Self-adaptive camera pose intelligent adjusting system

Abstract

The invention relates to the technical field of intelligent control, in particular to a self-adaptive camera pose intelligent adjusting system which comprises a dynamic recognition module, a pose adjusting module, a mechanism control module, an information filtering module and a feedback association module. According to the method, the foreground pixel brightness gradient and the boundary continuity in the image frame are extracted, the target displacement path can be recognized, the tracking information is constructed, the target coherent perception under rapid scene change is guaranteed, the attitude parameters and the target motion direction and speed are combined, the offset correction instruction is generated, the response stability of the dynamic target is improved, and the target tracking accuracy is improved. The execution structure feedback is used for correcting control output, reducing the driving error accumulation risk, eliminating non-target disturbance through image boundary definition and motion consistency, improving data purity and combining a view angle execution state and path information to realize synchronous control of tracking and view angle adjustment; and the response efficiency and the following robustness of the system in a dynamic environment are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and particularly to an intelligent adaptive camera pose adjustment system. Background Art

[0002] The technical field of intelligent control includes aspects such as the automated management, state recognition, and response adjustment of devices and processes. The core of this technical field lies in obtaining external environment or internal system state information through sensing elements, performing data analysis and judgment based on preset logic or models, and thus driving the actuator to complete corresponding operations. This field systematically involves multiple links such as sensor signal acquisition, control logic calculation, and drive execution response, and is widely applied in multiple application scenarios such as industrial automation, smart home, traffic control, and robot navigation. Its main goal is to improve the stability, efficiency, and intelligent response ability of system operation.

[0003] Among them, an intelligent adaptive camera pose adjustment system refers to a control system that can automatically adjust the camera pose according to the environment or target state. In response to problems such as shooting angle deviation, viewing dead angle, or target tracking deviation of the camera in a dynamic scenario, the system completes the adjustment of the camera pose through methods such as image information analysis, spatial geometric parameter calculation, and coordinated control of the execution device. The system judges the rationality of the current camera viewing angle based on the image edge recognition and target contour judgment method, then corrects the three-dimensional pose parameters in combination with the angle change law and position change trajectory, and completes the automatic adjustment of the camera pitch angle, horizontal rotation angle, and zoom distance through a multi-axis drive actuator, realizing continuous viewing angle following and pose optimization of the target.

[0004] In the process of automatically adjusting the camera viewing angle in the prior art, it mainly relies on the static analysis of the single-frame image edge and target contour, lacking the in-depth analysis of the continuous movement path of the target in the image sequence, and it is difficult to maintain the stability of target locking in complex background or multi-target interference scenarios. In terms of pose parameter adjustment, the existing system corrects based on a preset angle change model, failing to fully combine the movement direction and speed change of the target, resulting in response lag or even viewing angle loss when quickly switching scenes or the target suddenly moves. The feedback execution link fails to collect the real-time response state of the execution structure in a closed-loop manner, causing the drive deviation to gradually accumulate during continuous adjustment, manifested as slight drift or even distortion in the camera pose adjustment. The ability to dynamically separate the foreground and background regions in the image information is weak, often resulting in incorrect recognition due to the movement interference of non-target objects in the background, reducing the accuracy of target following. For example, in complex street scene monitoring, multiple targets such as pedestrians and vehicles appear simultaneously, and the system cannot eliminate the disturbance of non-critical targets, resulting in frequent jumps in pose adjustment, affecting video continuity and analysis stability, and restricting the improvement of the real-time performance, robustness, and control accuracy of the prior art in a dynamic environment. Summary of the Invention

[0005] The object of the present invention is to solve the disadvantages existing in the prior art, and to propose an intelligent adaptive camera pose adjustment system.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An intelligent adaptive camera pose adjustment system includes: The dynamic recognition module obtains camera pose data, extracts the video frames captured by the image acquisition component, extracts the pixel brightness gradient and boundary contour continuity of the foreground area in the frame, recognizes the target position transformation data of the front and rear frames within the image stabilization structure, judges the motion trend of the area block and recognizes the continuous motion path, and generates target tracking path information; The pose adjustment module, based on the target tracking path information, extracts the current pitch and rotation angle values of the camera, combines the target movement direction and speed interval, analyzes the pose deviation degree and the required angle adjustment amplitude, and generates a pose correction instruction set; The mechanism control module, according to the pose correction instruction set, collects the current response state of the execution structure, compares the target angle and displacement requirements in the instruction, recognizes the driving error range, adjusts the output amplitude and direction of the controller, and obtains the mechanism execution adjustment value; The information filtering module calls the mechanism execution adjustment value, extracts the boundary clarity and motion consistency features of the foreground and background areas in the image, recognizes the proportion of the non-target dynamic area, and eliminates the boundary noise area, and generates a background disturbance exclusion result.

[0007] As a further solution of the present invention, the target tracking path information includes the position change amplitude, the trajectory curvature feature, and the displacement direction angle, the pose correction instruction set includes the pitch angle adjustment amount, the rotation angle correction value, and the angle deviation threshold, the mechanism execution adjustment value includes the controller output gain, the rotation angle of the execution structure, and the real-time response displacement, and the background disturbance exclusion result includes the foreground boundary clarity score, the background area consistency coefficient, and the dynamic interference ratio.

[0008] As a further solution of the present invention, the dynamic recognition module includes: The pose estimation sub-module obtains camera pose data, extracts the three-dimensional pose data and the central pixel coordinates of the camera for each frame, analyzes the pose rotation angle difference and the translation vector change rate in the continuous frames, judges whether the change exceeds the pose stability threshold, and generates a pose adjustment trigger signal; The foreground perception sub-module, based on the pose adjustment trigger signal, extracts the foreground brightness gradient and boundary contour continuity of the current frame, combines the spatial offset and direction change of the target edge points of the front and rear frames, judges the perception integrity trend of the foreground area under the pose change, and obtains the image perception integrity parameter; The path linkage sub-module dynamically corrects the current pose according to the image perception integrity parameter, adjusts the focus orientation and rotation angle of the camera, calls the offset difference between the position of the center point of the corrected image and the original image, and obtains the pose adjustment execution path.

[0009] As a further solution of the present invention, the pose adjustment module includes: The angle extraction sub-module extracts the pitch angle and rotation angle data in the image frame based on the target tracking path information, combines the pixel matrix change and the image edge position sequence, identifies the pitch angle and rotation angle, and obtains the angle change numerical set; The direction matching sub-module calls the angle change numerical set, detects the direction consistency and speed change trend according to the target movement direction vector and speed interval, calculates the pose direction offset intensity value, and combines the direction matching result and the speed jump interval to obtain the direction offset adjustment parameter value; The instruction generation sub-module calls the direction offset adjustment parameter value, divides the gears according to the offset amplitude and speed interval, determines whether it is in the adjustment response interval, maps the correction action parameters, and generates a pose correction instruction set.

[0010] As a further solution of the present invention, the mechanism control module includes: The state recognition sub-module collects the current response state of the execution structure according to the pose correction instruction set, extracts the attitude angle vector and displacement component, identifies the dimensional deviation distance, and generates the attitude displacement difference amount; The error evaluation sub-module calls the attitude displacement difference amount, judges the matching offset, extracts the dynamic change amplitude and response trend in the response state, compares the change rate and the response rate threshold, analyzes the driving precision difference interval, and obtains the driving error range value; The control adjustment sub-module calls the driving error range value, combines the angle response sensitivity, displacement response lag amplitude, and adjustment feedback period amplitude, corrects the adjustment amplitude and direction output by the controller, calculates the control output correction amplitude, corrects the controller signal, and obtains the mechanism execution adjustment value.

[0011] As a further solution of the present invention, the information filtering module includes: The clarity extraction sub-module calls the mechanism execution adjustment value, extracts the gray gradient of the image boundary, screens the pixel points that meet the clarity threshold, and obtains the boundary clarity change rate; The dynamic ratio judgment sub-module calls the boundary clarity change rate, detects the consistency of the regional motion vector, counts the area ratio of the non-target area, and obtains the non-target area occupancy ratio; The disturbance elimination sub-module collects the boundary noise pixel distribution according to the non-target area occupancy ratio, performs a comparison by combining the foreground and background gray-scale differences and the error amplitude, calculates the disturbance elimination coefficient, filters out the noise area, extracts the mask layer, and generates the background disturbance elimination result.

[0012] As a further solution of the present invention, the system further includes a feedback association module: Based on the background disturbance elimination result, the feedback association module combines the updated target tracking path and the current pose execution state, evaluates the response synchronization between the camera view movement and the target path, and generates the camera pose response matching degree; The camera pose response matching degree includes the target path tracking deviation, the pose adjustment delay amount, and the synchronization coordination index.

[0013] As a further solution of the present invention, the feedback association module includes: Based on the background disturbance elimination result, the view angle offset extraction sub-module extracts the updated camera pose sequence and the starting pose data after background disturbance elimination, analyzes the attitude Euler angle difference and displacement change between consecutive frames, and performs normalization processing in combination with the time stamp to obtain the camera view angle offset trend; According to the camera view angle offset trend and the spatial direction change sequence of adjacent pose segments in the target tracking path, the path linkage comparison sub-module analyzes the cosine value of the included angle between the direction vectors and the velocity amplitude change rate at the corresponding time points, and judges the response lag situation in the synchronization interval to obtain the path response synchronization feature; The adjustment strategy generation sub-module calls the path response synchronization feature and the camera speed change rate at the previous moment, sets the weight parameter according to the response lag time difference and the included angle offset degree, performs weighted processing on each pose adjustment amount, and generates the camera pose response matching degree.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by dynamically extracting the pixel brightness gradient and boundary contour continuity of the foreground region in the image frame, the target displacement path in the stable structure can be identified with high precision, and the real-time tracking path information can be constructed according to the continuous movement trend of the target, so as to maintain a coherent perception of the target position under rapid scene changes, extract the current attitude parameters of the camera, and combine the target movement direction and speed range to generate an offset correction instruction for the pitch and rotation angles, enabling the camera to quickly adapt to the target behavior changes and improve the following stability of the dynamic target. Responding to the feedback data of the response execution structure, comparing the expected and actual angle deviations, and further adjusting the control output amplitude and direction, effectively avoiding the control drift caused by the accumulation of driving errors. Based on the boundary clarity and motion consistency characteristics of the image foreground and background regions, the disturbance effects of non-target regions are eliminated, the specificity of target recognition is strengthened, the data purity of pose adjustment decisions is ensured, and the synchronization degree of the perspective response is evaluated by combining the actual execution state of the camera and the updated information of the target path, realizing real-time closed-loop control between the tracking path and the perspective adjustment. The above operations act together in a multi-level link of target capture, attitude correction, response adjustment and interference elimination, greatly improving the system's rapid recognition ability for dynamic scenes, the accuracy of real-time attitude adjustment, and the robustness and visual consistency during the target following process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the dynamic recognition module in the present invention; Figure 3 is the flow chart of the attitude adjustment module in the present invention; Figure 4 is the flow chart of the mechanism control module in the present invention; Figure 5 is the flow chart of the information filtering module in the present invention; Figure 6 is the flow chart of the feedback association module in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Please refer to Figure 1 , an adaptive camera pose intelligent adjustment system includes: The dynamic recognition module acquires camera pose data, extracts the video frames captured by the image acquisition component, extracts the pixel brightness gradient and boundary contour continuity of the foreground area in the frame, identifies the target position transformation data between the front and rear frames within the image stabilization structure, determines the motion trend of the region block and identifies the continuous motion path, and generates target tracking path information; The pose adjustment module, based on the target tracking path information, extracts the current pitch and rotation angle values of the camera, combines the target movement direction and speed range, analyzes the pose deviation degree and the required angle adjustment amplitude, and generates a pose correction instruction set; The mechanism control module, according to the pose correction instruction set, acquires the current response state of the execution structure, compares the target angle and displacement requirements in the instruction, identifies the driving error range, adjusts the output amplitude and direction of the controller, and obtains the mechanism execution adjustment value; The information filtering module calls the mechanism execution adjustment value, extracts the boundary clarity and motion consistency features of the foreground and background regions in the image, identifies the proportion of non-target dynamic regions, and eliminates the boundary noise regions, and generates a background disturbance exclusion result; The feedback correlation module, based on the background disturbance exclusion result, combines the updated target tracking path and the current pose execution state, evaluates the response synchronization between the camera view movement and the target path, and generates a camera pose response matching degree.

[0019] The target tracking path information includes the position change amplitude, trajectory curvature characteristics, and displacement direction angle. The pose correction instruction set includes the pitch angle adjustment amount, rotation angle correction value, and angle offset threshold. The mechanism execution adjustment value includes the controller output gain, the rotation angle of the execution structure, and the real-time response displacement. The background disturbance exclusion result includes the foreground boundary clarity score, the background region consistency coefficient, and the dynamic interference ratio. The camera pose response matching degree includes the target path tracking deviation, the pose adjustment delay amount, and the synchronization coordination index.

[0020] Please refer to Figure 2 , the dynamic recognition module includes: The pose estimation sub-module obtains the camera pose data, extracts the three-dimensional camera pose data and the central pixel coordinates of each frame, analyzes the difference in pose rotation angles and the change rate of translation vectors in consecutive frames, determines whether the change exceeds the pose stability threshold, and generates a pose adjustment trigger signal; When obtaining the three-dimensional camera pose data, it is necessary to first obtain the camera position and rotation angle of each frame of image, and use sensors (such as IMU) and computer vision methods to obtain the data. For example, using visual SLAM technology, the pose is calculated by matching the image feature points between consecutive frames. Specifically, for each frame of image, the feature points are extracted, and the least squares optimization algorithm is used to calculate the camera pose of the current frame, including the three-dimensional translation vector and the rotation matrix. After extracting the pose data, it is necessary to calculate the difference in rotation angles and the difference in translation vectors between adjacent frames. The rotation angle can be represented by quaternions or Euler angles, and the translation vector is directly calculated by the three-dimensional coordinate difference between adjacent frames. Using a preset stability threshold, for example, when the difference in rotation angles is greater than 0.5 degrees and the difference in translation vectors is greater than 0.1 meters, it is considered that a significant change has occurred. If it is detected that the pose change exceeds the threshold, the pose adjustment signal is triggered. At this time, the camera position and orientation can be re-evaluated to ensure that the camera operates in a stable position. For example, if the difference in rotation angles between two consecutive frames is 0.6 degrees and the translation difference is 0.12 meters, the pose adjustment signal is triggered.

[0021] Based on the pose adjustment trigger signal, the foreground perception sub-module extracts the foreground brightness gradient and boundary contour continuity of the current frame, combines the spatial offset and direction change of the target edge points between the front and back frames, judges the perception integrity trend of the foreground area under pose changes, and obtains the image perception integrity parameter; Analyze the foreground brightness gradient and boundary contour continuity. According to the trigger signal of the pose change, extract the foreground area of the current frame, and then calculate the brightness gradient of each pixel in the foreground area, which can be achieved through image convolution operations. Use the Sobel operator or Canny edge detection to obtain the gradient value of the brightness change, analyze the boundary contour of the foreground area, use the image contour detection algorithm, such as Canny edge detection, to extract the contour edge and calculate the continuity of the contour. If the connection degree between boundary points is weak, it indicates that the perception continuity of the foreground area is poor. On this basis, combine the spatial offset and direction change of the target edge points between the front and back frames for multi-frame analysis, and judge the perception integrity trend of the foreground area under pose changes. For example, if the spatial displacement of the edge points in the front and back frames exceeds 0.15 meters and the direction of the edge changes greatly, it indicates that the integrity of the foreground perception decreases, and the image perception integrity parameter will be updated to reflect the perception loss of the foreground area.

[0022] The path linkage sub-module dynamically corrects the current pose according to the image perception integrity parameter, adjusts the focus orientation and rotation angle of the camera, calls the offset difference between the position of the center point of the corrected image and the original image, and obtains the execution path for pose adjustment; To dynamically correct the current pose of the camera according to the image perception integrity parameter, it is first necessary to judge whether the orientation of the camera has changed significantly according to the analysis results of the aforementioned image perception integrity. If the image perception integrity is low, it means that the current pose needs to be adjusted, and the adjustment mechanism is triggered according to the value of the perception integrity (for example, when the integrity is lower than 50%). The adjustment of the camera orientation can be achieved by controlling the rotation angle, and the change of the rotation angle is based on the change amplitude of the perception integrity. If the integrity drops by more than a certain threshold (such as a 20% drop), the rotation angle is adjusted through a dynamic algorithm. The correction is made according to the center point and offset difference of the image. Using the offset difference between the position of the center point of the adjusted image and the original image, a new correction path is calculated. For example, if the offset of the original image is 0.05 meters and the offset of the center point of the adjusted image is 0.07 meters, the difference in path adjustment is 0.02 meters, which means that the camera needs to adjust its position along this path to obtain the execution path for pose adjustment, so as to ensure the improvement of the image quality and perception integrity after pose adjustment.

[0023] Please refer to Figure 3 , the pose adjustment module includes: The angle extraction sub-module extracts the pitch angle and rotation angle data in the image frame based on the target tracking path information, combines the pixel matrix change and the image edge position sequence, identifies the pitch angle and rotation angle, and obtains the angle change value set; It is calculated by analyzing the spatial position change of the target in consecutive video frames. The data is used to understand how the target moves in three-dimensional space. For example, in security monitoring, the camera needs to adjust its angle in real time to keep the focus on the intruder. First, the edge and pixel matrix change data will be extracted from each frame of the image, which involves edge detection and optical flow estimation in image processing technology. Through this technology, the moving objects in the image and their speed and direction can be identified. By analyzing the change trend and position data in the image sequence, the change values of the pitch angle and rotation angle of the target are calculated to adjust the pitch and rotation angles of the camera in real time, ensuring that the camera's viewing angle is aligned with the moving target. For example, in a security scenario, if a person walks through the monitoring area from left to right, the rotation angle of the camera will be adjusted according to the person's moving speed and trajectory to keep the target in the center position of the field of view, generating an angle change value set, which is a series of values of pitch and rotation angles, reflecting how the camera should adjust its angle to track the target from one time point to another.

[0024] The direction matching sub-module calls the set of angular change values, and according to the target movement direction vector and the speed interval, detects the direction consistency and the speed change trend, using the formula: ; Calculate the attitude direction offset intensity value, and combine the direction matching result with the speed jump interval to obtain the direction offset adjustment parameter value; Among them, represents the attitude direction offset intensity value, represents the path speed expected value, represents the current speed value, is the expected direction angle, is the current rotation angle change value, is the expected pitch angle, is the current pitch angle change value, represents the path frame 's position value, represents the path frame 's position value, is the number of path frames; After calling the set of angular change values, it is necessary to numerically express the changes in the pitch angle and rotation angle of the camera in the current frame, and perform a linkage analysis with data such as the actual movement direction and speed difference of the target to form the attitude direction offset intensity value. The target speed can be obtained by dividing the change in the path node position between consecutive frames by the time interval. Let the path node position points , , the sampling interval , then the speed calculation is , for example, when , then ; The current speed comes from the real-time calculation of the instantaneous change amount of the path. Collect the spatial position difference between the front and rear frames at the current time point . Let the front and rear frames be respectively and , and the sampling interval is still , then ; The expected pitch angle is set by calculating the simulation path of the task target trajectory. Here it is , the expected rotation angle , the current change value of the camera is , , then the angle difference part is calculated as ; The change between the position values of consecutive path frame nodes is accumulated by the absolute value of the frame difference and then averaged; Suppose there is a node sequence, ; Then , the average path change is ; Substitute the above data into the formula: ; The result shows that the attitude direction offset intensity value is 7.12, indicating that there is a significant offset trend in the path direction of the current frame. It is necessary to enter the adjustment process to match the corresponding offset correction instruction set. In this calculation, the units of all parameters are unified as SI units: the speed adopts , and the angle is uniformly converted to dimensionless when input in degrees for geometric difference operations. The spatial position is as the basic unit, and the final result is the comprehensive value of the offset intensity, and the dimension is scalar without unit; By introducing the coupling expression of three physical variables: speed difference, direction angle error, and path continuity change, the dependence on single angle or speed data is avoided, and the response adaptability in the complex background of dynamic trajectories is enhanced.

[0025] The instruction generation sub-module calls the direction offset adjustment parameter value, divides the gears according to the offset amplitude and speed interval, judges whether it is in the adjustment response interval, maps the correction action parameters, and generates the pose correction instruction set; Convert the direction offset adjustment parameter value into specific camera adjustment instructions. The parameters are calculated and provided by the previous sub-module. Evaluate whether it is necessary to adjust the pose of the camera according to the offset parameters and the current speed interval. This involves judging whether the current offset exceeds the preset response threshold. For example, in urban traffic monitoring, if it is detected that a pedestrian or vehicle suddenly changes direction, the camera needs to quickly adjust its angle to track the target. Select the appropriate adjustment gear and parameter mapping. The gears are preset according to different speeds and offsets to optimize the reaction speed and accuracy of the camera. For example, in a mall monitoring, when it is detected that a child is moving in the crowd, the camera will select a more sensitive adjustment gear to track the fast-moving small target, generate the pose correction instruction set, and guide the camera to make physical position adjustments to ensure that the camera can effectively track and record the dynamic changes in the scene.

[0026] Please refer to Figure 4 , the mechanism control module includes: The state recognition sub-module collects the current response state of the execution structure according to the pose correction instruction set, extracts the attitude angle vector and displacement component, identifies the dimension deviation distance, and generates the attitude displacement difference; The starting position is based on the target angle and displacement requirements. By calculating the deviation distances in each dimension, this process is for the position adjustment scenario of the robotic arm in an automated assembly line. During implementation, the robotic arm first collects the current pose data, such as angles and position coordinates, against the preset target state. Then, it calculates the differences between the current state and the target state through real-time sensor feedback. The calculation includes the angular differences and displacement differences in each axis, and the specific values of each difference are determined through algorithms. For example, the angular error is calculated as Δθ, and the displacement errors are Δx, Δy, and Δz. The data is then used to adjust the control instructions to ensure that the robotic arm can accurately reach the preset position. Through continuous monitoring and immediate adjustment, the error rate can be significantly reduced, and the assembly accuracy can be improved, generating the pose displacement difference quantity, which is a specific numerical data representing the specific angles and displacements that the robotic arm needs to adjust to ensure consistency with the predetermined target.

[0027] The error evaluation sub-module calls the pose displacement difference quantity, judges the matching offset, extracts the dynamic change amplitude and response trend in the response state, compares the change rate with the response rate threshold, analyzes the driving precision difference interval, and obtains the driving error range value; The starting position calls the pose displacement difference quantity. For the error correction scenario in automated navigation, first, the obtained pose displacement difference quantity is used for preliminary analysis. How to adjust its algorithm parameters according to the magnitude of the quantity. For example, if the displacement difference quantity exceeds the preset response rate threshold, the adjustment ratio and direction need to be calculated, and then the path planning is optimized through the dynamic adjustment algorithm. This adjustment is not only based on real-time data but also considers the original data pattern to predict future trends and errors, thereby achieving more accurate navigation. The calculations in this process involve multiple parameters, such as thresholds, change rates, and quantization criteria for continuous trends. Through actual data-driven simulations, the real-time update and application of parameters are ensured. Finally, the generated driving error range value is an accurate data indicating the optimal error range that can be achieved under the current technical conditions.

[0028] The control adjustment sub-module calls the driving error range value, combines the angular response sensitivity, displacement response lag amplitude, and adjustment feedback period amplitude, and corrects the adjustment amplitude and direction of the controller output. Using the formula: ; Calculate the control output correction amplitude, correct the controller signal, and obtain the mechanism execution adjustment value; Among them, represents the control output correction amplitude, represents the driving error range value, represents the angular response sensitivity, represents the displacement response lag amplitude, represents the direction correction weight vector, represents the feedback period Adjustment feedback cycle amplitude Represents the feedback cycle Table response lag compensation ratio Is the number dimension of the feedback cycle; The starting position calls the drive error range value, combines the angular response sensitivity, displacement response lag amplitude, and adjustment feedback cycle amplitude to construct a control factor, corrects the adjustment amplitude and direction of the controller output, and uses a formula for calculation to obtain the control output correction amplitude E. This process is applied to dynamic adjustment scenarios such as real-time pose control of a camera in adaptive photography. To ensure the consistency of operations between parameters, first perform dimensional unification processing on all parameters, unify the angular unit to radians, the displacement unit to millimeters, and time-related parameters to seconds. The following separately describes the acquisition methods and example values of each parameter, and substitutes them into the formula for complete calculation: A: Drive error range value, which represents the comprehensive measure of the current pose deviation of the camera from the target pose. It is jointly determined by the angular difference and displacement difference, and is obtained by calculating the differences on each axis between the target and actual responses and performing square sum averaging processing. Assuming the actual angular deviation is 0.07 rad and the displacement deviation is 1.5 mm, then set A = 1.85 according to the comprehensive weighting rule (unified to numerical dimension); R: Angular response sensitivity, which is obtained by the change in angular response caused by a unit input excitation. After adjusting the control input by 0.1 unit, the angular change is 0.015 rad, so R = 0.015; B: Displacement response lag amplitude, which represents the hysteresis amount of the displacement response after the control output changes, and can be measured by the average response delay per unit control output. If the average response lag is 2.1 mm, set B = 2.1; F: Direction correction weight vector, which is used to adjust the error deflection direction and is quantified by the clustering result of the original response bias characteristics. The main bias direction can be quantified as 1.2; : Adjustment feedback cycle amplitude, which represents the change amount of the control signal within each feedback cycle. Set 5 consecutive feedback cycles for the system, and the control amount adjustment for each cycle is 0.8, 0.9, 1.0, 1.1, 1.2 respectively, with the unit unified to unit signal strength; : Response lag compensation ratio, which represents the lag error compensation amount per cycle, in sequence of 0.3, 0.4, 0.35, 0.45, 0.5, with the unit being error compensation unit; The steps are as follows: Calculate the square root term: ; Weighted term: (A + square root term) ; Multiply by F: ; Calculate the denominator: ; Finally, obtain the correction amplitude of the control output ; This result indicates that in the current response state, the controller output should be dynamically adjusted by an amplitude of 0.677 units and maintain the pose correction in the direction indicated by the direction correction vector F to obtain the adjustment value executed by the mechanism; The advantage of the formula is that by integrating the angle response sensitivity R and the displacement lag amplitude B into the square sum and square root operations, the sensitivity of the system to the two deviation sources is enhanced, and with the actual control input and the compensation coefficient as the benchmark for balancing, it can dynamically adapt the output signal strength in the multi-cycle feedback structure to form a stable and controllable adjustment mechanism.

[0029] Please refer to Figure 5 , the information filtering module includes: The clarity extraction sub-module calls the adjustment value executed by the mechanism, extracts the gray gradient of the image boundary, filters the pixel points that meet the clarity threshold, and obtains the boundary clarity change rate; Call the adjustment value executed by the mechanism. This adjustment value executed by the mechanism is based on the edge detection technology in the image processing algorithm. By extracting the change of the gray gradient of the boundary in the image and using the technology to locate the clear and blurred boundaries in the image. In a surveillance camera, for example, a camera used for traffic surveillance, this technology can identify the clear boundaries of vehicles and road signs to ensure that accurate visual information is obtained for traffic management. By calculating the clarity change amplitude between adjacent boundary pixels and comparing it with the preset clarity change threshold, if the clarity change between pixel points exceeds the threshold, the pixel points are identified as clear boundaries. Compare the gray gradient of the boundary pixel points with the threshold to determine whether the conditions for clear boundaries are met. For example, in a specific surveillance scenario, the threshold can be set to 20% of the gray change. When the detected gray gradient change of the boundary pixel exceeds this threshold, the pixel points will be marked and used for further analysis. In this way, the pixel points that meet the clarity threshold can be filtered out, effectively improving the image processing quality and the accuracy of surveillance. Through the refined image processing process, the boundary clarity change rate is obtained, and the data can be used to further optimize the image processing flow and improve the accuracy and efficiency of image recognition.

[0030] The dynamic ratio judgment sub-module calls the boundary clarity change rate, detects the consistency of the regional motion vector, and statistically obtains the non-target area ratio to get the non-target area occupancy ratio; The change rate of call boundary clarity is used to monitor the motion vectors of corresponding pixels between image frames. The steps are mainly used to identify dynamic regions in the image. For example, in a surveillance video, it is used to capture fast-moving objects. By analyzing the pixel changes between consecutive frames in the image sequence, the motion vector of each pixel point is calculated, and statistical analysis is performed on the vectors to determine whether the motion of the pixel points is consistent. If the motion vectors of most pixel points in a region are consistent, it indicates that the motion in this region is uniform and is caused by camera jitter or wind blowing grass, rather than the movement of the target object. By setting the area ratio threshold of the non-target region in the consistent region, such as setting that the non-target motion region shall not exceed 30% of the total image area, the non-target dynamic region can be effectively screened out. For example, the non-target region includes the dynamic background caused by wind blowing leaves. Through this technology, it is possible to prevent misidentifying natural actions as important events. The ratio of the non-target region obtained by this method can provide important reference data for subsequent image analysis and event judgment.

[0031] The perturbation exclusion sub-module collects the distribution of boundary noise pixels according to the ratio of the non-target region, and performs a comparison by combining the foreground and background gray-scale differences and the error amplitude. The formula is used: ; Calculate the perturbation rejection coefficient, screen out the noise region, extract the mask layer, and generate the background perturbation exclusion result; Among them, represents the perturbation rejection coefficient, represents the maximum value of the foreground gray-scale gradient, represents the average value of the background gray-scale gradient, represents the offset error amplitude of the noise region, represents the error change amplitude of the noise region, represents the ratio of the non-target region, represents the noise density value of the boundary region of the th segment, represents the number of segments for dividing the noise region; Operate according to the ratio of the non-target region. The ratio of the non-target region ω is calculated by the ratio of the area of the consistent motion vector region in the image to the total image area. For example, in a 1920×1080 pixel image, if the consistent non-target region occupies 345600 pixels, then ω = 345600 / (1920×1080)≈0.166. Collect the characteristic values of the noise pixel distribution in the image boundary region. First, divide the image boundary into z = 8 segments with a unit of 16×16 pixels; Sample the noise density within each segment , and the noise is defined as the pixel points whose intensity fluctuates by more than ±15 gray levels within three frames. For example, in the first segment, if the number of fluctuating pixel points is 204 and the total number of pixels in the region is 256, then , the values of the remaining segments are , , , , , , , and then obtain ; Next, collect the boundary gray - level features. The maximum value of the foreground gray - level gradient is the maximum value obtained by taking the first - order derivative of the boundary foreground pixels. Let its value be gray - level / pixel. The average value of the background gray - level gradient χ is the average value of the gray - level gradients of the background pixels in the region. Take gray - level / pixel; Analyze the offset error of the noise region. The maximum amplitude of the offset error is defined as the farthest distance that the center of the noise deviates from its geometric center, with the dimension of pixel. Let pixels, and the amplitude of the error change is the standard deviation of the noise pixel error. Let pixels; Substitute into the calculation: ; This perturbation rejection coefficient represents the significant degree of noise interference in the boundary region. By setting the perturbation rejection threshold to 5.5, judge that the current Q is higher than the threshold, and perform the rejection operation. Extract and mask the noise in the boundary region to form a clean mask layer for subsequent image fusion processing. Finally, establish the background perturbation exclusion result. By combining the noise distribution density with the offset error , , and introducing the ratio of the non - target region occupancy in the denominator for joint operation, the discrimination ability for complex background perturbation scenarios is enhanced. This result indicates that there are observable perturbations in the current image boundary region, and masking processing is required to ensure the accuracy of boundary recognition.

[0032] Please refer to Figure 6 , the feedback correlation module includes: The perspective - offset extraction sub - module, based on the background perturbation exclusion result, extracts the updated camera pose sequence and the starting pose data after background perturbation exclusion, analyzes the Euler angle difference and displacement change of the pose between consecutive frames, and performs normalization processing in combination with the time stamp to obtain the camera perspective - offset trend; Extract the camera pose sequence through the background disturbance exclusion algorithm. The exclusion of background disturbance is achieved by comparing the environmental changes between frames and using filtering or adaptive algorithms to remove irrelevant dynamic changes, obtaining stable pose data. The data includes attitude and position, which are composed of rotation and displacement in the coordinate system. According to the extracted pose data, calculate the attitude Euler angle difference and displacement change between consecutive frames. The Euler angle difference is obtained by calculating the rotation angle change between each frame, while the displacement change is obtained by directly calculating the Euclidean distance between the camera positions. Use timestamps to normalize the differences. Normalization is to eliminate the influence of the time length between different frames and map all displacement and angle differences to a unified time scale. Suppose between two frames of images, the displacement change of the camera is 0.5 meters, the rotation angle difference is 15 degrees, and the time interval is 2 seconds. Then it can be normalized to a displacement of 0.25 meters per second and a rotation of 7.5 degrees per second, and further obtain the camera's viewing angle offset trend.

[0033] The path linkage comparison sub-module analyzes the cosine value of the included angle between the direction vectors and the change rate of the speed amplitude at the corresponding time points according to the camera viewing angle offset trend and the spatial direction change sequence of adjacent pose segments in the target tracking path, and judges the response lag situation in the synchronization interval to obtain the path response synchronization characteristics. Extract the viewing angle data at adjacent time points from the camera viewing angle offset trend. The data is used to represent the spatial direction of the camera relative to the target tracking path. Extract the direction change sequence of adjacent pose segments in the target tracking path. For each pair of adjacent pose segments, calculate the cosine value of the included angle between their direction vectors. The direction vector is obtained by converting the rotation matrix or Euler angle of each pose, and the vector indicates the spatial direction of the path. The cosine value of the included angle is obtained by calculating the dot product of the two vectors. If the two direction vectors are exactly the same, the cosine value is 1; if they are exactly opposite, it is -1. Suppose the direction vectors of two adjacent poses of the target path are (1, 0, 0) and (0, 1, 0) respectively, then the cosine value of the included angle is 0 because they are perpendicular. Calculate the change rate of the speed amplitude, that is, obtain the speed increment at each time point through the change of speed and time. The change rate of speed can be obtained by performing a difference calculation on the speed data. Through the cosine value of the included angle and the change rate of the speed amplitude, analyze the response lag situation in the synchronization interval. The lag phenomenon is manifested as the moment when the change of the cosine value of the included angle lags behind the change of the speed, and further obtain the path response synchronization characteristics.

[0034] The adjustment strategy generation sub-module calls the path response synchronization characteristics and the camera speed change rate at the previous moment, sets the weight parameters according to the response lag time difference and the included angle offset degree, and performs weighted processing on the adjustment amount of each pose to generate the camera pose response matching degree. Obtain the path response synchronization feature and the speed change rate of the camera at the previous moment. The data is used to judge the response of the camera to path changes. The response lag time difference represents the lag time of the camera adjustment, and the included angle deviation degree refers to the deviation degree between the camera direction and the path direction. According to the information, set the proportion parameter. When the lag time is long, a larger proportion can be set for the lag time difference, and when the included angle deviation is large, a larger proportion can be set for the included angle deviation. When the proportions of the two are synthesized, the weighted adjustment amount of each pose can be obtained. For example, if the lag time difference is 0.5 seconds and the included angle deviation is 10 degrees, these two factors can be combined according to a predetermined weighting formula to obtain the final adjustment amount. Assuming that the proportion of the lag time difference is 0.6 and the proportion of the included angle deviation is 0.4, the final adjustment amount of each pose will be calculated by the weighted synthesis of these two proportions, thereby generating the camera pose response matching degree.

[0035] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent adaptive camera pose adjustment system, characterized in that, The system includes: The dynamic recognition module obtains the camera pose data, extracts the video frames captured by the image acquisition component, extracts the pixel brightness gradient and boundary contour continuity of the foreground region in the frame, recognizes the target position transformation data between the front and rear frames within the image stabilization structure, determines the motion trend of the region block and recognizes the continuous motion path, and generates the target tracking path information; The attitude adjustment module, based on the target tracking path information, extracts the current pitch and rotation angle values of the camera, combines the target movement direction and speed range, analyzes the attitude deviation degree and the required angle adjustment amplitude, and generates a pose correction instruction set; The mechanism control module, according to the pose correction instruction set, acquires the current response state of the execution structure, compares the target angle and displacement requirements in the instruction, recognizes the drive error range, adjusts the output amplitude and direction of the controller, and obtains the mechanism execution adjustment value; The information filtering module calls the mechanism execution adjustment value, extracts the boundary clarity and motion consistency features of the foreground and background regions in the image, recognizes the proportion of non-target dynamic regions, and eliminates the boundary noise regions, and generates a background disturbance exclusion result.

2. The adaptive camera pose intelligent adjustment system according to claim 1, wherein The target tracking path information includes the position change amplitude, the trajectory curvature feature, and the displacement direction angle. The pose correction instruction set includes the pitch angle adjustment amount, the rotation angle correction value, and the angle deviation threshold. The mechanism execution adjustment value includes the controller output gain, the rotation angle of the execution structure, and the real-time response displacement. The background disturbance exclusion result includes the foreground boundary clarity score, the background region consistency coefficient, and the dynamic interference ratio.

3. The adaptive camera pose intelligent adjustment system according to claim 1, wherein The dynamic recognition module includes: The pose estimation sub-module obtains the camera pose data, extracts the three-dimensional pose data and the central pixel coordinates of the camera for each frame, analyzes the pose rotation angle difference and the translation vector change rate in consecutive frames, determines whether the change exceeds the pose stability threshold, and generates a pose adjustment trigger signal; The foreground perception sub-module, based on the pose adjustment trigger signal, extracts the foreground brightness gradient and boundary contour continuity of the current frame, combines the spatial offset and direction change of the target edge points between the front and rear frames, determines the perception integrity trend of the foreground region under the pose change, and obtains the image perception integrity parameter; The path linkage sub-module, according to the image perception integrity parameter, dynamically corrects the current pose, adjusts the camera focus orientation and rotation angle, calls the offset difference between the position of the corrected image center point and the original image, and obtains the pose adjustment execution path.

4. The adaptive camera pose intelligent adjustment system according to claim 3, wherein The attitude adjustment module includes: The angle extraction sub-module, based on the target tracking path information, extracts the pitch angle and rotation angle data in the image frame, combines the pixel matrix change and the image edge position sequence, recognizes the pitch angle and rotation angle, and obtains the angle change value set; The direction matching sub-module calls the angle change value set, according to the target movement direction vector and speed range, detects the direction consistency and speed change trend, calculates the attitude direction offset intensity value, and combines the direction matching result and the speed jump range to obtain the direction offset adjustment parameter value; The instruction generation sub-module calls the direction offset adjustment parameter value, divides gears according to the offset amplitude and speed range, determines whether it is in the adjustment response interval, maps the correction action parameters, and generates a pose correction instruction set.

5. The adaptive camera pose intelligent adjustment system according to claim 4, characterized in that The mechanism control module includes: The state recognition sub-module collects the current response state of the execution structure according to the pose correction instruction set, extracts the attitude angle vector and displacement components, identifies the dimensional deviation distance, and generates the attitude displacement difference. The error evaluation sub-module calls the attitude displacement difference, judges the matching offset, extracts the dynamic change amplitude and response trend in the response state, compares the change rate with the response rate threshold, analyzes the driving precision difference interval, and obtains the driving error range value. The control adjustment sub-module calls the driving error range value, combines the angle response sensitivity, displacement response lag amplitude, and adjustment feedback cycle amplitude, corrects the adjustment amplitude and direction output by the controller, calculates the control output correction amplitude, corrects the controller signal, and obtains the mechanism execution adjustment value.

6. The adaptive camera pose intelligent adjustment system according to claim 5, wherein, The information filtering module includes: The clarity extraction sub-module calls the mechanism execution adjustment value, extracts the gray gradient of the image boundary, screens the pixel points that meet the clarity threshold, and obtains the boundary clarity change rate. The dynamic ratio judgment sub-module calls the boundary clarity change rate, detects the consistency of the regional motion vector, counts the area ratio of the non-target area, and obtains the non-target area occupancy ratio. The disturbance elimination sub-module collects the boundary noise pixel distribution according to the non-target area occupancy ratio, performs a comparison by combining the foreground and background gray differences and the error amplitude, calculates the disturbance elimination coefficient, screens the noise area, extracts the mask layer, and generates the background disturbance elimination result.

7. The adaptive camera pose intelligent adjustment system according to claim 1, wherein The system also includes a feedback correlation module: Based on the background disturbance elimination result, the feedback correlation module combines the updated target tracking path and the current pose execution state, evaluates the response synchronization between the camera view movement and the target path, and generates the camera pose response matching degree. The camera pose response matching degree includes the target path tracking deviation, pose adjustment delay, and synchronization coordination index.

8. The adaptive camera pose intelligent adjustment system according to claim 7, wherein The feedback correlation module includes: Based on the background disturbance elimination result, the view angle offset extraction sub-module extracts the updated camera pose sequence and the starting pose data after background disturbance elimination, analyzes the attitude Euler angle difference and displacement change between consecutive frames, and performs normalization processing in combination with the timestamp to obtain the camera view angle offset trend. According to the camera view angle offset trend and the spatial direction change sequence of adjacent pose segments in the target tracking path, the path linkage comparison sub-module analyzes the cosine value of the included angle of the direction vector and the change rate of the speed amplitude at the corresponding time point, and judges the response lag in the synchronization interval to obtain the path response synchronization feature. The adjustment strategy generation sub-module calls the path response synchronization feature and the camera speed change rate at the previous moment, sets the weight parameter according to the response lag time difference and the included angle offset degree, performs weighted processing on each pose adjustment amount, and generates the camera pose response matching degree.

Images