection image; The orientation extraction unit clips the bounding box region of the target organism based on the target detection box, obtains the coordinates of the head key points and the center point of the torso through the orientation key point recognition model, constructs the orientation vector and calculates the orientation angle of the orientation vector, and determines the orientation angle as the predicted escape direction.

[0012] Furthermore, the same-direction chasing judgment module includes a coordinate angle transformation unit and a dual-condition counting unit; The coordinate angle transformation unit obtains the motion direction angle of the mobile inspection device and the offset angle of the camera, transforms the motion direction angle to the image plane coordinate system, and adjusts it to the set angle range; The dual-condition counting unit calculates the spatial angle between the escape direction and the movement direction, determines the proportion of the intersection area between the target detection box and the preset direction judgment area, and judges whether a chasing state in the same direction is formed based on the continuous counting results of the cumulative counter.

[0013] Furthermore, the deportation direction execution module includes a target orientation calculation unit and a deportation angle correction unit; The target orientation calculation unit calculates the deflection angle of the target organism relative to the image center based on the center coordinates of the target detection box and the center coordinates of the image, and obtains the spatial angle between the deflection angle and the movement direction angle. The expulsion angle correction unit determines the expulsion direction angle based on the relationship between the spatial angle and the preset angle threshold, and outputs the expulsion direction angle to the expulsion module after adjusting it to the set angle range.

[0014] Furthermore, the continuous tracking and speed control module includes a disengagement status judgment unit and a disengagement parameter update unit; The detachment state judgment unit calculates the percentage of the intersection area between the target detection box and the preset direction judgment area frame by frame, and judges whether the target organism has stably deviated from the forward path based on the counting result of the detachment judgment cumulative counter. The deflection parameter update unit compares the travel speed with the minimum effective deflection speed, controls the mobile inspection device to perform deceleration or pause operations, calculates the deviation between the target deflection angle and the deflection direction angle, updates the deflection direction angle when the deviation exceeds the preset threshold, and sends an alarm message to the background monitoring system if the target is not deflected within the set time.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention identifies target organisms and extracts their orientation information to predict the direction of their escape, avoiding blind triggering of expulsion actions; by combining the spatial angle between the moving direction of the inspection device and the escape direction of the organism, as well as the position of the organism in the forward path area, it accurately determines whether a state of chasing in the same direction has formed, and promptly identifies the risk of inspection blockage; for the same-direction chasing scenario, it adjusts the expulsion direction parameters to maintain a sufficient angle between the expulsion direction and the moving direction of the device, guiding the organism to shift to the outside of the inspection route, breaking the state of chasing in the same direction from the root; during the expulsion process, it continuously tracks the position of the organism and dynamically updates the expulsion direction in conjunction with the movement speed adjustment, realizing coordinated control of inspection movement and organism expulsion, without stopping the inspection and waiting for the expulsion to be completed; it can quickly and effectively expel organisms outside the forward path, avoiding continuous blockage of the inspection process, ensuring smooth execution of continuous inspection operations, and improving overall inspection efficiency and operational stability. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the AI ​​recognition and expulsion control method of the mobile inspection device of the present invention. Detailed Implementation

[0017] 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, and 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.

[0018] The present invention provides the following technical solution: The AI ​​recognition and expulsion control system of the mobile inspection device includes: an image acquisition module, a biometric recognition and orientation analysis module, a same-direction pursuit judgment module, an expulsion direction execution module, and a continuous tracking and speed adjustment module; The image acquisition module acquires inspection images in real time during the inspection process of the mobile inspection device; The biometrics and orientation analysis module identifies target organisms in the inspection images, obtains the target organism's location and orientation information, and predicts the escape direction of the target organism after it is driven away based on the orientation information. The same-direction pursuit judgment module obtains the moving direction of the mobile inspection device, and determines whether a same-direction pursuit state has been formed based on the spatial angle between the escape direction and the moving direction, as well as the positional relationship between the target creature and the preset direction judgment area. When a chasing state is formed in the same direction, the driving direction execution module determines the driving direction parameters of the driving module, so that the spatial angle between the driving direction and the moving direction is greater than or equal to the preset angle threshold, and controls the driving module to perform the driving action. The continuous tracking and speed adjustment module continuously tracks the target organism during the driving process, determines whether the target organism has broken away from the same-direction pursuit state, and reduces the moving inspection device's speed or stops moving and readjusts the driving direction parameters if the target organism has not broken away. After the target organism deviates from the forward path, the original inspection speed and original inspection route are restored.

[0019] The biometrics and orientation analysis module includes a target detection unit and a pose and orientation extraction unit; The target detection unit inputs the inspection image into the trained target organism detection model and outputs the category of the target organism and the position coordinates of the target detection box in the inspection image; The orientation extraction unit clips the bounding box region of the target organism based on the target detection box, obtains the coordinates of the head key points and the center point of the torso through the orientation key point recognition model, constructs the orientation vector and calculates the orientation angle of the orientation vector, and determines the orientation angle as the predicted escape direction.

[0020] The same-direction chasing judgment module includes a coordinate angle transformation unit and a dual-condition counting unit; The coordinate angle transformation unit obtains the motion direction angle of the mobile inspection device and the offset angle of the camera, transforms the motion direction angle to the image plane coordinate system, and adjusts it to the set angle range; The dual-condition counting unit calculates the spatial angle between the escape direction and the movement direction, determines the proportion of the intersection area between the target detection box and the preset direction judgment area, and judges whether a chasing state in the same direction is formed based on the continuous counting results of the cumulative counter.

[0021] The drive-away direction execution module includes a target orientation calculation unit and a drive-away angle correction unit; The target orientation calculation unit calculates the deflection angle of the target organism relative to the image center based on the center coordinates of the target detection box and the center coordinates of the image, and obtains the spatial angle between the deflection angle and the movement direction angle. The expulsion angle correction unit determines the expulsion direction angle based on the relationship between the spatial angle and the preset angle threshold, and outputs the expulsion direction angle to the expulsion module after adjusting it to the set angle range.

[0022] The continuous tracking and speed control module includes a disengagement status judgment unit and a disengagement parameter update unit; The detachment state judgment unit calculates the percentage of the intersection area between the target detection box and the preset direction judgment area frame by frame, and judges whether the target organism has stably deviated from the forward path based on the counting result of the detachment judgment cumulative counter. The deflection parameter update unit compares the travel speed with the minimum effective deflection speed, controls the mobile inspection device to perform deceleration or pause operations, calculates the deviation between the target deflection angle and the deflection direction angle, updates the deflection direction angle when the deviation exceeds the preset threshold, and sends an alarm message to the background monitoring system if the target is not deflected within the set time.

[0023] Please see Figure 1 The AI-based identification and decoy control method for mobile inspection devices includes the following steps: S1. Acquire inspection images from the mobile inspection device, identify target organisms in the inspection images, and acquire the location and orientation information of the target organisms in the inspection images; based on the orientation information of the target organisms in the inspection images, predict the escape direction of the target organisms after they are driven away. S2. Obtain the moving direction of the mobile inspection device. Based on the spatial angle between the predicted escape direction of the target creature and the moving direction of the mobile inspection device, and the positional relationship between the target creature's position information in the inspection image and the preset direction determination area, determine whether the mobile inspection device and the target creature meet the chasing formation conditions. If the chasing formation conditions are met, determine that a chasing state in the same direction has been formed. The chasing formation conditions are that the spatial angle is less than a preset angle threshold, and the target creature's position in the inspection image is within the preset direction determination area. S3. For the determination of forming a chasing state, the driving direction parameter of the driving module is determined according to the escape direction of the target organism after being driven away and the moving direction of the mobile inspection device, so that the spatial angle between the driving direction and the moving direction of the mobile inspection device is greater than or equal to the preset angle threshold; the driving module is controlled to perform the driving action according to the driving direction parameter. S4. After the driving action is executed, the target creature is continuously tracked, and the mobile inspection device is judged based on the real-time relative position between the target creature and the mobile inspection device to determine whether the mobile inspection device and the target creature have broken away from the same-direction chasing state. If they have not broken away, the mobile inspection device's speed is reduced or its movement is stopped, and the driving direction parameters are readjusted until the target creature deviates from the mobile inspection device's forward path, at which point the original inspection speed and original inspection route are restored.

[0024] S1 includes the following: The mobile inspection device acquires inspection images, which are then input into a trained target organism detection model. The target organism detection model outputs the detection results of the target organisms in the inspection images, including the category of the target organisms and the position coordinates of the target detection boxes in the inspection images. The position coordinates of the target detection box in the inspection image are used as the position information of the target organism. The position information includes the planar coordinates of the target organism in the inspection image. The planar coordinates are taken as the origin of the upper left corner of the inspection image, in pixels, and are represented by the center coordinates of the target detection box. Based on the position coordinates of the target detection box in the inspection image, the bounding box region of the target organism is cropped from the inspection image and used as the input region for posture key point extraction. Posture key point extraction is performed on the bounding box region to obtain the head key point coordinates (uh, vh) and torso center point coordinates (ub, vb) of the target organism, where the head key point is the geometric center point of the head region of the target organism, and the torso center point is the geometric center point of the torso region of the target organism. The posture key point extraction is implemented using a pre-trained posture key point recognition model, which takes the bounding box region of the target organism as input and the head key point coordinates and torso center point coordinates as output. Construct an orientation vector Vh in the image plane, where Vh = (uh - ub, vh - vb), where uh and vh are the x and y coordinates of the head keypoint in the image plane, respectively, and ub and vb are the x and y coordinates of the torso center point in the image plane, respectively. The starting point of the orientation vector Vh is the torso center point coordinates (ub, vb), and the ending point is the head keypoint coordinates (uh, vh). The direction of the orientation vector Vh in the image plane is the natural orientation of the target organism. Calculate the orientation angle θh of the orientation vector, where θh = atan2(uh - ub, vh - vb), where atan2 is the arctangent function in the four quadrants, and the value of θh ranges from -180° to 180°, with the horizontal rightward direction of the image as 0° and counterclockwise as positive. Use the orientation angle of the orientation vector as the orientation information of the target organism, and use the orientation information θh as the predicted escape direction of the target organism after being driven away.

[0025] In this embodiment, the mobile inspection device has a built-in visual recognition function, which completes biological detection and pose key point recognition through existing deep learning networks. Biological detection can employ conventional target detection networks such as YOLO, SSD, and Faster R-CNN to identify targets in the inspection images and output the category information of the target organism and the coordinates of the detection box. Pose key point recognition can employ conventional pose estimation networks such as HRNet, OpenPose, and Stacked Hourglass to extract key points from the image region of the target organism and output the coordinates of the organism's head key points and torso center point.

[0026] S2 includes the following: The current movement direction of the mobile inspection device is obtained. The movement direction is provided in real time by the navigation system of the mobile inspection device and is represented by the movement direction angle θm of the device on the horizontal ground. The movement direction angle θm is 0° with due north as the reference direction and clockwise as the reference direction, and the value range is 0° to 360°. The offset angle θc of the current orientation of the inspection device's camera relative to due north is obtained. The offset angle θc is determined by the fixed installation angle of the camera or the real-time rotation angle of the pan-tilt unit. θm is converted to the image plane coordinate system. The conversion relationship is θm' = θm - θc. The converted movement direction angle θm' is 0° with the horizontal rightward direction of the image as the reference direction and counterclockwise as the reference direction. The angle is adjusted to the range of -180° to 180° by adding or subtracting 360°. Based on the predicted escape direction θh of the target organism and the angle θm' of the moving direction of the mobile inspection device after conversion, the angle difference Δθ = θh - θm' is calculated. By adding or subtracting 360°, the angle difference Δθ is adjusted to the range of [-180°, 180°] to obtain Δθn. Then the spatial angle θ1 = |Δθn|, and the value range of θ1 is from 0° to 180°. A rectangular region centered on the geometric center of the inspection image is obtained and used as the preset direction determination region, with a horizontal half-width of Dx and a vertical half-height of Dy. The specific values ​​of Dx and Dy are pre-calibrated based on the inspection image resolution and camera field of view. Based on the obtained target detection box center point coordinates (xc, yc) and the target detection box width w and height h, the ratio P of the intersection area Si of the target detection box and the preset direction determination region to the total area Sb of the target detection box is calculated. Here, the total area Sb of the target detection box = w × h, and the intersection area Si of the target detection box and the preset direction determination region is... =max(0, min(xc+w / 2, W / 2+Dx)-max(xc-w / 2, W / 2-Dx))×max(0, min(yc+h / 2, H / 2+Dy)-max(yc-h / 2, H / 2-Dy)), where W is the width of the inspection image, H is the height of the inspection image, and the ratio P=Si / Sb. When P>P0, the target organism is determined to be located within the preset direction determination area, i.e., within the forward path direction of the mobile inspection device; otherwise, it is determined to be located outside the preset direction determination area. Here, P0 is the preset area ratio threshold. An initial cumulative counter C is set to 0. The chasing formation condition is determined frame by frame. The calculated spatial angle θ1 is compared with a preset angle threshold θ0, which is pre-set based on the mobile inspection device's speed and scene safety requirements. When θ1 < θ0, the first condition is satisfied; otherwise, it is not. The determination of whether the target organism is within a preset direction determination area is used as the second condition. Within a single frame, if both the first and second conditions are satisfied, the cumulative counter C is incremented by 1; if either condition is not satisfied, the cumulative counter C is reset to 0. A preset frame count threshold N is set based on the image acquisition frame rate and determination stability requirements. When the value of the cumulative counter C is greater than or equal to the preset frame count threshold N, the mobile inspection device and the target organism are determined to meet the chasing formation condition, forming a chasing state in the same direction.

[0027] In this embodiment, assuming the inspection image size is 640 pixels × 480 pixels, the horizontal half-width of the preset direction determination area is Dx = 150 pixels, and the vertical half-height is Dy = 120 pixels, then the left boundary of the preset direction determination area is 640 / 2 - 150 = 170, the right boundary is 640 / 2 + 150 = 470, the upper boundary is 480 / 2 - 120 = 120, and the lower boundary is 480 / 2 + 120 = 360. In a certain frame, the center point of the target detection box is (500, 300), the width of the detection box is 100 pixels and the height is 80 pixels. Then the left boundary of the detection box is 500-50=450, the right boundary is 500+50=550, the upper boundary is 300-40=260, and the lower boundary is 300+40=340. Substitute into the formula to calculate: Horizontal overlap width = min(550,470) - max(450,170) = 470 - 450 = 20 pixels; Vertical overlap height = min(340,360) - max(260,120) = 340 - 260 = 80 pixels; The intersection area Si = 20 × 80 = 1600 square pixels; The detection box area Sb = 8000 square pixels; The area percentage P = 1600 / 8000 = 0.2; If the preset area percentage threshold P0 is 0.5, then 0.2 < 0.5, and it is determined that the target organism is not located in the preset direction determination area, that is, the target organism is not in the forward path direction of the mobile inspection device.

[0028] S3 includes the following: To determine if a unidirectional chasing state has formed, the calculated spatial angle θ1 is obtained. Since θ1 < θ0 when a unidirectional chasing state is formed, the driving direction is adjusted so that the spatial angle between the driving direction and the moving direction is greater than or equal to the preset angle threshold θ0. The coordinates of the target detection box center point (xc, yc) and the coordinates of the inspection image center (W / 2, H / 2) are obtained, and the deflection angle θp of the target organism relative to the image center in the image plane is calculated, where θp = atan2(yc-H / 2, xc-W / 2). The value of θp ranges from -180° to 180°, with the image horizontal to the right. θp is the deflection angle of the target organism's current direction relative to the front of the mobile inspection device. It is directly calculated from the center position of the target detection box and represents the spatial orientation of the target organism at the current moment. Unlike the escape direction θh predicted based on the orientation of the organism's head in S1, θp does not involve the prediction of the organism's movement trend, but reflects the actual position and orientation of the target organism in the image plane. When determining the driving direction, θp should be used as the pointing basis for the driving direction, so that the driving module acts towards the current orientation of the target organism, thereby driving the organism laterally away from the direction of movement of the mobile inspection device. Calculate the angle difference Δθp = θp - θm'. Adjust the angle difference Δθp to the range of [-180°, 180°] by adding or subtracting 360°, and take the absolute value to obtain the spatial angle θ2 between the direction of the target organism and the moving direction angle θm'. If θ2 ≥ θ0, the target driving direction angle θd is directly set to θp. If θ2 < θ0, the target driving direction angle θd is determined according to θd = θp + sign(θp - θm') × (θ0 - θ2 + Δ), where sign is the sign function, which takes 1 when the input is greater than 0, -1 when it is less than 0, and 0 when it is equal to 0. Δ is the redundant angle preset according to the control accuracy of the driving mechanism. The obtained θd is adjusted to the range of [-180°, 180°] by adding or subtracting 360°, and the driving module is controlled to perform the driving action according to the adjusted target driving direction angle θd.

[0029] In this embodiment, when θp = θm', the target organism is exactly in front of the direction of movement. At this time, the deflection direction is determined according to the horizontal coordinate of the target organism in the image. If xc > W / 2, the target organism is located to the right of the center of the image, that is, slightly to the right in actual space. The deflection direction should be deflected to the right, and θd = θm' + θ0 + Δ. If xc < W / 2, the target organism is located to the left of the center of the image, that is, slightly to the left in actual space. The deflection direction should be deflected to the left, and θd = θm' - θ0 - Δ. If xc = W / 2, the target organism is strictly located in the exact center of the image, that is, the target direction and the direction of movement are completely coincident and there is no left or right deflection. The default is to deflect to the right, and θd = θm' + θ0 + Δ.

[0030] S4 includes the following: After the driving action is executed, the mobile inspection device continuously tracks the target organism. It achieves tracking by acquiring the coordinates (xc, yc) of the center point of the target detection box frame by frame. Based on the position change of the center point of the target detection box in multiple consecutive frames, it determines whether the target organism has deviated from the forward path of the mobile inspection device. The departure determination cumulative counter Ce is set with an initial value of 0. In each frame of the inspection image, the intersection area ratio P of the target detection box and the preset direction determination area is calculated according to the area ratio calculation formula in S2. When P≤P0, it is determined that the target organism has deviated from the preset direction determination area, and Ce=Ce+1. When P>P0, it is determined that the target organism is still within the preset direction determination area, and Ce=0. A preset threshold M for the number of frames for determining escape is set according to the sensitivity requirements for escape determination. When Ce≥M, it is determined that the target creature has stably deviated from the forward path of the mobile inspection device, the same-direction pursuit state has been released, the driving action is stopped, the mobile inspection device resumes the original inspection speed and original inspection route, and the cumulative counter C in S2 is reset to 0, and the frame-by-frame determination of the pursuit formation condition is re-executed; when Ce<M, it is determined that the target creature has not stably deviated from the preset direction determination area, and the same-direction pursuit state has not been released. The current travel speed v of the mobile inspection device is obtained and compared with the preset minimum effective deportation speed vmin. vmin is preset according to the effective operation conditions of the deportation module. When v > vmin, the mobile inspection device is controlled to reduce its travel speed to vmin. When v ≤ vmin, the mobile inspection device is controlled to pause its movement. The deflection angle θpc of the target organism relative to the center of the image in the current frame is obtained. The angular deviation Δθe between θpc and the target deportation direction angle θd in S3 is calculated. The angular deviation Δθe is adjusted to the range of [-180°, 180°] by adding or subtracting 360° and the absolute value is taken to obtain the deviation angle θdc. θdc is compared with the preset deviation threshold θdt, where θdt is preset according to the target organism's movement speed and the response accuracy of the deportation module. When θdc > θdt, calculate the spatial angle θ2 between the current frame target creature's direction and the movement direction angle θm' in the same way as in S3, and update the target removal direction angle θdn according to θdn = θpc + sign(θpc - θm') × (θ0 - θ2 + Δ). After adjusting θdn to the range of [-180°, 180°] by adding or subtracting 360°, control the removal module to perform the removal action according to the updated removal direction angle. When θdc ≤ θdt, maintain the current removal direction unchanged; continue to execute the detachment judgment process. The preset total removal duration is preset according to the target creature category and scene safety requirements. If Ce does not reach M within the preset total removal duration, send an alarm message to the background monitoring system to prompt manual intervention.

[0031] In this embodiment, the mobile inspection device is set to determine that a chasing state has been formed in the same direction, and the gimbal is controlled to rotate according to the determined target driving direction angle θd=35°. The driving module begins to drive away the target organism in a directional sound wave manner.

[0032] After the driving action is executed, the mobile inspection device continuously tracks the positional changes of the target organism by acquiring the coordinates of the center point of the target detection box frame by frame. In the first to third frames after the driving action is started, the area proportions P are 0.6, 0.55, and 0.48, respectively, all greater than the area proportion threshold P0=0.5. The escape determination cumulative counter Ce remains at 0, indicating that the target organism has not deviated from the preset direction determination area and the same-direction pursuit state is still continuing.

[0033] At this time, the mobile inspection device obtains the current travel speed v=1.2m / s and compares it with the preset minimum effective repulsion speed vmin=0.2m / s. Since v>vmin, the control device reduces the travel speed to 0.2m / s to prolong the time that the repulsion module can act on the target organism, while preventing the device from approaching the target organism too quickly.

[0034] After the device speed decreases, the target organism begins to move to the right. In frames 4 to 7, the area proportions P are 0.45, 0.38, 0.30, and 0.25, respectively, all less than P0=0.5. The cumulative counter Ce for the departure judgment is accumulated to 4 frame by frame. The deflection angle θpc of the target organism relative to the center of the image in the current frame is obtained. Assuming that θpc=25° is calculated, the angular deviation Δθe=25°-35°=-10° between θpc and the current driving direction angle θd=35° is calculated. The deviation angle θdc=10° is obtained, which is greater than the preset deviation threshold θdt=5°, indicating that the position of the target organism has changed significantly and the original driving direction has deviated from the current position of the target organism.

[0035] The control module recalculates the spatial angle between the current frame's target organism's direction and its movement direction angle θm' in the same way as S3: θ2 = |θpc - θm'| = 25°. Since θ2 = 25° is less than θ0 = 30°, the target removal direction angle is updated as follows: θdn=θpc+sign(θpc-θm')×(θ0-θ2+Δ)=25°+1×(30°-25°+5°)=35°, the control module continues to execute the driving action according to the target driving direction angle.

[0036] In frame 8, the target creature's area ratio P further decreases to 0.15, and the escape determination cumulative counter Ce accumulates to 5; the preset escape determination frame number threshold M is 5, and Ce reaches M, indicating that the target creature has stably deviated from the preset direction determination area, and the same-direction pursuit state has been released; the control module stops the driving action, the mobile inspection device resumes the original inspection speed of 1.2m / s and the original inspection route, the cumulative counter C in S2 is reset to 0, and the frame-by-frame determination of the pursuit formation condition is restarted.

[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0038] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-based identification and decoy control method for a mobile inspection device, characterized in that: The method includes the following steps: S1. Acquire inspection images from the mobile inspection device, identify target organisms in the inspection images, and acquire the location and orientation information of the target organisms in the inspection images; based on the orientation information of the target organisms in the inspection images, predict the escape direction of the target organisms after they are driven away. S2. Obtain the moving direction of the mobile inspection device. Based on the spatial angle between the predicted escape direction of the target creature and the moving direction of the mobile inspection device, and the positional relationship between the target creature's position information in the inspection image and the preset direction determination area, determine whether the mobile inspection device and the target creature meet the chasing formation conditions. If the chasing formation conditions are met, determine that a chasing state in the same direction has been formed. The chasing formation conditions are that the spatial angle is less than a preset angle threshold, and the target creature's position in the inspection image is within the preset direction determination area. S3. For the determination of forming a chasing state, the driving direction parameter of the driving module is determined according to the escape direction of the target organism after being driven away and the moving direction of the mobile inspection device, so that the spatial angle between the driving direction and the moving direction of the mobile inspection device is greater than or equal to the preset angle threshold; the driving module is controlled to perform the driving action according to the driving direction parameter. S4. After the driving action is executed, the target creature is continuously tracked, and the mobile inspection device is judged based on the real-time relative position between the target creature and the mobile inspection device to determine whether the mobile inspection device and the target creature have broken away from the same-direction chasing state. If they have not broken away, the mobile inspection device's speed is reduced or its movement is stopped, and the driving direction parameters are readjusted until the target creature deviates from the mobile inspection device's forward path, at which point the original inspection speed and original inspection route are restored.

2. The AI ​​recognition and expulsion control method for the mobile inspection device according to claim 1, characterized in that: S1 includes the following: The mobile inspection device acquires inspection images, which are then input into a trained target organism detection model. The target organism detection model outputs the detection results of the target organisms in the inspection images, including the category of the target organisms and the position coordinates of the target detection boxes in the inspection images. The position coordinates of the target detection box in the inspection image are used as the position information of the target organism. The position information includes the planar coordinates of the target organism in the inspection image. The planar coordinates are taken as the origin of the upper left corner of the inspection image, in pixels, and are represented by the center coordinates of the target detection box. Based on the position coordinates of the target detection box in the inspection image, the bounding box region of the target organism is cropped from the inspection image and used as the input region for posture key point extraction. Posture key point extraction is performed on the bounding box region to obtain the head key point coordinates (uh, vh) and torso center point coordinates (ub, vb) of the target organism, where the head key point is the geometric center point of the head region of the target organism, and the torso center point is the geometric center point of the torso region of the target organism. The posture key point extraction is implemented using a pre-trained posture key point recognition model, which takes the bounding box region of the target organism as input and the head key point coordinates and torso center point coordinates as output. Construct an orientation vector Vh in the image plane, where Vh = (uh - ub, vh - vb), where uh and vh are the x and y coordinates of the head keypoint in the image plane, respectively, and ub and vb are the x and y coordinates of the torso center point in the image plane, respectively. The starting point of the orientation vector Vh is the torso center point coordinates (ub, vb), and the ending point is the head keypoint coordinates (uh, vh). The direction of the orientation vector Vh in the image plane is the natural orientation of the target organism. Calculate the orientation angle θh of the orientation vector, where θh = atan2(uh - ub, vh - vb), where atan2 is the arctangent function in the four quadrants, and the value of θh ranges from -180° to 180°, with the horizontal rightward direction of the image as 0° and counterclockwise as positive. Use the orientation angle of the orientation vector as the orientation information of the target organism, and use the orientation information θh as the predicted escape direction of the target organism after being driven away.

3. The AI ​​recognition and disengagement control method for the mobile inspection device according to claim 2, characterized in that: S2 includes the following: The current movement direction of the mobile inspection device is obtained. The movement direction is provided in real time by the navigation system of the mobile inspection device and is represented by the movement direction angle θm of the device on the horizontal ground. The movement direction angle θm is 0° with due north as the reference direction and clockwise as the reference direction, and the value range is 0° to 360°. The offset angle θc of the current orientation of the inspection device's camera relative to due north is obtained. The offset angle θc is determined by the fixed installation angle of the camera or the real-time rotation angle of the pan-tilt unit. θm is converted to the image plane coordinate system. The conversion relationship is θm' = θm - θc. The converted movement direction angle θm' is 0° with the horizontal rightward direction of the image as the reference direction and counterclockwise as the reference direction. The angle is adjusted to the range of -180° to 180° by adding or subtracting 360°. Based on the predicted escape direction θh of the target organism and the angle θm' of the moving direction of the mobile inspection device after conversion, the angle difference Δθ = θh - θm' is calculated. By adding or subtracting 360°, the angle difference Δθ is adjusted to the range of [-180°, 180°] to obtain Δθn. Then the spatial angle θ1 = |Δθn|, and the value range of θ1 is from 0° to 180°. Obtain a rectangular region centered on the geometric center of the inspection image, and use it as the preset direction determination region, with a horizontal half-width of Dx and a vertical half-height of Dy. Based on the obtained target detection box center point coordinates (xc, yc) and the target detection box width w and height h, calculate the ratio P of the intersection area Si of the target detection box and the preset direction determination region to the total area Sb of the target detection box, where the total area of ​​the target detection box Sb = w × h, and the intersection area Si of the target detection box and the preset direction determination region is calculated as: max(0, min(xc + w / ) 2, W / 2+Dx)-max(xc-w / 2, W / 2-Dx))×max(0, min(yc+h / 2, H / 2+Dy)-max(yc-h / 2, H / 2-Dy)), where W is the width of the inspection image, H is the height of the inspection image, and the ratio P=Si / Sb. When P>P0, the target organism is determined to be located within the preset direction determination area, i.e., within the forward path direction of the mobile inspection device; otherwise, it is determined to be located outside the preset direction determination area; where P0 is the preset area ratio threshold. An initial cumulative counter C is set to 0. The chasing formation condition is determined frame by frame. The calculated spatial angle θ1 is compared with a preset angle threshold θ0, which is pre-set based on the mobile inspection device's speed and scene safety requirements. When θ1 < θ0, the first condition is satisfied; otherwise, it is not. The determination of whether the target organism is within a preset direction determination area is used as the second condition. Within a single frame, if both the first and second conditions are satisfied, the cumulative counter C is incremented by 1; if either condition is not satisfied, the cumulative counter C is reset to 0. A preset frame count threshold N is set based on the image acquisition frame rate and determination stability requirements. When the value of the cumulative counter C is greater than or equal to the preset frame count threshold N, the mobile inspection device and the target organism are determined to meet the chasing formation condition, forming a chasing state in the same direction.

4. The AI ​​recognition and expulsion control method for the mobile inspection device according to claim 3, characterized in that: S3 includes the following: To determine if a unidirectional chasing state is formed, the calculated spatial angle θ1 is obtained. Since θ1 < θ0 when a unidirectional chasing state is formed, the driving direction is adjusted so that the spatial angle between the driving direction and the moving direction is greater than or equal to the preset angle threshold θ0. The coordinates of the center point of the target detection box (xc, yc) and the coordinates of the center of the inspection image (W / 2, H / 2) are obtained, and the deflection angle θp of the target organism relative to the image center in the image plane is calculated, and θp = atan2(yc-H / 2, xc-W / 2). The value of θp ranges from -180° to 180°, with the horizontal rightward direction of the image as 0° and counterclockwise as positive. Calculate the angle difference Δθp = θp - θm'. Adjust the angle difference Δθp to the range of [-180°, 180°] by adding or subtracting 360°, and take the absolute value to obtain the spatial angle θ2 between the direction of the target organism and the moving direction angle θm'. If θ2 ≥ θ0, the target driving direction angle θd is directly set to θp. If θ2 < θ0, the target driving direction angle θd is determined according to θd = θp + sign(θp - θm') × (θ0 - θ2 + Δ), where sign is the sign function, which takes 1 when the input is greater than 0, -1 when it is less than 0, and 0 when it is equal to 0. Δ is the redundant angle preset according to the control accuracy of the driving mechanism. The obtained θd is adjusted to the range of [-180°, 180°] by adding or subtracting 360°, and the driving module is controlled to perform the driving action according to the adjusted target driving direction angle θd.

5. The AI ​​recognition and expulsion control method for the mobile inspection device according to claim 4, characterized in that: S4 includes the following: After the driving action is executed, the mobile inspection device continuously tracks the target organism. It achieves tracking by acquiring the coordinates (xc, yc) of the center point of the target detection box frame by frame. Based on the position change of the center point of the target detection box in multiple consecutive frames, it determines whether the target organism has deviated from the forward path of the mobile inspection device. The departure determination cumulative counter Ce is set with an initial value of 0. In each frame of the inspection image, the intersection area ratio P of the target detection box and the preset direction determination area is calculated according to the area ratio calculation formula in S2. When P≤P0, it is determined that the target organism has deviated from the preset direction determination area, and Ce=Ce+1. When P>P0, it is determined that the target organism is still within the preset direction determination area, and Ce=0. The threshold M for the number of frames for detachment determination is preset. M is set in advance according to the sensitivity requirements for detachment determination. When Ce≥M, it is determined that the target organism has stably deviated from the forward path of the mobile inspection device, the same-direction pursuit state has been released, the driving action is stopped, the mobile inspection device resumes the original inspection speed and the original inspection route, and the cumulative counter C in S2 is reset to 0, and the frame-by-frame determination of the pursuit formation condition is re-executed. When Ce < M, it is determined that the target organism has not yet stably deviated from the preset direction determination area, and the same-direction pursuit state has not been lifted; The current travel speed v of the mobile inspection device is obtained and compared with the preset minimum effective deportation speed vmin. vmin is preset according to the effective action conditions of the deportation module. When v > vmin, the mobile inspection device is controlled to reduce its travel speed to vmin. When v ≤ vmin, the mobile inspection device is controlled to pause its movement. The deflection angle θpc of the target organism relative to the image center in the current frame is obtained. The angular deviation Δθe between θpc and the target deportation direction angle θd in S3 is calculated. The angular deviation Δθe is adjusted to the range of [-180°, 180°] by adding or subtracting 360°, and the absolute value is taken to obtain the deviation angle θdc. θdc is then compared with... A preset deviation threshold θdt is compared, where θdt is preset based on the target organism's movement speed and the response accuracy of the expulsion module. When θdc > θdt, the spatial angle θ2 between the current frame's target organism's direction and the movement direction angle θm' is calculated in the same way as in S3. The target expulsion direction angle θdn is updated according to θdn = θpc + sign(θpc - θm') × (θ0 - θ2 + Δ). After adjusting θdn to the range of [-180°, 180°] by adding or subtracting 360°, the expulsion module is controlled to perform the expulsion action according to the updated expulsion direction angle. When θdc ≤ θdt, the current expulsion direction remains unchanged. Continue executing the detachment determination process. The preset total detachment duration is pre-set based on the target organism type and scene security requirements. If Ce fails to reach M within the preset total detachment duration, an alarm message is sent to the background monitoring system to prompt manual intervention.

6. An AI recognition and expulsion control system for a mobile inspection device, applied to the AI ​​recognition and expulsion control method for a mobile inspection device as described in any one of claims 1-5, characterized in that: The system includes: an image acquisition module, a biometric recognition and orientation analysis module, a same-direction chasing judgment module, a drive-away direction execution module, and a continuous tracking and speed adjustment module; The image acquisition module acquires inspection images in real time during the inspection process of the mobile inspection device; The biometric identification and orientation analysis module identifies target organisms in the inspection images, obtains the location and orientation information of the target organisms, and predicts the escape direction of the target organisms after they are driven away based on the orientation information. The same-direction pursuit judgment module obtains the moving direction of the mobile inspection device, and determines whether a same-direction pursuit state has been formed based on the spatial angle between the escape direction and the moving direction, as well as the positional relationship between the target organism and the preset direction judgment area. When a chasing state is formed in the same direction, the driving direction execution module determines the driving direction parameters of the driving module, so that the spatial angle between the driving direction and the moving direction is greater than or equal to a preset angle threshold, and controls the driving module to perform the driving action. The continuous tracking and speed adjustment module continuously tracks the target organism during the driving process, determines whether the target organism has broken away from the same-direction chasing state, and reduces the moving speed of the mobile inspection device or stops moving and readjusts the driving direction parameters if the target organism has not broken away. After the target organism deviates from the forward path, the original inspection speed and the original inspection route are restored.

7. The AI ​​recognition and disengagement control system for the mobile inspection device according to claim 6, characterized in that: The biometrics and orientation analysis module includes a target detection unit and an orientation extraction unit; The target detection unit inputs the inspection image into the trained target organism detection model and outputs the category of the target organism and the position coordinates of the target detection box in the inspection image; The posture orientation extraction unit trims the bounding box region of the target organism according to the target detection box, obtains the coordinates of the head key points and the center point of the torso through the posture key point recognition model, constructs the orientation vector and calculates the direction angle of the orientation vector, and determines the direction angle as the predicted escape direction.

8. The AI ​​recognition and disengagement control system for the mobile inspection device according to claim 6, characterized in that: The same-direction chasing judgment module includes a coordinate angle transformation unit and a dual-condition counting unit; The coordinate angle conversion unit obtains the motion direction angle of the mobile inspection device and the offset angle of the camera, converts the motion direction angle to the image plane coordinate system and adjusts it to the set angle range; The dual-condition counting unit calculates the spatial angle between the escape direction and the movement direction, determines the proportion of the intersection area between the target detection box and the preset direction judgment area, and judges whether a chasing state in the same direction is formed based on the continuous counting results of the cumulative counter.

9. The AI ​​recognition and disengagement control system for the mobile inspection device according to claim 6, characterized in that: The drive-off direction execution module includes a target orientation calculation unit and a drive-off angle correction unit; The target orientation calculation unit calculates the deflection angle of the target organism relative to the image center based on the center coordinates of the target detection box and the center coordinates of the image, and obtains the spatial angle between the deflection angle and the movement direction angle. The expulsion angle correction unit determines the expulsion direction angle based on the relationship between the spatial angle and the preset angle threshold, adjusts the expulsion direction angle to the set angle range, and then outputs it to the expulsion module.

10. The AI ​​recognition and disengagement control system for the mobile inspection device according to claim 6, characterized in that: The continuous tracking and speed adjustment module includes a disengagement state judgment unit and a disengagement parameter update unit; The detachment state judgment unit calculates the percentage of the intersection area between the target detection box and the preset direction judgment area frame by frame, and judges whether the target organism has stably deviated from the forward path based on the counting result of the detachment judgment cumulative counter. The deflection parameter update unit compares the travel speed with the minimum effective deflection speed, controls the mobile inspection device to perform a speed reduction or pause operation, calculates the deviation between the target deflection angle and the deflection direction angle, updates the deflection direction angle when the deviation exceeds a preset threshold, and sends an alarm message to the background monitoring system if the target is not deflected within a set time.

Images